Garry's Opinionated OpenClaw/Hermes Agent Brain
TypeScript
Merge remote-tracking branch 'origin/master' into capy/gbra95-guard-definer
garrytan0f0c8888eb74
diff --cc CHANGELOG.mdindex 48b6de58a2,9db3815653..48540d4358--- a/CHANGELOG.md+++ b/CHANGELOG.md@@@ -10,40 -10,43 +10,77 @@@ credits are retained; no result has beeidentifiers and attribution are available in the pre-removal Git revision`6040075c6cb95be5881cc2e1b76ef7d71f4e5d29` (retained on 2026-09-23).- ## [0.60.467.0] - 2026-10-26++## [Unreleased]++**A temporary table can no longer slip purged or forgotten content past gbrain's guards. Before, any database session, including gbrain's own role, could create an empty table named like a purge list and have a purged claim, page or take stored again, or a forgotten fact saved active.**++gbrain keeps lists of what you purged (`gbrain forget --purge`) and what you forgot (`gbrain forget`), and database guards check every write against them. Those guards looked tables up in a way that let a session's own temporary tables come first. So a session could create an empty temporary table with the same name as a purge list, and the guard read that empty table instead and let the write through. Every gbrain function now looks at temporary tables last. Nothing else about the guards changed.++### How to get it++Upgrade and let the migration run (`gbrain upgrade`, or `gbrain apply-migrations --yes`). Then check:++```bash+gbrain doctor --json # function_search_path: ok+```++### What changed++| Situation | Before | Now |+|---|---|---|+| A session creates a temporary table or view named like a purge or withdrawal list, then saves purged content | the purged claim, page, take or derivation was stored again, and a forgotten fact was saved active, for the owner and any other writer | refused (or dropped, or stamped forgotten), exactly as without the temporary table |+| The migrating role does not own some gbrain functions | — | migration v277 updates the ones it owns, names the rest on stderr and carries on; `gbrain doctor` (`function_search_path`) prints the `ALTER FUNCTION` to run as each owner |++### What to watch for after you upgrade++- **Still open:** on Postgres, a write role with no row-level-security policy on a purge or withdrawal list reads it as empty, so its purged or forgotten writes still go through. Until gbrain#7659 lands, give such a role a `SELECT` policy on those lists for the sources it writes.+- An older gbrain that runs `gbrain init` or `gbrain upgrade` against the brain undoes this for the functions it re-creates. Doctor flags them; `gbrain init` or `gbrain upgrade` with this release, or the printed `ALTER FUNCTION`, puts it back.+- Procedures in `public` are left alone.++### Itemized changes++- `src/core/function-search-path.ts`: `FUNCTION_SEARCH_PATH_HARDENING_SQL` appends `pg_temp` to every gbrain function in `public` (`PINNED_FUNCTION_NAMES`, `prokind = 'f'`) the current role owns that pins `pg_catalog, public` or `pg_catalog`; another application's functions in the same database keep their search_path. ALTER only, no body changes. It is the last statement of the schema blob, runs in migration v277 (`function_search_path_pg_temp`), at the end of every migration run in `src/core/migrate.ts` (`initSchema`, and the locked `gbrain apply-migrations` / `gbrain upgrade` path; also when a migration fails, so an upgrade that stops early does not leave a re-created guard without it; best-effort, so a lock it cannot take after the migrations committed is named on stderr and left to doctor instead of failing the boot) and after `gbrain reindex-search-vector` re-creates the search-vector trigger functions.+- `src/commands/doctor/checks/function-search-path.ts`: new `function_search_path` check (fail) naming each function still without `pg_temp`, with the `ALTER FUNCTION ... SET search_path` to run as its owner. It never suggests an owner change.+- The purge and withdrawal guards are unchanged: `SECURITY INVOKER`, same bodies, same `BEFORE ROW` timing.+- Tests: `test/e2e/function-search-path-postgres.test.ts` (temporary table and view shadows for the owner and a non-owner writer, the migration from the previous version, a migrating role that owns all but one function, schema replay, lock timeout, an upgrade that fails at v276 through `initSchema` and through `runMigrations`), `test/function-search-path-pglite.test.ts`, `test/migrate-search-path-best-effort.serial.test.ts` (a failing search_path step after committed migrations does not fail the boot), `test/doctor-function-search-path.test.ts`.++ ## [0.60.467.0] - 2026-10-26++ **A CLI flag written as `--flag=value` is never silently ignored any more. Where a command reads that spelling, it applies; where it does not, the command stops with exit 2 and shows the spelling it reads, instead of running as if you had not passed the flag.