Craft Anything: understand how any software works, and craft your own.
Roadmap
The shared plan for reaching the product described in the README, for people and agents alike.
docs/STATUS.md says what works today; this file says what comes next, who owns it and what an
agent may pick up; roadmap-history.md keeps what's finished. Rebuilt on
2026-11-02 from main at 2e4afc5, with v0.3.0 the latest release.
How to use this file
- Each item has an id (
R<milestone>.<n>), an owner, a status and a list of pending changes. Pending changes are small, independently mergeable units. Ids are never reused or renumbered: commits, code comments and PRs quote them. - Claiming: before you start, put
— claimed: <name>, branch <branch>on the pending change and push your branch early (or open a draft PR) so others see it. Don't take a change someone else has claimed.— in PR #nmarks one a PR already covers: review that PR rather than starting again. The full working rules are inAGENTS.md("Working rules", mandatory). Items owned by maintainers are steered by the project maintainers: agents may take their unclaimed pending changes, but design decisions marked decision wait for a maintainer. - Finishing: tick the change (
[x]) where it is, add the commit or PR, updatedocs/STATUS.mdif what works changed, and add a line tolog/devlog.md. Maintainers move ticked lines toroadmap-history.mdwhen a release ships. - New work: add a
- [ ]line under the item it belongs to; a new item takes the next free id. - Status:
todo·in progress·blocked (<why>)·done. - Follow
AGENTS.md(layering, never crash, everything is a tool,platform::processfor every external program) and the architecture rules below.
Architecture rules for new work
The project should be easy to fork and bend. New features land behind a seam, not inside a
match:
- Traits plus a registry, not enums. Target kinds, checks, builders, asset providers and external engines are traits with a registry that holds built-in implementations. Adding one means adding a file and one registration line (as tools already work).
- Data over code for catalogues. Stacks, game-engine build recipes, model ids and prices,
and engine install pins live in typed tables (or TOML files embedded with
include_str!), not in string matches or prompt text. - Headless first; flows are tools. Everything works without the GUI. Anything a person
can start from the UI (rebuild, follow-ups, answering questions, engine installs, launching a
result) is reachable through
Engine::calland the CLI, so MCP, agents, scripts and other frontends get it for free. Orchestration, persistence and process control never live inui-eguiorapps/*. - Versioned files. Every file we write and read back (
craft.json,config.toml,project.toml) carries a version, has a JSON schema, and has a migration path. - Structured results. Decisions such as "did the build pass" read typed fields, never English text prefixes.
Milestones at a glance
| Milestone | Goal | Owner | Where it stands (2026-11-02) |
|---|---|---|---|
| M0 Foundations | Docs match the code, licences checked, one live end-to-end run | maintainers | Done |
| M1 Extensibility | Seams for target kinds, checks, builders, asset providers, engines, plugins | maintainers | Done, except engine pins (R1.5) |
| M2 Product completeness | Every kind of target gets study → build → check → refine with a real check | maintainers | Websites complete; apps and games build and run; API-key builds have never run live (R5.5); agents other than Claude Code are only partly wired (R2.2) |
| M3 Engine depth | More tools, and the known partial ones | open | 314 tools; the open lane |
| M4 Release | Packaging, signing, a managed tier | maintainers | v0.3.0 released, signed, on every platform; a macOS CLI question and the managed tier remain |
| M5 Works for real people | What people hit now that the project is public | maintainers | Comes first |
Open work
Every unticked - [ ] line in this file is open work, under the item it belongs to.
cargo xtask pending lists them all, grouped by item (--count for the totals). When you find
work that's missing, add it as a - [ ] line under the right item in the same PR, rather than
leaving it in a PR description or a chat.
Order and dependencies
Work in phases. Items in the same phase can run in parallel. A phase says what comes first; what stops an item from starting is its Depends on column, which must be done first. A soft dependency should come first but doesn't block. Issues are ranked by people affected and how badly, after leaving out what's already fixed.
| Phase | Item | Depends on | Why here |
|---|---|---|---|
| 0 Ship | R5.1 The next release | — | The Windows start-up crash and a build-breaking resume bug are fixed on main but not released |
| 0 | R5.2 Keeping contributions moving | — | main isn't protected, first-time contributors' CI is held, and a waiting PR fixes a safety bug |
| 1 Builds that work | R5.3 A builder that can't make progress stops | — | Costs people money or quota on every backend; small |
| 1 | R5.4 Local models can build | R5.3 (soft) | Largest open gap, with no fix anywhere yet |
| 1 | R2.2 Every builder gets the same tools (that group), and Approvals | decision for Approvals | Agents other than Claude Code start but can't do the job, and Clean room's rule doesn't reach them |
| 1 | R5.5 API-key builds fail in words people can use | — | This path has never been run live, and its errors say nothing |
| 1 | R5.6 Tools report only what they saw | — | Confirmed false successes on macOS; golden rule 3 |
| 2 Proof | R5.7 Live checks owed | — | Built but never run for real; each can reopen work above. Lines run when the machine or key they need is at hand, Clean room and the quick start first |
| 3 Depth | The rest of R2.2 | — | Detection waits for reports from 0.3.0 or later; ACP after the adapters work |
| 3 | R2.4 Runtime study, R2.5 Behaviour checks | scripted input (R2.4) before R2.5's sheet check | Makes "ready" mean more for apps and games |
| 3 | R2.7, R2.8 | — | Small gaps left in finished work |
| 3 | R2.9 Everyday use, R2.10 Pinned builds | — | Open to anyone at any time; most lines stand alone |
| 3 | R1.5 | — | Small leftovers |
| Any | M3: R3.3, R3.9, R3.11 | — | Open lane |
| Later | R4.1 macOS CLI, R4.2 Managed tier | decision | Not before M5 |
M5 Works for real people
R5.1 The next release
Owner: maintainers · Status: todo
Two reported failures, the Windows start-up crash (#414) and a builder resuming another's
session (#471), are fixed on main and still broken for everyone on 0.3.0. This is the
cheapest help we can give anyone, so nothing below should hold it up for long.
- #490 merged (
549ae5e) with the CLI fix the review asked for: a string-or-object input isany, so a JSON object given to--leftstays an object (tested both ways). What it reports is 0.2.0's, fixed by #93 (f6725c2): on Windows the published 0.3.0 CLI with Claude Code 2.1.179 mounted 314 of 315 tools (the 315th is the approval tool, which Claude Code doesn't list but calls), and a real Deep build reached Ready withcore_action_approveanswering five prompts. The reporter was asked for--version. macOS is still owed (R5.7) - Cut the next release (
docs/releasing.md). The version is 0.4.0 (#526) andmainatfb7ac86is green, but pushingreleaseis refused: its ruleset blocks every update and lists no one who may bypass it. A maintainer who can push it (or adds a bypass) goes next. 0.3.0 lacks0748ef1(#414: Windows starts DX12 only, the crash behind "it won't open"),1e8f661(#471: a builder never continues another builder's session) and3357daf(#430: Claude Code's message on stdin) - See that the launcher offers the new version: #74, #104 and #420 installed through it, and nothing in this repository says how it picks one
- After it ships, close what it fixes: #166, #346, #420, #478. #74, #104 and #222 give no
detail, so they may have another cause: keep them open until their fixes are validated
against the published release, following
docs/releasing.md. Ask #201 whether3861132(#379) ended its early compression (a guess at the link) - Close the tickets 0.3.0 already fixed, and the spam: #47, #53, #66, #95, #112, #113, #130,
#133, #138, #152, #179, #182, #185, #202, #214, #329, #344, #377, #378, #400, #449; spam #194,
#195. #113, #329 and #377 are 0.2.0's missing CLI, fixed by
1b54cbf, not what #490 reports. Not yet: #64 (its cause is fixed, its retries aren't: R5.3), #142 (R5.3), and #462, which saysnone, not #490's failure, and gives no version: ask for it - Sometimes starting the app shows nothing: after the release, see from
desktop.logwhether it still happens. DX12-only may have ended it; a GOG Galaxy Vulkan overlay layer was seen loading into the app - A program that is found but can't start (an npm script whose
nodeis missing) says where it is and what to look at, not "isn't installed" (#152, #95); on Windows a program found is started by its path, not its name - Live-check #152 on Windows 11 with OpenCode from npm, and on a machine with the agent but no Node.js
- A start that dies before its window (a native driver crash leaves no dialog or log, #74, #104,
#33) is remembered:
startingbesidedesktop.logis written before the window is asked for and removed once it's open, and a start that finds it draws with WARP that once (apps/craft-anything/src/start_guard.rs) - Live-check the start fallback on Windows: make a start die before its window (kill the
process at launch), start again, and see
desktop.logsay it drew without the graphics card and logdrawing withan adapter of typeCpu(WARP), and that the window opens. Passed on 2026-11-04 (mainat869eb85, release build, Windows 11, RX 6900 XT): a start killed right afterdrawing withthe Radeon leftstarting; the next logged "the last start never opened its window", drew with "Microsoft Basic Render Driver" (Cpu, DX12), loggedwindow openand removedstarting— claimed: 00xJS, branch docs/r5.1-start-fallback-live-check - The software retry still lets wgpu start every DX12 adapter before
pick_adapterruns, so a driver that crashes while its device is created crashes the retry too. wgpu has no way to skip hardware adapters; needs a different route (another renderer, or a helper that starts the app with only WARP visible) - Add to
docs/releasing.md: before publishing, a Deep build through Claude Code from the - #794: on some systems DX12 refuses egui's own pipeline (
egui_pipeline, 0x80070057 fromCreatePipelineState) on AMD, Intel and NVIDIA cards and on WARP too (lightcraft#812, pdfcraft#1046, wgpu#10771), so the software retry failed the same way every start, and the dialog only said to update the driver. Each failed start now moves on to the next renderer (DX12 → WARP → Vulkan → GL,apps/craft-anything/src/renderer.rs), keeps the one that worked for this app version, and starts again by itself after a wgpu error. Checked with a build that failed every DX12 start; the real failure wasn't reproduced (STATUS.md) — claimed: 00xJS, branch bugfix/r5.1-renderer-fallback - Live-check #794's fallback on a machine where DX12 refuses
egui_pipeline(the reporter's 780M, or lightcraft#812's) - A window that fails after it was shown no longer says "The next start will draw with …": its
start mark is already gone, so the next start draws as before (#909 review); the dialog says
it only when the window was never shown (
next_start_note) — claimed: 00xJS, branch bugfix/r5.1-retry-text-after-shown - Add to
docs/releasing.md: before publishing, a Deep build through Claude Code from the finished package, on a Mac and on Windows. Starting the CLI from the package (#91) doesn't show whether Claude Code accepts its tools — PR #982; claimed: doubektomas, branch docs/r5.1-packaged-deep-build-check. The checklist is documented; live package runs remain release validation work. - Add to
docs/releasing.md: after a release, close every ticket marked as waiting for it, so they don't stay open markedfixed— PR #1006; claimed: doubektomas, branch docs/r5.1-post-release-tickets. The checklist is documented; closing tickets waits for publication of the release containing their fixes.
