chore: no-mistakes(document): document FM_UPSTREAM_* env knobs in configuration reference - #803
Closed
Amplify-Logic wants to merge 13 commits into
Closed
Conversation
…efuse ambiguous delimiter shapes (#1) * fix(herdr-lab): place --session before the child-argv delimiter `agent start ... -- <child argv...>` always got a trailing --session appended after --, so Herdr never consumed it and the child process received it instead - meaning Herdr could fall through to the live default session for agent-start calls. fm_herdr_lab_raw now places exactly one --session immediately before -- for that shape and still appends it for every ordinary call. fm_herdr_lab_cli rejects multiple delimiters, a delimiter on any command other than agent start, and agent start without a delimiter plus non-empty child command, before any Herdr call is made. Updated the brief scaffold and backend doc wording that claimed the selector is always trailing. * no-mistakes(document): note run's delimiter-shape refusals in herdr-backend doc * no-mistakes(review): refuse option before child-argv delimiter; collapse raw invocation * no-mistakes(review): only treat first -- as delimiter; child argv untouched
…uid#685 vocabulary (#2) Commit 1182883 (kunchenguid#685) intentionally changed the secondmate charter scaffold's role declaration from 'a persistent domain supervisor' to 'a persistent second mate' and updated the captain-translation-contract test, but missed the assertion in tests/fm-brief.test.sh. The stale expectation left the secondmate-charter case failing. Point the role-declaration assertion at the current wording so it verifies the charter still declares its role without weakening the check or reverting the intended vocabulary.
…er to doc indexes (#3) * feat(primary): add guarded runtime profiles and Kimi K3 support * no-mistakes(review): make Kimi registry merge atomic and clean temp files * no-mistakes(document): add Kimi protocol and fm-primary launcher to doc indexes
* feat: add human-readable Herdr worker presentation * no-mistakes(review): capability-gate herdr presentation and skip secondmate worker rename
…4.5) (#6) * feat(harness): add verified cursor worker adapter (Cursor CLI + Grok 4.5) Adds `cursor` (Cursor CLI `agent` 2026.07.16-899851b) as a verified WORKER adapter, after empirical verification on 2026-07-19. Worker only: primary support was not verified and is not claimed, the mirror of the Kimi primary-only boundary. Kimi stays primary-only; no Kimi worker support here. Verification found two defects in shared composer monitoring that would have made an adapter monitoring cannot trust, both fixed in the shared owners: - Reverse-video cursor cell. cursor draws the terminal cursor as SGR 7 over the idle placeholder's FIRST character. Reverse video is neither dim/faint nor a dark foreground, so fm_composer_strip_ghost kept it and an idle composer reduced to a lone "A" -> `pending`, deferring every away-mode escalation forever. A plain FM_COMPOSER_IDLE_RE does not fix this, because the idle regex was only matched against ghost-stripped content; it is now matched against the plain row too, which is styling-independent. - #{cursor_y} does not point at cursor's composer (it parks the cursor in the bottom status area, observed 5 rows off). Reading that row found an empty status line, so a composer holding real unsubmitted text classified as `empty` - a FALSE-EMPTY, the dangerous direction, since the away-mode injector picks targets by emptiness and would type over pending input. The tmux path now locates the composer structurally, matching herdr's existing scan. Other integration surfaces: - fm-spawn: launch template (--yolo autonomy, --workspace pinned to the task worktree, and deliberately no -w so cursor never allocates a second worktree), plus a per-task gitignored .cursor/hooks.json `stop` hook verified to fire once per turn with no extra hook-trust grant. - Effort axis: cursor has no effort flag - effort is a SUFFIX on the model id - so cursor_model_with_effort folds it in. xhigh/max cap at high, an explicit tiered model is never retiered, and -fast variants are never implicit. - fm-harness: CURSOR_AGENT=1 detection, tested BEFORE CLAUDECODE because cursor does not clear an inherited CLAUDECODE=1 and a cursor worker under a claude-hosted firstmate would otherwise be steered with claude's vocabulary. - Busy signature `ctrl+c to stop` in both owners. The spinner VERB is deliberately not matched: it flips Working -> Running mid-turn, so matching it reads a tool-executing pane as idle. - tmux liveness: cursor's wrapper execs node, but the versioned cursor-agent bundle path survives exec -a in argv, so a `node` COMM resolves to `alive`. Any other bare node stays `unknown`, never dead; pi's gap is unchanged. Backends reviewed rather than assumed: tmux verified; herdr composer-safe by construction (structural scan + shared idle default) with liveness unverified; zellij/orca/cmux not exercised with cursor. Recorded in docs/cursor-harness.md. Also documents that cursor executes claude-format hooks (.claude/settings*.json, Stop -> stop), so cursor must never be launched from the firstmate primary checkout, and that --model mutates the account-global default. A live fm-spawn dispatch was NOT run: it allocates a real pooled worktree and writes live fleet state outside the task worktree. Every underlying behaviour was verified by raw launches and the integrated path (launch, turn-end hook, composer, liveness) was confirmed together on a real pane; the first supervised fm-spawn --harness cursor run remains firstmate's gate. Evidence: docs/cursor-harness.md. Regressions: tests/fm-cursor-adapter.test.sh (20 assertions). bin/fm-lint.sh clean; composer/tmux/herdr/orca/cmux suites pass. * no-mistakes(review): scope cursor composer-row scan to identified cursor panes * no-mistakes(document): document cursor worker adapter in config, architecture, and indexes
* fix(spawn): refuse missing harness binaries before launch Resolve and probe verified launch executables before creating task endpoints, preventing absent CLIs from leaving dead worker shells. Co-authored-by: Cursor <cursoragent@cursor.com> * no-mistakes(review): add stdin-closed timeout-bounded version probe with override knob * no-mistakes(review): escalate timed-out version probe to SIGKILL after grace * no-mistakes(review): kill probe process group so timed-out wrapper descendants die --------- Co-authored-by: Cursor <cursoragent@cursor.com>
Define one fleet-status contract for guarded Claude, Pi, and Kimi surfaces while preserving Cursor's worker-only boundary.
* test(watcher): wait for healthy peer readiness Wait for the fixture's SIGTERM handler before exercising restart so startup scheduling cannot turn the healthy-peer case into the child-owned path. Co-authored-by: Cursor <cursoragent@cursor.com> * no-mistakes(review): clean up fixture processes on healthy-peer assertion failures --------- Co-authored-by: Cursor <cursoragent@cursor.com>
Keep startup dispatch validation and secondmate liveness aligned with the verified worker adapter set, including Cursor's model-suffixed effort range. Co-authored-by: Cursor <cursoragent@cursor.com>
…alation (#11) * fix(herdr): do not escalate busy cursor panes as waiting-on-human Herdr can report agent_status=blocked while a cursor worker is mid-turn (ctrl+c to stop footer). Corroborate blocked/idle against that busy signature before immediate stale escalation or poll-path idle. Co-authored-by: Cursor <cursoragent@cursor.com> * no-mistakes(test): fix jq-1.6 and preflight fixture gaps in failing tests * no-mistakes(docs): corroborate herdr idle/blocked in architecture summary Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * no-mistakes(lint): silence SC2329 on indirectly-invoked test stub --------- Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Detect when a configured upstream remote has commits this home lacks, and report count plus subjects through the existing bootstrap diagnostic path so forks stop silently drifting behind without auto-merging third-party changes. Co-authored-by: Cursor <cursoragent@cursor.com>
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
Author
|
Closing this PR — it targeted the wrong repository by mistake (fork PR routing opened against the upstream parent) and is not an intended contribution. Sorry for the noise; the change will land on Amplify-Logic/firstmate instead. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Intent
Re-open the PR against Amplify-Logic/firstmate:main for the already-validated upstream-drift detection work on HEAD c91630e.
Same commits already cleared review/test/document/lint on run 01KY1SHNE9CH54BRA3J29JPSJW; skipping those steps only because the pipeline agent (claude) is blocked by the org monthly spend limit, and the sole remaining need is a correctly-targeted Amplify-Logic PR with the no-mistakes body signature plus CI green.
CRITICAL: PR target must be Amplify-Logic/firstmate:main, never kunchenguid/firstmate. Wrong upstream PR #801 was closed; hand-opened Amplify-Logic #14 was closed for missing signature.
Feature intent unchanged: detect and surface upstream Firstmate drift at session start (detection only, no auto-merge, firstmate-only scope).
What Changed
c91630e no-mistakes(document): document FM_UPSTREAM_* env knobs in configuration reference
0c3bb92 feat(bootstrap): surface fork upstream drift at session start
Testing
Pipeline
Updates from git push no-mistakes
✅ **intent** - passed
✅ No issues found.
✅ **Rebase** - passed
✅ No issues found.
⏭️ **Review** - skipped
Step was skipped.
⏭️ **Test** - skipped
Step was skipped.
⏭️ **Document** - skipped
Step was skipped.
⏭️ **Lint** - skipped
Step was skipped.
✅ **Push** - passed
✅ No issues found.