**++ Security: many commands read only the two-argument spelling (`--flag value`) of a flag that flag validation accepted in either form, so the `=` form passed the check and was then dropped. The worst case was a scoped access token: `gbrain auth create <name> --scopes=read` minted a full-access token. A switch given a value had the same problem in reverse: `--yes=n` or `--dry-run=true` was read as if the value were not there. Now every `=` token on a command-line-only command is handled before the command runs: kept when the command reads the `=` form on that path, split into `--flag value` when the command's help declares a value, and refused with the two-argument fix otherwise. A switch never takes a value.++ ### What changed++ | You run | Before | Now |+ |---|---|---|+ | `gbrain auth create <name> --scopes=read` | a full-access token | exit 2, no token; `--scopes read` works as before |+ | `--yes=n`, `--dry-run=true`, `--apply=false`, `-y=n` | the value was ignored and the switch was read as given or not given | exit 2: write the switch on its own, or leave it out |+ | a value flag the command's help declares, written with `=` (serve bind address, public URL, port, token lifetime and stdio idle timeout; sync interval, source and repository; import workers; autopilot install repository and supervisor; onboard and resumed doctor remediation spend caps; apply-migrations reviewed inventory and forced retry; migrate target engine and URL) | ignored | applies |+ | an `=` spelling the command reads itself (for example recall's source, the consent spending cap and plan hash) | worked | still works |+ | an `=` spelling the command does not read | ignored, and the command ran without it | exit 2 with the spelling to use |+ | `gbrain agent run "<prompt>" --detach=true` | read as prompt text; the job ran in follow mode | exit 2: write `--detach` on its own |+ | one of agent run's own flags written with `=` inside the prompt (`gbrain agent run fix the --model=opus bug`) | read as prompt text; the job ran without the flag | exit 2: write the flag before the prompt, or put `--` before the prompt to keep it as text |+ | free text that only looks like an `=` flag (`gbrain capture "-n=3 is the answer"`, anything after `--` in an agent run prompt) | passed to the command | still passed to the command |++ ### What to watch for after you upgrade++ - Scripts that pass values with `=`: a value that used to be ignored now applies, and a spelling the command does not read now stops with exit 2. In particular a serve given an all-interfaces bind address with `=` now listens on every interface, and a public URL given with `=` now sets the OAuth issuer, so clients authorized under the old localhost issuer reconnect.+ - A few `=` spellings that only some paths of a command read now exit 2 with the two-argument fix: the stale-extraction source id of `gbrain extract`, the checkout path of the database-free `gbrain sources pull`, and the values of the managed `gbrain sources` lifecycle subcommands.+ - An agent run prompt that contains one of agent run's own flags written with `=` (for example `--model=` or `--tools=`) now exits 2. Write the flag before the prompt, or put `--` before the prompt to send it as text.+ - An upgrade restart of a `gbrain jobs supervisor` started with `=` values it ignored starts it again without them.++ ### Itemized changes++ - `src/cli/flag-equals.ts` (new): `splitFlagEquals` (keep, split or refuse per token), `EQUALS_READERS` (paths whose handler reads the `=` form, keyed by command or `command sub`; `sync --status` reads as `sync status`, `forget --purge` is its own path, `repos` uses the `sources` keys through `READER_ALIASES`), `OWN_EQUALS_PARSERS`, `COMMAND_SWITCHES`, `COMMAND_VALUES`, `END_OF_OPTIONS`, `findUnknownFlag` (moved from `main.ts`) and `directEntryArgs` for the direct `bun run src/commands/auth.ts` entry. Tests: `test/cli-flag-equals.test.ts` (every registered flag of every command-line-only command), `test/serve-flag-equals.test.ts` (real CLI on a keyless PGLite brain: serve access and surface, token minting, destructive commands, schema add-type JSON output).+ - `src/cli/main.ts`: `cliOnlyFlagArgs` runs the gate once after routing and the handler-free help printers, before strict arguments, flag validation and dispatch, and also covers the hand-parsed `search` subcommands and the validation-exempt paths.+ - `src/commands/auth.ts`: the direct entry strips and applies the global flags and runs the same gate and unknown-flag check as `gbrain auth`.+ - `src/commands/schema-add-type.ts`, `src/commands/schema.ts`, `src/commands/sources-writer-movement.ts`, `src/commands/eval-longmemeval.ts`: their own parsers refuse a value on a switch; `schema add-type` and `sources writer movement` also take `--json=true` as the JSON switch.