R5.2 Keeping contributions moving
Owner: maintainers · Status: in progress
- Fix PR #1030's Windows descendant fixture startup and stop the watched launch before startup assertions can fail; cover the failure cleanup path — claimed: Codex, branch bugfix/r5.2-watched-launch-tests
- Run PR #1030's shell-free parent/descendant fixture and startup-timeout cleanup test on Windows; the contributor reproduced startup failure and leaked processes on 07671fe
- Clarify PR #1030's test Rustdoc: post-stop heartbeat sampling, child-fixture guard, and the direct-run no-write check — claimed: Codex, branch bugfix/r5.2-watched-launch-tests
- Address PR #1030 review: take the heartbeat baseline after stopping, guard the descendant stand-in against direct ignored-test runs, and document the test helpers — claimed: Codex, branch bugfix/r5.2-watched-launch-tests
-
regression_preview_limits_connections_and_releases_listener(platform) fails on macOS withInvalid argumentin its socket fixture, also reported before PR #1030's review fixes. Investigate the fixture separately from the watched descendant shutdown test
Review comes before new work while the queue is long, smallest and clean PRs first. A PR is merged only when it's green and a maintainer has fully reviewed it, code as well as wording.
- Maintainers: protect
main(PRs required, squash only); the only ruleset coversrelease - Auto mode's delete check is a safety bug that shows up as Mac test failures. It splits a
command on spaces, so
rm -rf "C:/x/Game Backup/y"from a project atC:/x/Gamegoes ahead unasked, and it lower-cases paths, so on a Mac or Linux a delete inside a project with capitals in its path is refused;auto_mode_keeps_going_but_never_does_the_dangerous_thingsfails on a Mac for that.real_binary_end_to_end_if_presentfails there too, expecting/bin/lsto be Linux: both fixed (#357) - decision First-time contributors' CI: GitHub holds their runs for approval, so their PRs
show no
gaterun (#490 and #413 were among them). Approve each push, or change the repository's approval setting - #193 changes the language list that #486 (merged) changed too: it needs a rebase. #468 looks redundant with #500: ask its author
- decision Wide PRs with no roadmap line: #416 (project goals, 34 files, cites an R2.11 that doesn't exist; that id stays free for it) and #503 (two asset providers and spending controls, 37 files, cites the finished R1.4). Decide whether we want them before more is built on them
- decision Replies don't reach people who file through the ticket relay (each such ticket says so), yet the bot asks them for details. About half the open issues came that way. Get replies back to them, or stop asking
-
which::tests::usual_windows_dirs_come_after_path(platform) failed on Windows (10 found, 9 expected, "nothing twice"): it didn't count Antigravity's second folder, fixed by #770 -
capture::console_tests::reads_what_the_program_drew_and_stops_at_the_deadlinefailed once with the whole platform suite running and passes alone (passed 3 whole-suite runs on Windows 11 on 2026-11-04). Find what it depends on -
process::launch_tests::a_launcher_that_hands_off_keeps_its_group_until_stoppedfailed on a busy Windows machine: it killed the run 500 ms after the start, and acmdthat hadn't exited by then made a cancelled run, not a hand-off. The test now waits for the launcher's mark, and a kill that finds the program already exited reports its exit, not a cancel — claimed: 00xJS, branch test/r5.2-hand-off-race -
website_tools::tests::a_browser_that_exits_at_once_is_reported_at_once(engine) failed once on Windows with the whole engine suite running (66 s under load) and passes alone, three times in a row. Find what it times -
a_helper_is_judged_by_whether_it_runs(ai crate) sometimes fails when several crates' tests run at once and passes alone: it writes a script and runs it straight away, which a busy machine can refuse while another test's child still holds the file open -
a_remote_query_expands_only_the_remote_working_directorycompares the shell's$PWDwith the fixture's canonical path, so macOS's/var→/private/varalias does not fail the test. Remote expansion and special-character quoting checks are retained — PR #949 — claimed: Codex, branch test/r5.2-macos-remote-query-path -
which::tests::usual_windows_dirs_come_after_path(platform) failed on Windows with 10 found, 9 expected ("nothing twice"): #748's second Antigravity folder, counted since #770 -
capture::console_tests(platform) timed out now and then with the whole platform suite running. On Windows Server 2022 a program another thread starts while a console opens inherits its handles and holds it open until it ends, and closing a console that programs are still attached to can hang its host. Program starts and console opens (and Unix terminal opens, which have a shorter gap of the same kind) now take one lock (process::STARTING), and a timeout ends the program's whole tree before closing — claimed: SoldierPePe, branch test/r5.2-console-capture-flake - A console program that exits by itself and leaves programs attached to its console: closing it can still hang the console host on Windows Server 2022 (only a timeout ends the tree)
-
process::launch_tests::a_watched_program_can_be_stoppedleaves itssleep 30(Windows:ping -n 30) running, onmaintoo:stopends the shell but not what the shell started — in PR #1027 -
a_helper_is_judged_by_whether_it_runs(ai crate) sometimes failed when several crates' tests ran at once: a script just written can't start while another thread's child still holds it open ("Text file busy"). #787 tries again: on Linux 4 of 72 runs failed before it, none after -
a_remote_query_expands_only_the_remote_working_directory(ai crate) fails on a Mac: the temp folder is/var/…, a link to/private/var/…, and the shell's$PWDgives the second. Compare with the folder's canonical path - Revisit the two advisory exceptions in
deny.toml:rsa(no fix; we only verify with public keys) when rsa 0.10 lands, and unmaintainedzbus_names/zvariant(Linux-only via keyring and accesskit) when those move on.deny.toml's comments pointed at R0.3, which never held them; they name this line now — claimed: kuseidos, branch docs/r5.2-deny-points-at-its-own-roadmap-line - Check whether the release workflows still run actions that target Node.js 20. None is named
here any more: every job calls
storytold/craft-common-actions@main, so a fix belongs there - Correct STATUS.md's obsolete GUI-only limit: document the shared project feed, engine setup, launch and preview tools, with their consent and lifetime limits — PR #1014 — claimed: Codex, branch docs/r5.2-headless-status
- A spinner on screen no longer makes the window redraw at the display's rate: egui's own asks
for a frame at once every frame, ours at the animation's pace (
widgets::Spinner, #579) - Measure #579 on Windows 11 (0.3.0 portable): the app's CPU while idle, while a build runs and during a website capture, before and after; and why a website capture is slow and sits for a long time (no details were given, so nothing was changed for it)
-
remote_discovery_uses_the_remote_program_and_connection_contextfails on a computer with OpenCode installed: its made-up path is missing, soprogram_offalls back toopencode— claimed: justinjohn0306, branch test/remote-discovery-without-opencode - Idle CPU on an AMD Radeon, measured (2026-11-04,
mainat869eb85, release build, Windows 11 build 26300, RX 6900 XT, driver 32.0.21045.16010, Ryzen 5 2600X; the window open on a fresh profile, nothing changing). DX12 on the card: one core at ~100% for as long as the window is open; Vulkan: the same; GL: 0%; WARP: ~280% for the first 5 s after start, then 0%. The busy thread isn't the app's: on DX12 it starts inamdxc64.dll, on Vulkan inamdvlk64.dll, and it starts with the device, not the main thread. A bare eframe window does the same with every present mode and frame latency tried (AutoVsync, Fifo, Mailbox, Immediate; latency 1, 2, 3), so nothing in our frame pacing causes or avoids it; GL alone stays idle. Seen in the same way with PdfCraft and PhotoCraft (pdfcraft#3255) — claimed: 00xJS, branch docs/r5.2-idle-cpu-amd - decision On the one Radeon measured (RX 6900 XT, driver 32.0.21045.16010, Windows 11
build 26300), AMD's idle driver thread costs a core on DX12 and Vulkan; other Radeons
(integrated ones too), drivers and Windows builds haven't been checked. Say so in the
troubleshooting docs (with
WGPU_BACKEND=glas the way out, which loads AMD's OpenGL driver), report it to AMD, or leave it - Limit the decision line above to what was measured (one RX 6900 XT, one driver, one Windows build) and say other Radeons and drivers aren't checked (#912 review) — claimed: 00xJS, branch docs/r5.2-radeon-idle-scope
R5.3 A builder that can't make progress stops
Owner: open · Status: todo Every round spends the person's money or quota. One that can't do anything shouldn't start.
Preserve the interrupted request when an agent cannot restore its session: ACP starts a new session, or a registered CLI adapter falls back to Plain (#552 review). — claimed: Afloat16, branch bugfix/r5.3-interrupted-request-recovery-20261103
A build stops and says why after two builder turns in a row that changed nothing in
app/, on every backend (#44, #110, #417, #481). The count is saved in the project, so it holds across runs: a run ends at 7 checks (max_rounds) but check numbers carry on, so #44's check 28 and #417's 35+ look like several runs (a guess: Try again is the only restart found, and Auto mode doesn't start runs) — implementation:2fc2970948315b64d47359543b458509da42ab3fA builder whose tools didn't start isn't resumed.
