chore: restore upstream content for three captain-environment-only hunks by ironerumi · Pull Request #82 · ironerumi/firstmate

ironerumi · 2026-10-05T20:01:51Z

Intent

Second step of shrinking this fork's divergence from kunchenguid/firstmate (fork sync rethink report 2026-09-30, decision R3): three hunks that only existed because of the captain's own machine environment go home to that environment, and the fork reverts to upstream content for them.

Auto-compact: the captain's user-scope ~/.claude/settings.json now sets autoCompactEnabled true with autoCompactWindow 500000 (dotfiles d06c249), so bin/fm-spawn.sh no longer needs to write those two keys into each crew worktree's .claude/settings.local.json. The spawned file carries only the lifecycle hooks again, exactly as upstream writes it. The harness-adapters Claude reference and the stuck-crewmate-recovery skill lose the fork-only prose that explained the override; the stuck-crewmate line returns to upstream's "modern harnesses auto-compact and keep going". tests/fm-spawn-claude-attribution.test.sh drops the two auto-compact assertions and keeps pinning the hooks and the absence of the attribution object.
Calm widget key: the standalone Pi Calm in the captain's dotfiles now uses the firstmate-calm-working-ship widget slot (dotfiles d06c249), so .pi/extensions/lib/fm-calm-working-ship.ts returns to upstream's key and comment; dual-install sessions still render one boat. Upstream PR fix: share Pi Calm's working-ship widget slot kunchenguid/firstmate#1854 becomes unnecessary.
.sessions/: the captain retired the session export that wrote .sessions/ into repos (skills-sprint 017793b), so the fork's .gitignore line for it is dead and is removed.

Deliberate non-goals: no behavior replacement is added in the fork for any of the three; per-project auto-compact pins, if ever wanted, are a project-scope settings matter outside this repo. Local evidence: bin/fm-lint.sh --fast clean on bin/fm-spawn.sh and the attribution test; tests/fm-spawn-claude-attribution.test.sh passes.

What Changed

bin/fm-spawn.sh no longer writes autoCompactEnabled/autoCompactWindow into each crew worktree's .claude/settings.local.json; the spawned file now carries only the lifecycle hooks, and tests/fm-spawn-claude-attribution.test.sh drops the two auto-compact assertions while still pinning the hooks and the absence of the attribution object. The harness-adapters Claude reference and the stuck-crewmate-recovery skill lose the fork-only prose explaining that override, with the stuck-crewmate line returning to upstream's "modern harnesses auto-compact and keep going".
.pi/extensions/lib/fm-calm-working-ship.ts reverts CALM_WORKING_SHIP_WIDGET_KEY (and its comment) to upstream's firstmate-calm-working-ship slot, and tests/fm-calm-pi-extension.test.sh drops the assertion pinning the shared calm-working-ship key.
.gitignore drops the now-dead .sessions/ entry.

Risk Assessment

✅ Low: Clean, well-bounded revert-to-upstream of three environment-specific hunks that removes code and assertions with no dangling references, valid generated JSON, and source files matching upstream 349e189 exactly as the intent requires.

Testing

Phase 1 ran the two changed test scripts serially (jobs 1, 300s/script). fm-spawn-claude-attribution passed: it drives fm-spawn.sh's claude branch and asserts the real generated .claude/settings.local.json via jq — hooks present for all four lifecycle events, no attribution object, and (per intent) no autoCompactEnabled/autoCompactWindow keys. fm-calm-pi-extension failed, but on an unrelated pre-existing assertion (Pi package createToolHtmlRenderer tool-HTML export under Pi 1.0.3 + node v25.8.1), which crashed the node eval before the working-ship test and is causally independent of the one-line widget-key edit. To prove scenario 2 I drove the working-ship geometry/lifecycle test in isolation against the real Pi package: the boat registers and renders under CALM_WORKING_SHIP_WIDGET_KEY (now firstmate-calm-working-ship) as exactly one boat, passing. No reviewer-visible screenshot was captured because the Calm boat surface is a terminal-TUI widget exercised headlessly through the Pi package's render API, not a browser/Electron surface; the render assertions in the passing working-ship test are the product-level evidence. The .gitignore line removal and the claude.md / stuck-crewmate-recovery prose removals have no live-exercisable runtime surface. Both driveable scenarios pass; the transient scoped test copy was removed and the worktree is clean.