+ - `src/commands/agent.ts`: exports `RUN_VALUE_FLAGS` and `BOOLEAN_TAIL_FLAGS`, so the gate reads `agent run` where its parser reads flags (the leading flags and the trailing switches it hoists) and refuses agent run's own flags written with `=` inside the prompt, up to the first bare `--`.+ - `src/cli/strict-args.ts`: `strictSpecFor`, the one lookup of the strict spec for a subcommand, shared with the gate.+ - `src/core/upgrade-restart.ts`: the supervisor restart keeps global flags and `--job-isolation=` and drops the `=` values the older supervisor ignored. Test: `test/upgrade-restart.test.ts`.+ - `scripts/generate-flag-registry.ts`: the flag registry no longer picks up this module's flag tables. Test: `test/generate-flag-registry.test.ts`.+## [0.60.466.0] - 2026-10-26**A brain whose database is at schema v265 now upgrades instead of stopping with `column "pending" does not exist`, and on a brain with a large write journal each write claim is a quick index lookup again instead of a scan of the whole journal after every restart.**diff --cc scripts/module-size-limits.tsvindex 3fe73fd5eb,53be93b07a..70b8aaaf3f--- a/scripts/module-size-limits.tsv+++ b/scripts/module-size-limits.tsv@@@ -47,4 -47,4 +47,4 @@@ src/commands/doctor/checks/extraction-ssrc/core/search/evidence-delivery.ts 1733 ratchet +5 W5: EvidencePlan.rankCut resolved with the plan (logic in search/page-rank-cut.ts); +20 W5 whole_pages: wholeCost prices a block as delivery would emit it whole (needs the private Block/enrichBlock internals) and the cut is wired before allocate (cutWholePages lives in search/page-rank-cut.ts); new row: crossed the 1500 unlisted cap at the #6367 merge (master 1479, its explicit-budget cap) by the C1 date header threaded through every block path, cap included; pagePlanHits (search/page-plan-hits.ts) and the header loader (evidence-date.ts blockDateHeaders) were peeled first; peel candidate: allocateCapped; +8 at the #6396 merge: page evidence text cut through the fence-chunk trust overlay and the caller minTrust floor (re-summed: our 1504 + master +8); +119 Cat 40 next wave F2: the search.long_page_unit policy (withLongPagePolicy, selectLongPages size preflight with page/byte/deadline limits) and the long-page fetch and fallback in deliverEvidence, kept in the shared seam the plan names (gbrain-evals docs/plans/2026-10-11-cat40-next-wave, F2); +17 F2 rework to the frozen-comparison policy: search.long_page_min_chars (rendered-body characters, default 20,000) read in withLongPagePolicy, a character count in the preflight, the rendered-body confirmation in deliverEvidence, and the long-page row exemption in capDeliveredSnippets (re-summed at the 0.60.225.0 merge: master 1537 + F2 136); +39 Wave B PR 2 lane 2b: planSegmentedBlock joins the windows search/segmented-evidence.ts reads around a segmented page's hits into one run-marked document (needs the private Doc/Piece/buildDoc/locateAnchor internals), planBlock's unit cut split into blockFromDoc for both, sections bounded by their run, selectionCost counts a gap between runs, and the fetch skips the whole-body fence overlay for a segmented page; +4 segmented pages read at once before planning (segmentedReads; the read is search/segmented-evidence.ts, the statements page-body/read-range.ts); +17 GBRA-156: tail trim plan flag, DeliverOptions.query, DeliveryMeta.tail_trim, the trim call and the frozen-hit score/query library fields (logic in search/evidence-tail-trim.ts; interface fields cannot move)src/commands/config.ts 1523 ratchet new row: crossed the 1500 unlisted cap in wave 15 P7.4 #5461: set-time validation of think.pages_budget_chars / think.excerpt_ceiling_chars (range + parser live in core/think/pages-budget.ts), mirroring the cycle.consolidate.cluster_threshold hook; +1 wave 16 G5.1 (#6495): set-time refusal hook for search.keyword_statement_timeout_ms (check lives in commands/config/keyword-timeout-value.ts); +1 wave 18b lane 5 (#5012 item 6): search.contextual_retrieval prints the existing-pages hint after the write (text lives in commands/config/contextual-mode-hint.ts); +1 #6712 PR 2: retired/rollback sync key dispatch line (logic in core/sync-config-keys.ts handleSyncKeyConfig); +1 #6865: sync.slug_conflicts.auto_remove boolean validation at set time (refusal text in repair/slug-autoheal.ts); +1 #6948: sync.origin_twins.auto_resolve boolean validation at set time (refusal text in repair/origin-twins-heal.ts); +2 #6795: config set chat_model/expansion_model help line and the model-pin import (shadow report and file-pin unset live in commands/config/model-pin-shadow.ts); +1 #6712 PR 4b: sync.admission set-time hook (value logic in commands/config/sync-admission-value.ts); +4 #7130: set-time refusal hook for sync.max_deletes / sync.max_delete_ratio (and .source.