turn(rebuild/mod.rs) tries 3 more times ("Picking up where it left off", #377) becauseworth_retryingdoesn't know this error. Give it a typed signal, asConversation::session_lostdoes, not one more word in that function's list. Say what failed and stop (pitfalls: never retry what can't succeed) — commitf80a16a66834worth_retryingdecides on English words in a message (AGENTS.md: structured results). Move its other cases (sign-in, credit, a missing install) to the same typed signal — commitf80a16a66834Find what repeated #64's attempts and stop it: it logged 41, about 38 s apart, on 0.2.0, but
turnallows 3 retries 30, 60 and 90 s apart, there too, so something else starts themA rate limit is waited for only on agents with a session:
turnretries only whenagent.session_id()is set, which API backends and OpenCode's plain adapter never set (read in the code, not run). So #148 didn't reach #142, which was OpenCode. Wait and send the turn again for them too, then close #142 — commitf80a16a66834A builder step on an API backend stops after 25 failed build commands in a row (none succeeding between), says so and keeps the project (
MAX_FAILED_BUILDSinagent.rs, #399: 9.4M tokens in 19 minutes of the first stage, never compiled). Try again carries ondecision Whether a build gets a token or cost limit of its own, beyond rounds and that stop (#399): a model that changes code without ever running a build isn't caught by it
Check live that the stop fires on a real API-backend build and the message reads well. It fired on a local Ollama build (the same built-in loop) after two turns that changed nothing in
app/("The build stopped after two builder turns changed no files in app/…"); an API-key build is still owed#736: the build check and the stop message say when
app/is empty, so a builder that claims a build can't hide that nothing was written. v0.3.0 (the reporter's release) had no stop after two unchanged turns (#557, merged after it); main stops, and this adds the empty-folder wording to the build check (it failed first, before #749's website check) and to the stop message — claimed: claude, branch bugfix/r5.3-empty-app-loopA build whose checks keep finding no working build stops after 10 in a row, across "Try again" presses, until a new request starts the count again (
NOT_WORKING_CHECK_LIMIT, kept in craft.json asnot_working_checks). "Try again" used to start six more rounds every time with nothing counted across runs, so #736's build went on to check 40 and 49 times. A run at the limit stops before a builder turn. The unchanged-turn guard already stops a builder that writes nothing; #736's exact Windows run (gemini-cli) is not reproduced here. An emptyapp/is named as having no files yet (#771's wording) and the builder is asked to write the project and show its build's output, not to fix a build (NO_PROJECT_PROMPT). — claimed: fabox-os, branch bugfix/736-stop-empty-build-loopLive-check #736 on Windows with gemini-cli: an empty app/ and a builder that claims
npm run buildstops after two turns (or 10 checks) with the "no files yet" note, and Try again doesn't restart it
R5.4 Local models can build
Owner: open · Status: todo The built-in agent loop assumes a capable tool-calling model. Reports #24, #44, #110, #403, #417 and #498 say otherwise (#44's backend is a guess; #116 may be the same), and no local model build has been run live.
- A live local build through Ollama, small target, to Ready or to a clear stop, and note in
STATUS.md which models it worked with: Ollama 0.40.7 with
qwen2.5-coder:14bon Windows (RTX 5090, 32k context, no prompt cut) stops clearly at "No project to build found in app/ yet", twice. The lines below say why - A builder on a local model is offered 13 tools (about 10,000 characters of descriptions)
instead of 33-41 (23,000-30,000): the Build tools and the Core tools the brief names, not the
study set a model explores with instead of writing (#417,
LOCAL_BUILDER_CORE). Measured as text, not on a model; Ollama's default context (4k) was too small for the old list alone - Run #417's case live (Ollama,
qwen2.5-coder:14b, "a scientific calculator") before and after this: does it callbuild_file_writenow? If it still doesn't, the model can't build. Yes, with the three fixes below: over about 50 calls it wroteapp/Cargo.toml(eframe), fixed the two errors the checks reported, and check 3 said "Build succeeded". Butsrc/main.rswas still cargo's "Hello, world!": no calculator was written, and the screenshot check rightly failed it at the round limit (#753) - The system prompt told every model to call
core_brief_writeandcore_plan_updatefirst andcore_handoff_writelast, tools a lean local builder isn't offered, so the model asked only forcore_brief_writeand did nothing else for three checks. A lean builder's prompt leaves the memory steps out; live, the same model then used the Build tools - A call written after a sentence ("Let's list the files.\n
json {…}") isn't run: #554 reads a reply that is only the JSON. Half the model's calls came this way and the turn ended with nothing done. A reply that ends in completejsonfences, each a call to an offered tool, with only words before or between them, now runs them all (fenced_calls, idsreact_Nas other text calls); words after the last fence, other code, an unclosed fence or a tool that wasn't offered keep it text (#753) - A call followed by a summary isn't run, and the model is told to retry it without trailing
text: in a second live run of
#417's case both turns put "### Summary:" after their one
jsonfence, nothing ran, and the build stopped after two turns with no files. The next message now says that it wasn't run — PR #1028, branch bugfix/r5.4-fenced-call-summary - Keep the retry condition when
Action:orFinal Answer:appears between fenced calls, without rejecting ordinary words between calls (PR #1031 review) — claimed: Codex, branch bugfix/r5.4-fenced-call-followup - A local builder spends its rounds on the project file and never writes the app: run 4 had a
compiling
Cargo.tomlby check 3, its last, withmain.rsstill "Hello, world!". See whether more rounds get there, or whether a small model needs the window and the--screenshotoption written for it (a template in the Rust steps) - The model ran
cargo new appinapp/(commands run there), so the project landed inapp/app/and every check said "No project to build found in app/ yet". The check now says where a cargo or npm project one folder down is, and the Rust steps sayapp/Cargo.toml, not a folder insideapp/(a brief can't saycargo init: not every builder's commands start inapp/). Live, with calls after a sentence run too, the model then madeapp/Cargo.tomland the build compiled (#751) - The model reads file paths from the project folder and commands from
app/as one base: it rancargo init appandcd app && cargo buildafter writingapp/Cargo.toml. Say the folder inbuild_command_run's result (or its description) so the two can't be mixed up -
usage.jsonafter a local build says about 4,200 input tokens and 6.5 s of work, while Ollama logged about six requests of 4,000-5,000 tokens each: find what it leaves out. Cached tokens: Ollama 0.40 reports them apart (cached_tokens), and the built-in loop'sTurnDoneleft them out while coding agents count them. It counts them now; a live run on Ollama 0.40.7 had recorded 5,323 input tokens of 75,014 — claimed: SoldierPePe, branch bugfix/r5.4-usage-cached-tokens - Told that its
cargo new appmadeapp/app/, the model kept editingapp/app/Cargo.tomland runningcargo buildinapp/, and checks 2 and 3 still found the project one folder down (live, GPU). See whether naming the move gets it out, or whether the run should move it - The model reads file paths from the project folder and commands from
app/as one base: it rancargo init appandcd app && cargo buildafter writingapp/Cargo.toml.build_command_run's result now says where it ran (ran_in), a command that makes a project inapp/app/is told it goes inapp/itself, and one that starts withcd appwhere there's none is told to leave it out. Live (Ollama,qwen2.5-coder:14b) the first command was stillcargo new app, before any result could say so. On a GPU (Ollama 0.40.7) the note then reached the model; on the CPU a second run wrote the call with a summary after it, so nothing ran (the open line on a call followed by a summary) — claimed: SoldierPePe, branch bugfix/r5.4-build-command-folder -
usage.jsonafter a local build says about 4,200 input tokens and 6.5 s of work, while Ollama logged about six requests of 4,000-5,000 tokens each: find what it leaves out - An assistant turn with no text sends an empty string, not
content: null, which Ollama refused asinvalid message content type: <nil>(#403, #536) - Understand a tool call a model writes as JSON in its text (#417; implementation
dfe9a4db2dcb3e27edbd67311cbb4202584e7af6). The fallback protocol reads onlyAction:andAction Input:lines, so anything else ends the turn with nothing done - Preserve
lengthandcontent_filterstop reasons on native tool-call streams; a parsed call must not override truncation or refusal and execute tools. Pair other retained unexecuted calls with results before the conversation continues — PR #544 - Return a rejected tool argument's JSON Pointer when known and guidance to correct the call
(#498), keeping schema types strict
— commit
0c3ae5cca569 - Find how each local server is told its context size and set it, or stop with a plain message
when the prompt won't fit. We assume 24k and send no
num_ctx, which an OpenAI-style endpoint may ignore anyway; LM Studio refused a 12.6k prompt on its 8k default (#73) - When a server refuses a request for its context window and says the size (Ollama, llama.cpp, OpenAI-style), fit the conversation to that window and ask once more, else say how to raise it (#962)
- A per-backend context budget in
[[ai.backends]](LoopConfig::context_tokensexists), for services whose real window differs from 150k (#374) — implementation:6bd94b84828ec89b805cb8a246f557d0a778c037 - Find LM Studio on another port (#73): the port in its settings file (when it is there) and
the ports 1235-1239 are probed like the usual servers; OpenCode's own
baseURLs were already read (#83) - Check live that LM Studio on
--port 1235is offered, and where its settings file is on Windows, macOS and Linux (LM_STUDIO_SETTINGSis from memory of its folders, not verified) - See why Bionic or LM Studio can't be picked (#140)
- Do not select an embedding, ranking or transcription model as a chat backend when a local server has no chat models — claimed: Afloat16, branch bugfix/r5.4-nonchat-model-detection-20261103
- Recognize specialised model file paths and numeric task-name versions during local chat setup — claimed: Afloat16, branch bugfix/r5.4-model-task-identifiers-20261103
- The first run's AI step can add a model server on another computer by its address (llama.cpp, Ollama or LM Studio on a home server), the same form as Settings → AI
- A chat sent while a llama.cpp router is still loading the model ended after 31 s with "The answer stream ended early." and no next step (seen once live, 27B model); say the server may still be loading it and to try again in a minute — claimed: yashpatadiya1724-cmyk, branch bugfix/r5.4-stream-loading-remediation
R5.5 API-key builds fail in words people can use
Owner: open · Status: todo
-
rebuild --backend <an API backend>still adds the first agent app it finds to the person's settings and prints "Using Claude Code (…)":ensure_agent(rebuild_cmd.rs) runs whenever no agent app is set up, though this run won't use one. Run it only when the build needs one — claimed: AlexJoaquimPereira, branchbugfix/r5.5-no-agent-with-api-backend -
--backend <a name that is turned off or unknown>is refused with a specific message: a turned-off backend is named and must be turned on, and a name that matches nothing is named instead of saying to install an agent or add a key — claimed: AlexJoaquimPereira, branchbugfix/r5.5-disabled-backend-name; commit a380b3e - When a project's saved builder was turned off, Start builds with the default AI without
saying so: the UI picker filters the stale name out (
ui-egui/src/builders.rs) instead of reporting that the chosen AI was replaced. The Make picker says so in a toast, and every start (Start, Continue, a note, Try again) names the saved AI and the one building instead — claimed: AlexJoaquimPereira, branchbugfix/r5.5-stale-builder-notice -
rebuild --continue-in,rebuild_starton a project andrebuild_question_askname the saved builder that is off or gone and the AI building instead: the CLI prints it before the run,rebuild_startaddsbuilder_notice, and an answered question carries it — claimed: AlexJoaquimPereira, branchbugfix/r5.5-headless-builder-notice - The saved-builder replacement notice is English only for now; add it to the catalogs after a native speaker checks the wording
- Bound accumulated data across all lines of one server-sent event, so short lines cannot grow a streamed API response without limit — claimed: Afloat16, branch bugfix/r5.5-bounded-sse-events-20261103
- Show the provider's own reason: #369 bought $10 of credit and got only "OpenRouter answered 400: Provider returned error". Find the reason in the response body and show it — PR #542
- OpenRouter picks an upstream provider per request, and one without tool calling answered
"400: Provider returned error" (#369): a request with tools now sets
provider.require_parameters, and a 404 "no provider for these parameters" falls back to the text protocol like any server that rejects tools - A saved fal.ai key reaches an API-key build's asset tools; every call said "No fal.ai key" (#923). Outside a project the tools now say that instead. Not checked with a real fal.ai key or API build
- An API-key build works alone: it shares the app's session, whose
attachpoints at whichever project started last, so another build beside it sent its files and asset spending to the wrong project (build.parallelis 2 by default). Others wait their turn in the queue - Let API-key builds run side by side: give each an engine of its own, as a question has. This
also closes the gap left in working alone: the app and the
rebuild_starttool each keep a queue, so a build the in-app chat starts isn't kept apart from one the app started - Check #369 live with Kimi K3 and a key limited to Kimi/Moonshot models (needs the person's OpenRouter key): that the build starts, or that the error now names the provider's reason
- A live API-key build (needs a key on the test machine)
- A Google AI Studio (Gemini API) key is its own choice in Settings → AI and first run (#646):
the
geminibackend kind uses Gemini's OpenAI-compatible endpoint andGEMINI_API_KEY - Check a Gemini build live with a real Google AI Studio key (tool calls through the OpenAI-compatible endpoint)
- Failure logs for the other ways an AI stops (API backends, MCP server start-up), like the
ones agent CLIs write to
logs/in the config folder
R5.6 Tools report only what they saw
Owner: open · Status: todo Golden rule 3: the result is evidence. Each fix needs a Mac to test on.