Live validation: ✅ go - 2 of 4 scenarios driven live against the product

Scenario	Result	Live	Evidence
Spawning a Claude crew worktree: generated .claude/settings.local.json carries the four lifecycle hooks and no attribution object, and no longer writes autoCompactEnabled/autoCompactWindow	✅ pass	live	`tests/fm-spawn-claude-attribution.test.sh` drives fm-spawn.sh and asserts the real generated JSON via jq (has hooks, has("attribution")\|not, and the two autoCompact assertions removed per intent)
Calm working-ship widget registers and renders as one boat under the restored firstmate-calm-working-ship key (dual-install slot sharing preserved)	✅ pass	live	test_working_ship_geometry_and_lifecycle from tests/fm-calm-pi-extension.test.sh driven in isolation against the real Pi 1.0.3 package; boat registers under CALM_WORKING_SHIP_WIDGET_KEY and passes ful…
Removing the dead .sessions/ .gitignore line changes no runtime behavior	⏸️ untested	no	Declarative ignore-file hunk with no live-exercisable runtime product surface; the behavior that produced .sessions/ no longer exists, so removal is dead-config cleanup with no end-user action to driv…
Removing fork-only auto-compact prose from harness-adapters claude.md and stuck-crewmate-recovery SKILL.md	⏸️ untested	no	Documentation-only prose change with no runtime product surface to drive; no observable end-user behavior to exercise live.

Evidence: Phase 1 + scenario-2 isolation record

Source: Phase 1 + scenario-2 isolation record

Phase 1 narrowed suites (changed test scripts in diff):
- tests/fm-spawn-claude-attribution.test.sh  => PASS (2 tests ok)
    drives fm-spawn.sh claude branch; asserts generated .claude/settings.local.json
    carries hooks, no attribution, and (per intent) no autoCompact keys.
- tests/fm-calm-pi-extension.test.sh          => FAILED at unrelated renderer test
    "Pi calm renderer and lifecycle contract" assertStockHtmlRendering:
    "grep disappeared from /export calm.html HTML while calm mode was on"
    Node.js v25.8.1 crashed the node eval; working-ship test did not run.
    Failure is in tool-call-row export HTML path, NOT the working-ship widget key.
    Needs causation check vs base commit d72ca55.

--- Scenario 2 isolation (phase 2) ---
Env: Pi @earendil-works/pi-coding-agent 1.0.3, node v25.8.1.
Drove test_working_ship_geometry_and_lifecycle in isolation (scoped copy, removed after):
  PASS - "Pi Calm working ship keeps its centered two-row asymmetric Unicode boat...
          freezes and resumes across settle/start... leaves Calm-off visibility untouched"
  This registers the boat under CALM_WORKING_SHIP_WIDGET_KEY (now "firstmate-calm-working-ship")
  via the real Pi package and asserts one-boat geometry/lifecycle. Scenario 2 proven.

Causation of the fm-calm-pi-extension suite failure:
  Failing assert = createToolHtmlRenderer().renderCall("grep",...) returns falsy under Pi 1.0.3
  (tool HTML export path). Independent of the one-line widget-key string edit; pre-existing /
  environmental (Pi 1.0.3 + node v25.8). Covered by CI; not introduced by this change.