<id>) / purge.retention_hours (check lives in core/sync-reconcile.ts deleteSafetyConfigRefusal); +1 GBRA-150: sync.metafile_pages set-time route (body in config/sync-metafile-pages-value.ts); re-summed at the v0.60.330.0 merge (1520 measured); +2 #7242 minions.terminal_hook file-plane key and its set branch (logic in minions/terminal-hook.ts); +1 GBRA-161: generations.* set-time route (refusal in commands/config/generations-values.ts)src/commands/mcp-expose.ts 1512 ratchet hosted_option: Session carries the gbrain.io coaching notice on the first-receipt run and Tailscale/service friction exits (copy, refs and gating live in core/onboard/hosted-option.ts)- src/core/behavior-change-notice.ts 1508 ratchet the BEHAVIOR_CHANGES table grows by one row per disclosed change: the GBRA-95 RLS backfill row (+2), +3 rows for #7596 pass_interrupted, #7599 read-only serve and the --access=read-only ceiling, +1 G5 lane C trust_release row, and +2 the #7632 function search_path disclosure row; the behavior-change table split (#7607) retires this row-src/core/behavior-change-notice.ts 1507 ratchet the BEHAVIOR_CHANGES table grows by one row per disclosed change: the GBRA-95 RLS backfill row (+2), +3 rows for #7596 pass_interrupted, #7599 read-only serve and the --access=read-only ceiling, +1 G5 lane C trust_release row and +1 #7619 cli_flag_equals; the behavior-change table split (#7607) retires this row++src/core/behavior-change-notice.ts 1509 ratchet the BEHAVIOR_CHANGES table grows by one row per disclosed change: the GBRA-95 RLS backfill row (+2), +3 rows for #7596 pass_interrupted, #7599 read-only serve and the --access=read-only ceiling, +1 G5 lane C trust_release row, +1 #7619 cli_flag_equals and +2 the #7632 function search_path disclosure row; the behavior-change table split (#7607) retires this rowdiff --cc src/core/behavior-change-notice.tsindex 3d03602d09,fee1230e48..b18f787a6c--- a/src/core/behavior-change-notice.ts+++ b/src/core/behavior-change-notice.ts@@@ -1308,8 -1308,7 +1308,9 @@@ export const BEHAVIOR_CHANGES: Readonly{ since: '0.60.464.0', text: 'Security: `gbrain serve --access=read-only` (the `=` form) now sets the read-only ceiling. It passed flag validation before but started the serve with full write access, because only `--access read-only` was read. A config that used the `=` form now gets the read-only tool set and refused mutations it asked for; a repeated `--access` never widens a read-only value, and an invalid value in either form refuses. `--surface=<v>` likewise now sets the serve surface (it was ignored, leaving the full surface), and a repeated `--surface` takes the narrowest value.' },{ since: '0.60.464.0', text: 'A managed sync pass whose process was killed or crashed now reads `resumable` / `pass_interrupted` in `gbrain sync status` (with an `interrupted` block naming the pass and the dead holder) instead of `pass_unfinished`, and a second killed pass reads `blocked` / `pass_abandoned` at once. `gbrain doctor` `managed_sync_blocked` lists both (warn, fail after an hour). A live pass still committing is no longer counted abandoned because its row is old.' },{ since: '0.60.465.0', text: 'On a managed brain, `gbrain trust release h<id>` of a held fact whose entity page exists now writes the fact into that page\'s `## Facts` fence through the page-write door (a `trust_release` page write) and flips the hold in the same publication, so the released fact lives in the page\'s file and a re-import keeps it. Before, the release inserted a database-only row the file never held; a fence-guard hold that names an existing fence row created a database-only twin of it, and now indexes that row and leaves the file unchanged. The row is stamped `user_confirmed` with origin `trust_release` and the page keeps its trust tier. A hold decided after the release was accepted fails `preview_changed` and writes nothing; a remote request, or a deleted or private entity page, refuses and the fact stays held. A hold with no entity page, and unmanaged brains, release database-only as before.' },+ { since: '0.60.467.0', id: 'cli_flag_equals', text: 'A CLI flag value written with `=` that a command used to ignore now exits 2 (`invalid_params`) and names the two-argument spelling the command reads: `gbrain auth create --scopes=read` minted a full-access token. A switch never takes a value (`--yes=n`, `--dry-run=true`, `--apply=false` and `-y=n` exit 2 instead of being ignored). The `=` spellings a command reads on every path keep working, and so do the spending cap and plan hash the consent check of a command reads; a few that only some paths read now exit 2 with the two-argument fix, among them the source id given that way to `gbrain extract` for stale extraction, the checkout path given that way to the