- macOS: a denied Accessibility permission is reported as success with an empty tree (#240) — claimed: mavericzz, branch bugfix/r5.6-accessibility-denied; PR #887
- macOS:
runtime_ui_readcaptures a requested screenshot (#248): one on-screen window belonging to the pid, restricted bywindow_idwhen supplied; PNG/base64 up to 2048 px per side. Combined JSON (including escaping and the newline) stays within 16 MiB by trimming the tree with explicit truncation and partial coverage, preserving the image. Missing Screen Recording permission, a missing window or a capture failure produces an explicit screenshot limitation and partial evidence. Denied Accessibility permission with the tree enabled (the default) returns a tool error before capture;accessibility: falseskips the tree. Requires macOS 14+ and existing Screen Recording permission; opens no dialog — claimed: Codex, branch bugfix/r5.6-macos-screenshot - macOS:
runtime_ui_readhonoursaccessibility: false(#307): the Swift helper skips the tree and its permission check, returns no nodes without truncation, and the evidence records the flag and says the tree was skipped. Walkthroughs still require Accessibility — claimed: Codex, branch bugfix/r5.6-macos-ui-flags - macOS: call observation's lldb callback was a literal
\nin its Python, so the Python was invalid and it reported zero calls (#251). Fixed under R2.4: the callback is one valid line covering every breakpoint, and a failed lldb run is an error, not "0 calls"; only the live lldb run is still owed (R5.7) — claimed: mavericzz, branch docs/r5.6-stale-251; PR #1070
R5.7 Live checks owed
Owner: open · Status: todo The code for each of these is done; what's left is to run it for real and record the result in STATUS.md. A failure becomes a new line under its item. Grouped by what the run needs. Start with the Clean room build and the README quick start: Clean room is the default and has no live build yet, and the quick start is how every new person begins.
A signed-in coding agent:
- A live Clean room build of a website and of an app, checked end to end
- A live build in a picked folder each way (edit in place, start from), checked end to end
- Thresholds under pressure: a clone that's off by a few pixels (a page with fluid type or
clamp()spacing) to see whether 4 / 2 / 8 px is too strict - Screenshot check: in a live run the final check once got no picture from an Electron app within 90 s while the same check passed in the builder's turn and on a re-run; not reproduced. The app waited on animation frames, which a window behind others may not get; the Electron steps now say to capture after loading instead
- A live chat run with a big comparison result, to see Claude Code page it with
core_evidence_browse(and whether 80 KB is the right size) - Live check on Windows: rebuild a project while its frozen pin runs, and the build isn't blocked
An API key or a bill:
- Run Nano Banana Pro once with a real key (a picture, then a 3D model from it) and check the picture's size and the invoice ($0.15)
- fal bills pictures "by rounding up to the nearest megapixel"; estimates use the exact size. Confirm on a real invoice whether a 1024×1024 picture bills as 1 or 2 megapixels, then round in the table (needs a key)
- A live extraction run with Kimi K3.1 Pro (needs the person's OpenRouter key)
A second computer or another system:
- A live run of a project copied between two real computers with Claude Code (#451, #452); done
on one Mac with a second
CLAUDE_CONFIG_DIRstanding in for the other computer (STATUS.md) - Ghidra setup tried for real on Linux and macOS (the Java pins were checked by tests only)
- A Ghidra with no decompiler program for this computer (public downloads leave out macOS) is refused before the long analysis, with the command that builds it; checked on a real Mac
-
engines_installbuilds Ghidra's native programs on macOS, so a Mac install works unaided - Follow the README quick start on a clean machine (Windows, and Linux or macOS)
- Build the new DMG on a Mac (release dry run) and look at the window once
- #490 on macOS: a real Deep build through Claude Code with the published 0.3.0 CLI, with 2.1.179 and a current version. Windows passed; the Mac CLI was only checked to contain the fixed schema
- macOS CLI zip blocked by Gatekeeper when downloaded in a browser (#94), though Apple has its notarization tickets: find why the online lookup fails for people (it passes on our Mac, which may be exempt), on a machine that shows it
A real program, device or file:
- Cutter sessions: try them against a real Cutter, including whether
=h&keeps Cutter responsive (only a stand-in server so far) - Vineflower: run it once on a real JAR (no Java on the machine it was written on) — claimed: hasnain607, branch docs/r5.7-vineflower-live-run
- Verify the 23 unverified adb tools; record results in tool-status.md
- Android locales: try
only_localeson a real APK (only the test fixture so far) -
app_graphs_compareon the two real Electron releases again, to count how many of the 162 "removed" now pair (and pair theindex-<hash>chunks by structure if many stay) - A real IL2CPP
global-metadata.dat(needs a Unity build or a shipped game); synthetic so far. Compare a real APK's components against another tool's reading - Stand-in backend: a browser-made HAR and a real front end talking to it (only a hand-written HAR so far); GraphQL and tRPC on real bundles (only hand-written minified shapes)
- Check the Kimi Code, Factory Droid and WorkBuddy registrations in the real apps
M1 Extensibility
R1.5 External engine trait in L3
Owner: open · Status: in progress
- Find Chrome, Brave, Chromium and Edge installed as Flatpaks on Linux (system and per-user exports), and keep a Flatpak or Snap browser's private profile in the home folder its sandbox shares, since its /tmp is its own (#747). Unit-tested; not run on a real Flatpak
- Find browsers under their other Linux names, and say why one isn't found: a set path that isn't there, or a Flatpak's sandbox (#841). Unit-tested; not run on Linux or in a real Flatpak
- decision Let the Flatpak start browsers on the computer. It can't see system ones today;
a per-user Flatpak browser in
~/.local/share/flatpak/exports/binis found, but its start script needsflatpak, which the sandbox doesn't have.flatpak-spawn --hostwould need--talk-name=org.freedesktop.Flatpak, which lets the app run anything outside its sandbox (#841) - More pins once their hashes are checked against the publisher: Chromium (no published checksums found yet), Rizin for Linux (the tar.xz unpacks to ~800 MB, so it needs streaming extraction) and macOS (a .pkg)
R1.6 Headless first
Owner: maintainers · Status: in progress
- Give Linux-specific Clang guidance in Unreal's guided installer when the bundled compiler
is missing (the requirements summary is already platform-specific).
— claimed: AlexJoaquimPereira, branch
bugfix/r1.6-linux-unreal-clang - Give macOS the right C++ guidance when Xcode is missing: the status line and requirements summary now name Xcode, matching the guided step.
- Find Visual Studio's C++ tools on Windows when Visual Studio is installed outside Program
Files (through the Visual Studio Installer's
ProgramDatainstance state or developer environment variables, #669, PR #726) — claimed: riyaz, branchbugfix/r1.6-windows-vs-cpp-tools - Turn off web and capture tools by default in restricted MCP mode (#856, #855).
M2 Product completeness
R2.2 More coding agents
Owner: open · Status: in progress
- A saved agent program path that no longer exists (Codex desktop moves its CLI to a new build folder on every update) says so and names the path, instead of "isn't installed or isn't on your PATH" (#721); it stays a missing program, so a builder turn isn't retried
- A relative program path started in another folder is looked up where it was started before it's reported as missing (Unix: a script whose interpreter is missing was said to have no file there)
-
ai testnames the program it ran and the saved one it replaced, anddoctorsays a saved path has no program there instead of blaming PATH (#721) - Settings says when a saved agent program is gone and which one runs in its place; today a build moves to the newest build folder or the usual program without saying so, and the setting keeps the old path (read on a worker, not while drawing) (#721)
- GitHub Copilot through its own CLI login and ACP, without credential import:
discovery, MCP-only tool restrictions, model selection and fresh read-only questions;
unit tests and CLI 1.0.105 with an isolated home and loopback model, all five access modes
— claimed: qu-gg, branch
feature/r2.2-copilot-integration(recorded during PR preparation) - Grok Build as a coding agent over ACP (
grok agent stdio): discovery in~/.grok/bin, the person's owngrok login(a missingauth.jsonis unknown, so an API key still starts), model and reasoning effort, and our MCP server on the session with"type": "stdio"only for Grok. Clean room, deep chat, and questions are refused, because Grok can read and edit without a permission request. Verified against Grok 1.0.77's ACP handshake andsession/set_config_option(the CLI--modelflag does not stick aftersession/new). A live deep build on 2026-11-06 (Grok 1.0.77) finished, called the Craft Anything server, and answered one permission request. Clean room and deep chat were not run live; those runs are refused. See Grok Build. — branchfeature/r2.2-grok-agent - Review hardening: cap target handoff at 128 while preserving the active target; propagate context-discovery cancellation, normalize CLI-only efforts for Claude requests, and name the configured agent in shared picker warnings, with focused regressions and translations
- Local agent chats share a saved investigation and open-target metadata with MCP; background
refresh makes recorded findings visible in the workbench Evidence pane.
core_finding_editcorrects a finding with checked evidence and an optional revision guard, preserving confidence when omitted. Builder refresh failures warn instead of ending the turn; shared ACP errors use the current agent and its sign-in guidance. Engine, process-level MCP and headless rendering tests cover sharing, mode boundaries and finding ids; regressions cover confidence preservation, explicit overrides and agent-specific errors - Serialize project-backed investigation load, revision check, mutation and all rendered writes under one bounded, cancellable interprocess lock. Refuse malformed state rather than replacing it; separate-process regressions retain concurrent appends and reject one of two updates based on the same revision
- Refresh externally changed investigation/evidence records incrementally, avoiding repeated full reloads under the session lock while retaining mid-turn findings visibility
- Offer OpenCode locally and over SSH with its own visible chat models and supported thinking depths (Low, Medium, High, Xhigh, Max) in settings and headless clients. Choices are validated before saving; explicit choices are refused at execution until native MCP support is available. See OpenCode.
