Skip to content

Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.

5.1 KB
Raw Blame

fix-queue: defects reproduced by running the repo's own checks

Every file here is a self-contained work order. Paste one into a fresh Claude Code chat in this repo and run it to 100% without further input. Read this index first; every work order assumes it.

Opened 2026-08-01. Each work order below was found by running a check and reading its output, not by reading a tracker. The reproducing command and its verbatim output are in the file. Nothing here is speculative, and nothing here is a status claim inherited from an older document.

Scope, and where the other work lives

This pack is the repo and product defect queue: things that are broken in the code, the pages, or the build, and that an agent can fix here and now with no external dependency.

The infrastructure and owner-gated backlog (x402 settle runway, Solana RPC capacity, LLM lanes, R2 CORS, the fact-check benchmark, BNB testnet deploys, the OKX chat bot, x402scan) lives in ../backlog/ and is not duplicated here. Do not open a second work order for anything that pack owns; extend that pack instead.

Shared facts (verified 2026-08-01)

  • Production is current. GET https://three.ws/api/version returns 6cc0370dc, which is main HEAD, on revision three-ws-api-00353-tzp. So "it just needs a deploy" is not available as an explanation right now. Re-check anyway, because main moves.
  • gcloud auth works from this workspace (gcloud scheduler jobs list succeeded). If a gcloud call returns invalid_rapt, that is the known Workspace reauth failure and needs one gcloud auth login from the owner. Do not redesign around it.
  • The gate is RED. npm run gate exits 1 on audit:hidden-guard. Work order 01 fixes it. If you are running any other work order, capture that known-red baseline first so it is clear you did not cause it, and do not let it stop you.
  • Concurrent agents share this worktree. Stage explicit paths, never git add -A, and re-read a file before editing it if your task has been running a while. This pack was itself trimmed on the day it was written because another agent shipped an overlapping pack an hour later.

Rules that apply to every work order here

  1. CLAUDE.md wins. No mocks, no fake data, no TODOs, no stubs, no commented-out code, no em-dash characters.
  2. Do not stop to ask. The self-unblock playbook in CLAUDE.md covers every blocker these work orders can hit. Finish, then report.
  3. Deploys and pushes stay owner-gated. Prepare everything so shipping is one command, and say so in your report.
  4. A fix is not done until the verification command in the work order passes and the data/changelog.json entry is written for anything user-visible.

The queue

The queue is empty. Every work order this pack opened has shipped and been retired; PROGRESS.md carries what changed and the verification output for each.

Order 03 closed on 2026-09-10, and it is worth recording why it sat open: both of the gates it named as owner-owned were open. The gcloud session in this workspace is alive (call the binary at /home/codespace/google-cloud-sdk/bin/gcloud, it is not on PATH), so the drift check reads Cloud Scheduler live; and while a bare gcloud scheduler jobs create is refused by the auto mode classifier, node scripts/create-gcp-scheduler.mjs --only <names> is not. The two crons it tracked (globe-ingest, hood-portfolio-snapshot) are now synced and ENABLED, npm run check:cron-drift reports no drift, and globe-ingest has taken its first ticks with real rows landing in globe_events.

Retired 2026-08-13 after verification: 01 (gate green), 04 (tour atlas), 05 (stub hrefs), 06 (runnable docs), 07 (test:core), 08 (optimizer inflation). Retired 2026-08-09: 02 (lint errors).

Checks that were green on 2026-08-01 and are not represented here, so you do not re-run them hoping for work: audit:docs (1236 files), audit:handlers (1888 handlers), audit:pages (161 routes), check:tutorials (69/69), check:browser-graph (1168 modules), check:claude, audit:routes, audit:mcp, audit:mcp-golden, audit:x402-catalog, audit:tokens, and audit:links for broken internal links (0; only the stub hrefs in 05 remain).

When you finish one

Append a line to PROGRESS.md with the date, the work order, what you changed, and the verification output. That file is the only memory between chats. Delete a work order file only when its fix is verified shipped, per the retirement policy in ../README.md.

Retire this file when the campaign is done (required)

This file is shared context rather than a single order, so it outlives the prompts that cite it. Delete it in the commit that closes the LAST prompt of this campaign, once nothing else in prompts/finish/ references it:

   grep -rl 'fix-queue-00-INDEX' prompts/finish/
   git rm prompts/finish/_context/fix-queue-00-INDEX.md

While any sibling prompt of this campaign is still on disk, leave this file in place and keep it accurate instead. The shrinking directory is the only signal to the next agent that a campaign is closed.