database-free `gbrain sources pull`, and the values given that way to the managed `gbrain sources` lifecycle subcommands (add, remove, archive, restore, purge, set-path, reclone). A value written with `=` on a flag that the help of the command declares used to be ignored and now applies, so check scripts that pass these that way: an all-interfaces bind address given to `gbrain serve` bound loopback and now listens on every interface; a public URL given to `gbrain serve` now sets the OAuth issuer and resource, so clients authorized under the old localhost issuer must reconnect, and admin cookies become Secure on https; the serve port, token lifetime and stdio idle timeout, the sync interval, source and repository, the import worker count, the autopilot install repository and supervisor, and the target engine and URL the legacy `gbrain migrate` copier reads take effect the same way; a spend cap given to `gbrain onboard` (off included) now applies, and one given to a resumed `gbrain doctor` remediation overrides the cap its checkpoint recorded; and `gbrain apply-migrations` now honors an accepted reviewed inventory and a forced retry of a wedged migration. An upgrade restart of a `gbrain jobs supervisor` started with values written that way (which it ignored) starts it again without them. A script that passes a value with `=` to a command that does not read that form now stops with the fix instead of running without the value.' },+ // #7632 function search_path. `since` is stamped at the merge slot.- { since: '0.60.467.0', id: 'function_search_path_pg_temp', text: 'Every gbrain function that pins a search_path now ends it with `pg_temp` (`ALTER FUNCTION ... SET search_path`, no body changes). Before, the purge and withdrawal guards pinned `pg_catalog, public`, which leaves the session\'s temporary schema first for tables and views: any session, the owner included, could create an empty temp table or view named like a ledger (`fact_purges`, `page_purges`, `take_purges`, `fact_withdrawals`, `page_segment_derivation_purges`) and store a purged claim, page, take or derivation again, or save a withdrawn fact active. The guards are otherwise unchanged (SECURITY INVOKER, same bodies, same timing). Only gbrain\'s own functions change; another application\'s functions in the same database keep their search_path. Still open, not fixed here: on Postgres, a write role with no RLS policy on a ledger reads it empty, so its purged or withdrawn writes still go through (gbrain#7659). Where the migrating role does not own a function, the migration leaves it, names it on stderr and carries on, and `gbrain doctor` (`function_search_path`) prints the `ALTER FUNCTION` to run as its owner; that changes no owner, so later boots still succeed. An older gbrain binary that replays its schema against the brain (`gbrain init` or `gbrain upgrade` from it; an ordinary boot replays nothing) drops `pg_temp` from the functions it re-creates; doctor flags them, and `gbrain init` or `gbrain upgrade` with this release, or the printed `ALTER FUNCTION`, puts it back. An ordinary boot of this release does not.' },++ { since: 'TBD', id: 'function_search_path_pg_temp', text: 'Every gbrain function that pins a search_path now ends it with `pg_temp` (`ALTER FUNCTION ... SET search_path`, no body changes). Before, the purge and withdrawal guards pinned `pg_catalog, public`, which leaves the session\'s temporary schema first for tables and views: any session, the owner included, could create an empty temp table or view named like a ledger (`fact_purges`, `page_purges`, `take_purges`, `fact_withdrawals`, `page_segment_derivation_purges`) and store a purged claim, page, take or derivation again, or save a withdrawn fact active. The guards are otherwise unchanged (SECURITY INVOKER, same bodies, same timing). Only gbrain\'s own functions change; another application\'s functions in the same database keep their search_path. Still open, not fixed here: on Postgres, a write role with no RLS policy on a ledger reads it empty, so its purged or withdrawn writes still go through (gbrain#7659). Where the migrating role does not own a function, the migration leaves it, names it on stderr and carries on, and `gbrain doctor` (`function_search_path`) prints the `ALTER FUNCTION` to run as its owner; that changes no owner, so later boots still succeed. An older gbrain binary that replays its schema against the brain (`gbrain init` or `gbrain upgrade` from it; an ordinary boot replays nothing) drops `pg_temp` from the functions it re-creates; doctor flags them, and `gbrain init` or `gbrain upgrade` with this release, or the printed `ALTER FUNCTION`, puts it back. An ordinary boot of this release does not.' },];/** The newest disclosed change's release: the notice id moves only when a release adds rows. */