- Capture OpenCode settings and project together; refuse a busy session instead of treating it as no project, and preserve open model pickers and matching results during contention.
- Report cancelled model validation as stopped, while preserving error events for other failures.
- Finish rejected or cancelled model validation with a terminal event so headless chat exits; document functions touched by the OpenCode model-selection changes.
- Integrate the five-level
Efforttype from #395 for agent variants, and share a strict disk-based selection save between the hub and headless tools, preserving unrelated edits. - Rebase onto main after #395's squash merge, preserve extra model arguments without a saved selection, and render the remote-host placeholder visibly in the setup instructions.
- Hold the UI settings worker's session guard through its disk save so it cannot overwrite a newer model selection; reverify the existing terminal-event regression for validation failures.
- Label the settings-save guard so filesystem I/O is identifiable as a running job by the lock tracker.
- Finish a failed CLI worker-thread start with a terminal event after its error so headless chat stops polling.
- Synchronize the UI exit-time settings save with the session's latest AI table and other writers; an isolated exit-save regression preserves the latest model/depth and pending app settings.
- Never wait for a running tool's session lock at shutdown: defer non-AI app edits in a versioned sidecar when busy, then recover against the latest committed AI settings at startup.
- Discard superseded deferred settings after a successful full exit save; document model-only and model-with-variant command forms, and share transport selection between preflight and runs.
- Rebase after #567 and translate the ten model-selection strings and two agent tool titles in the Russian catalogue, restoring coverage and catalogue key parity.
- Supply and verify the run's Craft Anything MCP server in native OpenCode before enabling
explicit model/thinking execution: a config file in
OPENCODE_CONFIGwithrun --standalone(verified live locally). — claimed: advokat11, branch bugfix/r2.2-opencode-native-mcp - Hand native OpenCode runs over SSH the Craft Anything MCP server (the config file is local).
- Hand native OpenCode runs over SSH the Craft Anything MCP server (the config file is local).
One
sshrun checkscraft-anything-cliis on the host's PATH and writes the config on the host (~/.cache/craft-anything/, mode 600, atomic rename); the builder fails fast with the fix if the helper is missing. Verified: POSIX-shell regression test (the_remote_line_runs_on_a_posix_shell) and an SSH round-trip to localhost (probe, byte-exact write, mode, cleanup). Not run with a live agent over SSH. — claimed: jelassiaymen94, branch feature/r2.2-opencode-remote-mcp - Make
the_remote_line_runs_on_a_posix_shellaccept the shell's missing-command status: assert failure rather than 127, isolate PATH from installed helpers, and verify that a missing helper writes nothing in a fresh home. Fixture utilities use host lookup rather than assuming/bin; the shell uses the bounded process runner. Reproduced the old failure on macOS; the AI suite now passes (446 passed, 2 ignored). PR #1018 — claimed: Codex, branch test/r2.2-posix-remote-helper - Find OpenCode's CLI inside its Windows desktop app (
opencode-cli.exe), for detection and for runs (#1060) - Settings → AI can only add an agent it detected: add one by choosing its program, for a CLI kept somewhere we don't look (#1060)
- Verify OpenCode model discovery and an explicit thinking selection on a live SSH host, and complete a live generation with the selected model and variant (local catalogue verified).
"Found on this computer" has to mean "can do the job". Today only Claude Code gets everything. Every builder gets the same tools:
- A table of what each adapter gets (our tool server, approvals, resume, notes mid-turn, a
message of any length), as typed data, with a test that fails when one is left out.
AgentAdapteralready answers some, such asresumesandprompt_on_stdin. The lines below fill it in - OpenCode's plain adapter never passes our tool server, so its builder has none of our tools and hunts for a setting that doesn't exist (#481). Hand it the server, or build over ACP by default once that passes
- Aider has no adapter, so the plain one hands it the prompt as a file name, and it can open a browser sign-in nobody agreed to (#72). Give it an adapter, or stop offering it — claimed: Afloat16, branch bugfix/r2.2-aider-recommendations-20261103
- Add a non-interactive Aider adapter before recommending it again; detection currently leaves it out.
- Codex (#499): in a chat the tool server has no active target. The reporter attached a patch for chat continuity; look at that first
- Codex (#538): list every permitted tool from startup, since Codex doesn't refresh its callable
list after
enable_toolsetchanges it — PR #805 - Codex (#540):
rebuild_startwas refused bycodex exec's approval policy and showed asnull; setup and the builder now approve our tools for Codex (we still ask before running code) and a refusal explains the fix (docs/codex.md) - Codex (#540): run
codex execon a real install and checkrebuild_startgets through - Codex: resume support (
codex exec resume <id>), verified on a real install - Every adapter but Claude Code's still takes the message in argv (
prompt_on_stdin), so a long one can pass Windows' command-line limit as Claude Code's did (#402); hand each its message on stdin or in a file where its CLI reads one - A note sent during a build reaches the builder at its next step on other coding agents with tool hooks, and on the API-backend loop between tool calls (Claude Code only today)
- Session copies for other agents that keep sessions (ACP agents'
session/load) (#451) - Hermes Agent's native Windows install (
%LOCALAPPDATA%\hermes\bin) is searched, so an app started before the install, or by a launcher, finds it (#116) - Live-check #116 on Windows: install Hermes natively, start the app from a launcher, and see Settings → AI offer it; #116's "error after every message" has no text yet (asked)
- See whether Hermes still fails on 0.3.0 (#460: the fix its
fixedlabel names was already in 0.3.0 when it was filed, and it gives no version)
Approvals:
- decision One approval behaviour across agents, and whether we want an auto-approve mode
apart from Clean room and Deep (Auto mode only governs what's asked through us). Today it
differs by agent: Claude Code asks our approval tool; OpenCode's builder runs with
--auto; Hermes's-zturns its approvals off on every run, chats included. Clean room's approval rule doesn't reach an agent that never asks us - Clean room outside builds (#564): a chat or question with Claude Code refuses the agent's own
file, shell and web tools outright (
--disallowedTools), not through its defaults; an ACP chat refuses them through an approver instead of a consent card (docs/claude-code.md) - Say on each adapter, as typed data, what it can be held to (asks us, its own sandbox, nothing), in a build, a chat and a question, and have runs act on that, once the decision above is made. Today Codex, Gemini CLI, OpenCode, Antigravity, Hermes and Maki are not held to Clean room in a chat (their own sandbox, if any, is not established)
- Check live that a Clean room chat with Claude Code can't read a file outside its folder, and that an ACP chat in Clean room still reaches our tools (its permission requests for them are allowed, only its own read/write/command/fetch are refused)
- Check live what Hermes over ACP asks about: whether its file reads reach our approval policy
- Maki's plain
maki -pbuilder doesn't ask us before commands, and in Clean room its reads aren't checked (as with every agent that never asks us): say so in the approval decision above, or run Maki over ACP by default once that passes. Outside a build, tools the person allowed in Maki's own permissions (an MCP server's, or all of them withalways_yolo) still run; the adapter can't see that config - Maki over ACP writes inside its working folder without a permission request (its built-in
rule), and that folder is a build's project folder, so writes to our own records (inbox,
manifest) skip the approval policy;
maki acptakes no tool flags to stop it (0.7.0). Start it inapp/(paths in the prompt relative to that), or ask Maki for a way to make those writes ask; part of the approval decision above
ACP:
- Gemini CLI, opencode, goose via ACP or native adapters; each verified live, then turn ACP on by default for the ones that pass
- Over ACP, suppress
session/updatenotifications untilsession/loadanswers, keeping replayed messages and tool activity out of the current turn — commit38ffc39be084 - Over ACP, turns report 0 tokens: the protocol sends only
usage_update(context used and size). Read real counts where an agent keeps them (Maki:token_usagein its session file under~/.local/state/maki/sessions/) or say the count is unknown instead of 0
Finding them:
- #105, #135, #259, #385 and #423 predate 0.3.0, which looks in more places, and came through
the ticket relay, so nobody can ask them for
doctoroutput. Close them pointing at 0.3.0, and add search paths only for a new report that has that output - #423: an agent that is set up but switched off is never offered again, so "No coding agent was
found" stays on screen with Claude installed: the setup card and "Look again" leave out every saved
AI by name, on or off. Offer a switched-off agent as found and turn it on, keeping its options.
Cause is not confirmed on the reporter's machine (needs their config.toml); the fix covers the
switched-off case that the code leaves out (
is_on,leave_out_set_up,builder_backend). — claimed: fabox-os, branch bugfix/423-disabled-agent-not-offered - Say plainly when what's found is the Claude desktop app rather than the
claudeprogram (#259, a guess at its cause) - macOS: check that
Claude.appis where the desktop app is found (#259's hint is untried there) - #259's hint, the gaps left in review of #781: it's hidden when only
~/.ssh/configsuggestions are found, a Store install without itsclaude-desktopalias isn't seen, and a desktop app'sclaude.exeon PATH is taken for Claude Code (#824) — claimed: FuryNocturn, branch bugfix/r2.2-claude-desktop-hint-gaps - macOS:
/Applications/Antigravity.appended up as the agent's command (#467). Found: the "Where is the program?" picker (pick_file) returns the bundle, andset_commandsaved any path. Refuse a folder or app bundle when it's saved (set_command), and don't run one that was saved before (program_ofuses the usual name). Tests:an_app_bundle_is_refused_as_an_agents_program,program_ofcases. — claimed: fabox-os, branch bugfix/467-agent-program-app-bundle - Windows: look in
%LOCALAPPDATA%\agy\bintoo (#62) — claimed: Afloat16, branch bugfix/r2.2-antigravity-search-20261103 - Windows:
usual_windows_dirs_come_after_pathcounts Antigravity's second folder (#748 added it; the test runs only on Windows, which CI doesn't) - CI runs the tests on Ubuntu only, so
cfg(windows)tests (search paths, consoles, Job Objects) are checked only on contributors' machines: run them on a Windows runner too - "Claude Code on " is offered for any SSH host, whether or not it has Claude Code (#62)
- Gemini CLI's sign-in reportedly refuses individuals now (#400): check, and stop recommending it in the builder setup dialog if so
- DeepSeek Harness: an adapter runs
dsh --profile headless "<task>"and hands it our tool server as a generated--patchoverlay for that run (its own project, inbox and mode), since the harness reads tool servers only from profile layers; a question gets none and the person's own profile is left untouched. Verified with adapter and overlay tests and a live follow-up website build whose journal showsrebuild_check,core_unknown_addandcore_handoff_writegoing through our server — claimed: 0231071, branchfeature/dsh-agent-adapter - DeepSeek Harness: send tasks on stdin so multiline and long messages can start through the
Windows desktop launcher (see the argv note above)
— claimed: gtxy27, branch
bugfix/r2.2-dsh-stdin - DeepSeek Harness: no resume (its headless profile keeps no session), and a build on the harness's desktop profile is untried
-
setupwrites the plainmcpcommand, so it drops an--all-toolsan entry already has, anddoctorthen calls that entry "a different copy or shape of the command" and suggests the very command that breaks it. The installed skill now warns against runningsetupon such an entry;doctorandsetupshould keep the person's extra arguments instead — claimed: 18923837179wjt-dot, branch fix/setup-skill-warns-before-setup-drops-all-tools
R2.4 Runtime study for apps and games (with consent)
Owner: maintainers · Status: in progress
- UI tree on Linux (AT-SPI)
-
native_window_captureon macOS: no Screen Recording is refused before the program starts (it was "didn't show a window", which said to wait longer), and the name fallback matches whole words and never the system, sowinno longer pictures the menu bar -
native_window_capturekeeps looking for up to 10 s pastwait_secswhen no window is there yet, instead of deciding from one look; tried with a program that opens its window late - Close a pictured program politely before killing it: a helper it started through launchd outlives the process group, and can stay running with the program's data files open
-
apple_calls_observe: the breakpoint callback is one valid line (it had a literal backslash andnbetween statements, so lldb stopped before launching, #251), covers every breakpoint instead of only the first, a failed lldb run is an error and not "0 calls", and names or arguments with line breaks are refused (they'd add commands to lldb's command file). The callback body was compiled with Python; lldb itself was not run - Run
apple_calls_observeon a Mac against a small program withlldb(the #251 probe): the hit foralphais recorded, and a program lldb can't debug says so - Window capture tried on macOS and Linux (only Windows so far); games behind a launcher
- Short scripted input sessions (enigo) to record screens of the original
R2.5 Behaviour checks for non-web builds
Owner: open · Status: in progress
- Behaviour checks that use the sheet: run the build with scripted input and measure values (speed, jump height, fire rate) against it (needs R2.4's scripted input)
- Hold the self-test to what the check can see too: picture the game after its scripted
input and require it to differ from the
--screenshotpicture (a report alone is the build's own word)
R2.7 The request decides how to build
Owner: open · Status: in progress
- Make: a game whose request names a technology the stack table doesn't know (Phaser, Bevy…) still needs the default engine installed to start, though the builder is told to use what was named; ask the person, or let such a game start without an engine
- The website check's missing-site note names every place
site_diraccepts (dist/, build/, out/, or a plain index.html with no package.json) and the app's real place, "app/" or "the project folder" (#494) - A project edited in place (Work on my project, or
rebuild --existing edit) keeps what its files are built with, unless the request names another stack: an engine's project file, then package.json and Cargo.toml dependencies, thenfound_byinstacks.toml, and a project whose dependencies name no stack there is left to the builder. A request that read as a game used to ask for Unity even in a Godot project, and an app got the kind's default — claimed: xerenon, branch bugfix/r2.7-stack-from-files; PR #755 - A person's own project (edited in place, nothing to recreate) is judged by what it does for
itself: its own build, then its own tests, through
[[toolchain]]inengine/src/toolchains.toml(Rust, Node.js, Python, Go, .NET, Maven, Gradle, Make). Nothing is added to it for our sake: no--screenshot,--self-testor mechanics sheet. A site stack with nothing to recreate is a website, whatever the request says — claimed: xerenon, branch feature/r2.7-own-project-checks; PR #760 - Run the Go, .NET, Maven and Gradle rows of
[[toolchain]]on real projects, and every row on Windows (windows_program): Rust, Node.js, Python and Make were run for real, on a Mac only. Also a person's own Unity project: it's opened in batch mode, which should exit with an error when its scripts don't compile (not run: no Unity on that machine) - A person's own tests that already failed before their request fail the check too, so the builder is sent to fix them. If that proves a nuisance, run them once before the first turn and hold it only to what it broke
- New projects build through the shared
[[toolchain]]catalogue. Python root entrypoints, Go's rootmain.go, and one root.csprojalso launch for screenshots, self-tests and Play; existing packaged programs, Rust, Electron and game-engine paths keep priority — PR #950 — claimed: Codex, branch bugfix/r2.7-new-project-toolchains - Runtime window capture follows runner descendants (including
go run) and reads controls from the same process that supplied the picture, without falling back to unrelated names — PR #950 - Shared toolchain commands make project-local executables absolute before changing the working directory, preserving virtualenv paths for relative project placements — PR #950
- Launch a Go project with one
cmd/<name>/main.goas a package when the root has nomain.go— claimed: Afloat16, branch feature/r2.7-go-command-layouts - Extend launch recipes to Python package scripts and Java build outputs; build checks already use their toolchain commands
- Audit relative project placements in build, option-run and Play paths too: runtime window capture now normalizes its request paths, but earlier runner paths can still be relative
- "player" alone doesn't make a request a game: a video or music player is an app (#953) — claimed: justinjohn0306, branch bugfix/953-player-not-a-game
- System-tool and media-player phrases keep the app target: "a low level disk tool", "an MPEG 2 player" and "a single player app for movies". Independent game cues still apply — claimed: Afloat16, branch bugfix/r2.7-media-and-system-apps
R2.8 Clean room means clean room
Owner: maintainers · Status: in progress
- A remote (SSH) builder needs a craft-anything-cli that knows
--deepand--clean-room: check its version first and say so when it's older
R2.9 Everyday use: typing, projects, settings
Owner: maintainers · Status: in progress Gaps found while using the app for real projects.
Each project chooses, per kind of action, what the builder's approvals get (allow, ask or refuse): reading the web, commands, files outside the project, pushing and publishing, deleting outside the project, system changes, other tools. Unset kinds follow Auto mode as today. Found when Auto mode refused every
git pushwithout a card, so the person's OK in the prompt box never reached the check, and refusedgh repo viewas publishing too. Tried live by a maintainer on Windows 10 (a release build ofmainatb700e1d, 2026-11-06): a builder's push was approved and went through — claimed: tanish2k09, branch feature/r2.9-project-approvalsAgents that don't ask our approval tool (OpenCode
--auto, Hermes-z, Maki's-p) aren't held to a project's approval choices either; part of R2.2's approval decisionWhile "commands" are allowed (Auto mode's default), a builder can run any program. Commands that name
craft.json, the inbox or the approvals tool count as system changes, and our CLI started inside an agent calls tools as an agent, but a script that builds the path, or unsetsCRAFT_ANYTHING_AGENT, isn't caught. Keep the person's choices where a builder's shell can't write them, or say plainly that "commands: allow" trusts the builderA typed note ("yes, push it") still reaches the builder only as text: say in the card's place, when the builder is refused, how the person can allow that kind
The project page and a slash command change those choices (like
/btw): the Approvals menu under the prompt box and/approvals <kind> <allow|ask|refuse|default>— claimed: tanish2k09, branch feature/r2.9-approvals-uiThe new approvals strings in every catalog were written with an AI; not reviewed by native speakers yet — pt-BR claimed: marlonferreira1800-hue, branch docs/r2.9-approvals-pt-br-review
Keep New project's content column fixed while the vertical scrollbar animates in or out, hiding it when there is nothing to scroll, so switching tabs never shifts the page horizontally — claimed: Codex, branch bugfix/r2.9-home-scrollbar-gutter
Agent paragraphs with glossary terms and long file links wrap together without pushing their card or later messages under the sidebar (#523); Project and Ask bounds checked at two sizes — claimed: carl-dourado, branch hotfix/r2.9-523-message-width
Create atomic-write temporary files exclusively with short names distinct from the destination and private Unix permissions from creation. — claimed: Afloat16, branch bugfix/r2.9-atomic-temp-creation-20261103b
A prompt brought back with Up keeps its pictures (their paths are absolute, so they don't come back after the project is moved)
Change a project's builder from its page (a tool, so the CLI and MCP can too) — in PR #245
chat --backend <a name that is turned off>andcomparerefuse it before model or agent setup:AiHub::startchecksenabledand says how to turn it on or choose another — PR #1008; claimed: Codex, branch bugfix/r2.9-disabled-chat-backend (the API regression uses invalid tunnel settings to detect setup before the guard)Ctrl+V (Shift+Insert, Cmd+V) in a prompt box that takes pictures pastes a copied image (saved as a PNG) or copied picture files; attachments show as square picture tiles, the whole picture on hover, decoded on a worker
A picture over 20 MB dropped on a prompt box or picked in its dialog is refused when it's added, as a paste does (
workspace::attachable_picture), before anything is copiedWhen the builder decides a question, show the question and its choice as an answered card in the feed (the run sees only the tool call today; the builder's words say what it chose)
A question's own cost: a forked Claude Code session reports the whole session's running total, so a forked answer shows none; count questions in the project's usage
"Make this change" on an answer that suggests one (sends it as a normal message)
"Just ask" is told it has fewer tools than the builder and that the build may be running beside it, so it no longer reports the build's tools as disconnected or its work as cut off
"Change it" knows what was just talked about in "Just ask": the next message tells the builder the latest few questions and answers, once, and a Clean room build only those asked in Clean room (#857). Not checked in a live build
A note sent while a build is running tells it what was asked in "Just ask" (#857): the note's hand-over file carries the answers, and the after-turn path puts them before the notes; only answers the run's mode may show, marked told when read or sent — claimed: AlexJoaquimPereira, branch bugfix/r2.9-note-just-ask; commit 78d3963
Record which mode a builder's session ran in:
agent_sessionis kept across modes, so a Clean room build or question can resume a session from a Deep build and carry what it read, and such a question's answer is recorded as asked in Clean room. Don't resume one for Clean room#504: the builder's plan card can be folded away with its X, or with
project_plan_fold(between builds: a fold during a build can be overwritten by its next save). The steps stay in the conversation (FEED_VERSION4,FeedItem::PlanDismissed), and a later run's checklist shows again — claimed: claude, branch bugfix/r2.9-plan-card-dismissA run that ends with steps still open marks them as stopped on its card, instead of showing "Doing" until someone folds it (#504: Claude's runs left the plan unticked) — claimed: rust-craft, branch feature/R2.9-stop-open-steps; PR #876
#666: a long line in a feed's tool-work row (a Deep-mode decompile, a status note) ran off the window and pushed the feed left; the header title can't wrap in egui, so it's the note's first line now, cut at 90 characters, and the full note wraps in the body — claimed: claude, branch bugfix/r2.9-feed-text-overflow
decision A program started from the CLI (
project_play) keeps the shell's output pipe open, so the command doesn't return until it closes (#207). Every fix found needsunsafeor an unstable API: an exception to rule 9, or a limit we document. Meanwhile a reproducing test marked#[ignore]and a line in the CLI docs are welcomemacOS: the window drew Windows-style controls at the top right, since the app turned off the system's title bar (
with_decorations(false)); it now keeps the system's buttons, frame and resize edges under a see-through title bar, which is 32 pt there so the buttons sit centred in it (#20)macOS: the app's icon is bigger than the system's (#20)
Show a command's output while it runs: Claude Code reports a tool's result only when it ends, so a long
cargo buildis one quiet line until then. Watching the agent's child processes from outside could say what's running and for how longRelease the keep-awake helper before a rebuild reports
Finished, waiting up to five seconds for its supervised cleanup and logging a timeout. Real subprocess stand-ins checked on Linux (#541) — claimed: Afloat16, branch bugfix/r2.9-keep-awake-cleanup-20261103Recheck keep-awake cleanup with a real Windows CLI
rebuild: the helper once outlived it by more than 9 minutes and held the caller's output pipe; not reproduced in 8 more runs. Found: closing the desktop app during a build ends it without dropping the guard, and the helper (an endless PowerShell loop, started with a week's time limit) carried on. It now ends by itself once the process it serves is gone: the Windows loop checks that process every 30 s, and macOS usescaffeinate -w. Live on Windows: a CLI build killed outright, its helper ended 30 s later. Not run on a MacZoom (Cmd/Ctrl + / − / 0) between 50% and 200%: egui's own limits are 20% and 500%, where text is unreadable or the window has room for almost nothing — claimed: sanketdofe, branch feature/r2.9-zoom-bounds
Find why a site capture lost its browser ("the WebSocket is closed", 419 ms into
web_site_capturefor a second page, seen once; the same capture passed 41 s earlier)Delete a private browser's temporary profile when it closes or fails to start (#810) — PR #822
Stop private browsers that outlive the app after an abrupt exit, and investigate the Brave Helper crash (#810); both still need macOS reproduction
Claude Code's Ultracode switch (a setting, not an effort level; the effort environment variable we use can't turn it on)
A model the list doesn't offer (
opusplan,[1m]variants, one another harness serves) can be typed in beside the model list in Settings → AI, asai modelalready allowed (#470)The preview server answers only requests addressed to this computer (a Host and Origin check against DNS rebinding), streams large files instead of reading them into memory, sends no body for HEAD on a missing file, and caps open connections and the time a request may take to arrive (#597, #598, #600, #601, by Afloat16)
The command palette stays where the wheel scrolls it; it follows the selection only when the arrow keys or typing move it (#584)
Linux: draw on the GPU that drives the screen. egui-wgpu's default picks the discrete GPU, so on a laptop whose integrated GPU drives the panel the window took 20 s to open and the whole desktop lagged (#571). Pick by the connected displays in
/sys/class/drm: the built-in panel's GPU, else the only GPU with displays, else the discrete one;WGPU_POWER_PREFandWGPU_ADAPTER_NAMEstill decide when set. Checked live on an AMD + NVIDIA laptop only (#578) — claimed: Nghiathan13, branch bugfix/r2.9-linux-gpu-choiceLinux GPU choice on a desktop with monitors on two GPUs, and on an Intel + NVIDIA laptop: covered by unit tests on fake
/sys/class/drmfolders, not run on such a machine yet (#571)"Work on my project" on New project, beside "Make my own version" and "Just look inside": pick the folder of a project you already have, say what to change, and its files are edited in place, with nothing asked; the project takes the folder's name, an empty folder is refused and a folder that's already a project opens. Until now that was only reachable through "Choose a folder…" under Make's Start, on a screen that asks what to recreate — claimed: xerenon, branch feature/r2.9-work-on-my-project; PR #686
Leaving "Work on my project" forgets a folder it would edit in place and drops a folder dialog that ends later, so a version made on Make can't edit that project, and an address in its request isn't read as an original to recreate (review of #686) — claimed: xerenon, branch bugfix/r2.9-own-project-review; PR #750
build_files_listlists an edited-in-place project from the project folder itself, not a missing or nestedapp/, and leaves Craft Anything's own records out — claimed: AlexJoaquimPereira, branchbugfix/r2.9-build-files-in-place; commit 1ffeb40Choose Claude Code's folder (
CLAUDE_CONFIG_DIR) per AI, so one computer can build with two Claude accounts; an app opened from the Dock never sees the shell's variable. Settings → AI (a typed folder, and Add another) andai folder; live-checked from the CLI with an empty folder (not signed in, and the fix names the folder) overriding the shell's own variable, and by a build in the app with a typed~/folder. A sign-in that expired says so instead of "isn't signed in" (the folder's.claude.jsonstill names the account) — claimed: Claude (Opus 5.5), branch feature/claude-config-dirA live build with two Claude Code AIs signed in to two accounts, and one over SSH with a
~/folder (the remote line is unit-tested, not run on another computer)The Advanced console's status stays clear of its tabs when the console is narrow — claimed: justinjohn0306, branch bugfix/r2.9-console-status-overlap
Keep the builder-availability smoke regression independent of live agent discovery. — claimed: Afloat16, branch test/r2.9-builder-discovery-fixture-20261104 Test fixture implemented with controlled discovery; automated validation remains pending.
macOS: a menu bar with the app's own actions (open a file or folder, new project, settings, Simple / Advanced, Clean room / Deep, the command palette, help); the system's menu bar showed only the app menu — claimed: sanketdofe, branch feature/r2.9-macos-menu-bar
A Traditional Chinese README (
README.zh-hant.md) translated fromREADME.mdand linked from the language switcher in every README — claimed: monarchsofchaos004-gif, branch docs/readme-zh-hant; PR #926macOS: an Edit menu (Undo, Cut, Copy, Paste, Select All). The keys work in text boxes, but the system's Edit items need the text box to answer them, so they'd show greyed out today
The Advanced console's prompt box starts tall enough to read what's typed (#942) — claimed: justinjohn0306, branch bugfix/942-console-input-height
A French README (
README.fr.md) translated fromREADME.mdand linked from the language switcher in every README (#831)Vietnamese letters with stacked marks (ạ ả ấ ế ơ ư ữ…) aren't in the bundled DM Sans or egui's default fonts: without craft-fonts they show as boxes, and with it each such letter comes from Noto Sans CJK while the rest of the word is DM Sans. Bundle a DM Sans build that has the Vietnamese set (with its OFL notice) so Tiếng Việt renders in one face
R2.10 Pinned builds
Owner: maintainers · Status: in progress Keep a build that works at hand: pin it, and open it from the sidebar whatever the builder does to the project next.
- Say when a light pin ended up copying what it meant to share (a drive without hard links) — in PR #220
- Pins of one project sharing with each other, so a third frozen copy of the same assets takes no more space
- Pin a website (a frozen copy of the built site, each served on its own port)
- A pin started before the app restarted shows as running (keep its process id and start time in the pin, and check both), and can be stopped
- Cap a run's log (
last-run.log,play.log): it grows as long as the program prints — in PR #218 - Opening a pin that's already running brings its window to the front, instead of a message
- How much space pins take in Settings
M3 Engine depth
An open lane: anyone may take these at any time.
R3.3 Frida sessions for real
Owner: open · Status: todo
- Replace the in-memory session bookkeeping with real attach, resume and detach
R3.9 Game assets out
Owner: open · Status: in progress
- Unity: sounds out (the stored stream, usually FSB5; PCM ones as WAV)
- Unity: crunched textures (DXT1Crunched, DXT5Crunched; two of the games checked use them)
- Unity: the float formats (RHalf, RGHalf, RGBAHalf, RFloat, RGFloat, RGBAFloat) — claimed: Afloat16, branch feature/r3.9-unity-float-textures
- Unreal 5 and Oodle: IoStore packages, unversioned properties and Oodle-compressed paks
- Game files: save pictures from IoStore containers (needs the Unreal 5 support above)
- Unreal: sounds and meshes out (so a Deep-mode rebuild needs no outside converter for them)
R3.11 Single-file apps, game files and stand-in backends
Owner: open · Status: in progress
- Validate the remaining loopback fixture corrections for complete and fragmented request
headers, negative probes and stand-in cancellation on the current main
— claimed: Afloat16, branch test/loopback-fixture-stability-20261103
Revision
cd468aca, synchronized with main41d2175e, passed full Linux CI. The fragmented-header regression withholds the final byte until the next read and keeps the existing request and response assertions. The request-encoding, negative-probe, cancellation and builder-discovery smoke checks passed in the same run. - Verify stand-in session cleanup by the original answerer's lifetime rather than polling a released port that another listener may reuse — claimed: Afloat16, branch test/r3.11-standin-session-lifetime-20261103
- Stand-in backend: writes that change later reads (recorded answers are replayed as they are)
- Tell the builder about the stand-in backend in a website's brief, and start one from a recording
- Stand-in backend: retain 304 samples without a usable cached body for diagnostics, skip them
during replay, and report
cached_body_unavailablewhen the best matching group has no usable answer. Build results count them; cached bodies still answer as 200 (#543) — claimed: Afloat16, branch bugfix/r3.11-standin-empty-304-20261103 - Godot format 4 and IL2CPP metadata 38 and 39 are header-only; read their tables once a real file can be checked
- Bun: take source maps and bytecode out as files; a standalone
.pyz -
native_source_folder_writeuses the built-in decompiler only: use Ghidra in Deep mode when it's there - Keep stand-in response samples distinct when headers, content types or body-availability metadata differ.
Compare header names without case and ignore ordering between different names, preserving each name's value order.
At the sample cap, a usable answer can replace an unavailable 304 for the same path, query and request key;
other requests keep their missing-body evidence. Refused bodies cannot replace that evidence.
Compare complete samples before charging retained bytes, using #717's retained-budget accounting.
Historical revision
f79dc0bdpassed full Linux CI: all five original metadata regressions and nine header and capacity regressions passed. All 24 combined replay regressions pass locally on the refresh ontocf414ff4. See this PR's checks for the refreshed revision's complete gate. — claimed: Afloat16, branch bugfix/r3.11-standin-sample-metadata-20261103b -
native_publisher_showhashes the file minus the signature fields, not section by section; a file with unusual gaps between sections can read as different - Enforce stand-in per-body and total byte budgets on the text that is actually retained after redaction;
deduplicate consecutive samples before charging the capture budget.
— claimed: Afloat16, branch bugfix/r3.11-standin-redacted-budget-20261103b
Revision
d601bc4epassed full Linux CI: all four retained-budget regressions and six deduplication regressions passed.
R3.12 Format parser correctness
Owner: open · Status: in progress
Require a complete DNS question's type and class fields before retaining it, and keep truncated questions out of packet inventories while reporting the incomplete message — claimed: Afloat16, branch bugfix/r3.12-dns-question-class-review-20261104
Reject impossible IPv4 header and total lengths. — claimed: Afloat16, branch bugfix/r3.12-ipv4-header-lengths-20261105
Validate the version in IPv6 frames. — claimed: Afloat16, branch bugfix/r3.12-ipv6-version-20261105
Stop decoding extension headers inside noninitial IPv6 fragments. — claimed: Afloat16, branch bugfix/r3.12-ipv6-fragment-payload-20261105
Enforce the expanded DNS name wire-length limit. — claimed: Afloat16, branch bugfix/r3.12-dns-name-wire-length-20261105
Report truncated DNS TXT strings as incomplete. — claimed: Afloat16, branch bugfix/r3.12-dns-txt-truncation-20261105 Preserve empty ANY/NONE UPDATE prerequisites with zero TTL; keep TXT validation scoped to parsed answers/prerequisites. Preserve empty multicast/unicast DNS-SD TXT records using wire-label service instance names.
Read HTTP Host fields only from the header section. — claimed: Afloat16, branch bugfix/r3.12-http-host-boundary-20261105
Keep capture flow timestamps in chronological order. — claimed: Afloat16, branch bugfix/r3.12-capture-flow-time-range-20261105
Reject incomplete simple packet block payloads. — claimed: Afloat16, branch bugfix/r3.12-pcapng-simple-block-length-20261105 Count rejected lengths against the packet scan limit and summarize them once at the end of the walk.
Bound rejected packet records, validate timestamp fractions and check pcapng section-local interfaces. — claimed: Afloat16, branch bugfix/r3.12-capture-timing-framing
Keep ZIP directory entries within the declared directory. — claimed: Afloat16, branch bugfix/r3.12-zip-directory-bounds-20261105
Refuse split ZIP archives without their other volumes. — claimed: Afloat16, branch bugfix/r3.12-zip-multidisk-20261105
Check Unity bundle blocks against their declared output lengths. — claimed: Afloat16, branch bugfix/r3.12-unity-bundle-decompressed-length-20261105
Locate Unity bundle metadata within the declared bundle size. — claimed: Afloat16, branch bugfix/r3.12-unity-bundle-declared-size-20261105
Check that Unity bundle compressed blocks end before the trailing block table, or the declared bundle end for front-positioned tables, so bytes outside the data region cannot supply block contents. — claimed: Codex, branch bugfix/r3.12-unity-bundle-data-region; PR #1021
Reject Unity serialized objects outside their declared file. — claimed: Afloat16, branch bugfix/r3.12-unity-serialized-object-ranges-20261105
Preserve compression method slots in Unreal pak indexes. — claimed: Afloat16, branch bugfix/r3.12-unreal-pak-compression-slots-20261105
Require complete terminators in Unreal FString values. — claimed: Afloat16, branch bugfix/r3.12-unreal-fstring-terminators-20261105
Count every compression block spanned by an IoStore chunk. — claimed: Afloat16, branch bugfix/r3.12-iostore-unaligned-chunk-blocks-20261105
Reject truncated IoStore fixed headers. — claimed: Afloat16, branch bugfix/r3.12-iostore-complete-header-20261105
Reject invalid marshal long integer digits. — claimed: Afloat16, branch bugfix/r3.12-pyc-long-limb-validation-20261105
Check expanded RGBA image sizes and enforce output budgets before allocation. — claimed: Afloat16, branch bugfix/r3.12-pixel-rgba-output-overflow-20261105
Count empty tensor shapes consistently across axis orders. — claimed: Afloat16, branch bugfix/r3.12-tensor-zero-dimension-product-20261105
Find APK signatures through valid ZIP comments and framed central directories. — claimed: Afloat16, branch bugfix/r3.12-apk-sign-eocd-comment-boundary-20261105
Keep Mach-O load-command facts within their command. — claimed: Afloat16, branch bugfix/r3.12-macho-load-command-framing-20261105
Keep Mach-O function starts within their declared payload. — claimed: Afloat16, branch bugfix/r3.12-macho-function-starts-payload-boundary-20261105
Keep code-signature children and verification reads within their declared superblob. — claimed: Afloat16, branch bugfix/r3.12-codesign-superblob-length-boundary-20261105
Keep PE optional-header reads within the declared header. — claimed: Afloat16, branch bugfix/r3.12-pe-optional-header-framing-20261105
Keep PE resource metadata within the resource directory. — claimed: Afloat16, branch bugfix/r3.12-pe-resource-directory-boundary-20261105
Keep Go function tables and names within their section. — claimed: Afloat16, branch bugfix/r3.12-go-pclntab-section-boundary-20261105
Preserve complete Unicode characters in shortened .NET strings. — claimed: Afloat16, branch bugfix/r3.12-dotnet-user-string-surrogate-limit-20261105
Locate ELF core mapping names after the complete declared table. — claimed: Afloat16, branch bugfix/r3.12-elf-core-file-name-count-cap-20261105
Keep DOS relocations inside the declared executable header. — claimed: Afloat16, branch bugfix/r3.12-dos-relocation-header-boundary-20261105
Parsers refuse what they used to half-read: binary plists keep objects inside their object area and reject unrepresentable widths, base64 (plists, source maps) refuses damaged tails and incomplete quanta through one shared decoder, OIDs must be minimally encoded, and string offsets past the address space are left out. Tar keeps old GNU timestamps out of names and PAX sizes across other metadata records; CRC cpio checks member data; source maps decode percent-escaped data URLs and report indexed-map columns in the generated file (#606–#615, by Afloat16)
Bound and cancel CRC cpio verification across firmware probes, and reject non-hex checksum fields — claimed: Afloat16, branch bugfix/r3.12-bound-cpio-crc-scans-20261103
More parsers keep to the bytes they declare and refuse impossible values: signed LEB128 past 64 bits, damaged modified UTF-8 (valid text around it survives), widened Java instructions and invocation operands, DEX register lists and array widths, SQLite text in the database's encoding with cells kept inside their page and overflow cycles stopped, Android string pools, ZIP output limits for stored entries, decoded sizes and ZIP64 fields, ASAR header framing, ROMFS dot roots, cramfs, ext and JFFS2 sizes and holes, PyInstaller modules outside their PYZ, JSON config edits (comments kept, every duplicate key removed), and UDP, TCP, DNS over TCP, pcapng and TLS lengths (#688–#715, by Afloat16)
Check ClientHello vector lengths and extension values before reporting the handshake as complete. — claimed: Afloat16, branch bugfix/r3.12-tls-client-hello-values
Address the combined parser review: preserve partial Java reference coverage, enforce DEX signed 32-bit LEB128 limits, recover legacy AAPT UTF-8 lengths within their pool with a shared recovery probe limit, and distinguish missing ext indirect metadata from sparse holes. — claimed: Afloat16, branch bugfix/r3.12-parser-review-followups-main-20261104 Shared-budget revision
29c48e03passed full Linux CI. The merge with main41d2175eawaits a new CI run.Keep DNS answer names inside their declared RDATA, including compressed names and MX exchanges, and report malformed records as incomplete instead of inventing answer values, while preserving legal empty DNS UPDATE RRset prerequisites — claimed: Afloat16, branch bugfix/r3.12-dns-rdata-names-review-20261104
Keep address and service records attached to their owner and validate the record's wire fields. — claimed: Afloat16, branch bugfix/r3.12-dns-record-evidence
R3.13 Web inspection correctness
Owner: open · Status: in progress
- The HTML scanner finds script preloads in
reltoken lists, decodes asset URLs once (numeric and named references), keeps script and title text through lookalike closing tags, and skips raw-text elements, markup inside attribute values and custom tag names such as<script:widget>(#591–#596, by Afloat16) - HAR reports redact credentials whose query names are percent-encoded (and the stand-in backend matches older recordings the same way), redact declared secrets in every kept field (names, methods, hosts, versions, timings), count body bytes before text conversion, and template only real calendar dates in API paths (#602–#605, by Afloat16)
- Preserve empty query pairs when matching older stand-in recordings, without treating raw
query separators as pairs — claimed: Afloat16, branch
bugfix/r3.13-preserve-empty-query-pairs-20261103
M4 Release
R4.1 A macOS CLI that opens
Owner: maintainers · Status: todo
- decision A standalone macOS CLI that can carry its ticket, such as a stapled
.pkg; that needs a Developer ID Installer certificate in the release secrets
R4.2 Managed tier
Owner: maintainers · Status: todo
- decision Managed tier: account backend, usage metering (Claude API under our own key only)
- Before the managed tier or any paid offering: confirm the Claude Code orchestration question with Anthropic (claude-code.md)
Parked
Looked at and not committed to; none of this is open work. Say in the issue if you want one.
- From the tool survey: firmware image kinds (ESP32, U-Boot, device trees, Intel HEX), Erlang BEAM, GPU code objects, kernel modules, GameCube DOL, Mac crash reports, SAZ and qlog captures, protobuf bodies, R8 mappings, JNI links, MCP over HTTP
- Ideas from issues: resume by itself after a usage limit (#476, the one tied to a real problem), a source folder as a target (#480), actions after a build ends (#495), WebAssembly disassembly (#175), an HTTPS intercepting proxy (#164), VM and Windows Sandbox detonation (#144, #163), an adapter for another disassembler (#147)
- More providers and agents (#14, #63, #97, #162, #170, #186, #206, #470): R2.2's table comes first, so each new one is a small addition rather than another half-wired agent