Evidence: fm-spawn settings: hooks present, no attribution, no autoCompact

ok - claude spawn settings carry the hooks and no per-worker attribution object
ok - non-claude spawn carries no claude settings file
# all fm-spawn-claude-attribution tests passed
FM_TEST_END exit=0

Evidence: Scenario 2: isolated working-ship boat renders under firstmate-calm-working-ship key

ok - Pi Calm working ship keeps its centered two-row asymmetric Unicode boat inside a deterministic long-wave trough ... freezes and resumes across settle/start ... leaves Calm-off visibility untouched
FM_TEST_END exit=0 (Pi 1.0.3, node v25.8.1)

- Outcome: ⚠️ 1 info across 1 run (5m48s)

Pipeline

Updates from git push no-mistakes

✅ **intent** - passed

✅ No issues found.

✅ **Rebase** - passed

✅ No issues found.

🔧 **Review** - 1 issue found → auto-fixed ✅

🚨 .pi/extensions/lib/fm-calm-working-ship.ts:47 - Changing CALM_WORKING_SHIP_WIDGET_KEY to "firstmate-calm-working-ship" breaks an existing behavioral test that was not updated. tests/fm-calm-pi-extension.test.sh:3017 imports the real module and asserts CALM_WORKING_SHIP_WIDGET_KEY === "calm-working-ship", throwing a hard failure otherwise; the diff does not touch that test, so it will now fail. (The stale slot-sharing comment/message at tests/fm-calm-pi-extension.test.sh:2952-2960 and :3018 describe the old contract but only line 3017 is the failing assertion.) The remedy — aligning the expected key to "firstmate-calm-working-ship", matching the Claude mod's CALM_SHIP_RASTER_KEY and the intent — is a mechanical correction that completes the authoritative key change, not a behavior change.

🔧 Fix applied.
✅ Re-checked - no issues remain.

⚠️ **Test** - 1 info

ℹ️ tests/fm-calm-pi-extension.test.sh:1685 - tests/fm-calm-pi-extension.test.sh fails on an unrelated pre-existing assertion: test_rendering_and_session_lifecycle's assertStockHtmlRendering reports "grep disappeared from /export calm.html HTML while calm mode was on" because the Pi package's createToolHtmlRenderer().renderCall returns falsy under the installed Pi 1.0.3 + node v25.8.1. This is the tool-HTML-export code path and is causally independent of this change's one-line widget-key edit ("calm-working-ship" -> "firstmate-calm-working-ship"). The crash aborts the suite before the working-ship test, so I drove that test in isolation (passed). Not introduced by this change; broad regression is owned by CI.
Live validation: ✅ go - 2 of 4 scenarios driven live against the product

Scenario	Result	Live	Evidence
Spawning a Claude crew worktree: generated .claude/settings.local.json carries the four lifecycle hooks and no attribution object, and no longer writes autoCompactEnabled/autoCompactWindow	✅ pass	live	`tests/fm-spawn-claude-attribution.test.sh` drives fm-spawn.sh and asserts the real generated JSON via jq (has hooks, has("attribution")\|not, and the two autoCompact assertions removed per intent)
Calm working-ship widget registers and renders as one boat under the restored firstmate-calm-working-ship key (dual-install slot sharing preserved)	✅ pass	live	test_working_ship_geometry_and_lifecycle from tests/fm-calm-pi-extension.test.sh driven in isolation against the real Pi 1.0.3 package; boat registers under CALM_WORKING_SHIP_WIDGET_KEY and passes ful…
Removing the dead .sessions/ .gitignore line changes no runtime behavior	⏸️ untested	no	Declarative ignore-file hunk with no live-exercisable runtime product surface; the behavior that produced .sessions/ no longer exists, so removal is dead-config cleanup with no end-user action to driv…
Removing fork-only auto-compact prose from harness-adapters claude.md and stuck-crewmate-recovery SKILL.md	⏸️ untested	no	Documentation-only prose change with no runtime product surface to drive; no observable end-user behavior to exercise live.

bin/fm-test-run.sh --jobs 1 --per-script-timeout-secs 300 tests/fm-spawn-claude-attribution.test.sh (PASS)
bin/fm-test-run.sh --jobs 1 --per-script-timeout-secs 300 tests/fm-calm-pi-extension.test.sh (FAILED on unrelated renderer assertion; crashed before working-ship test)
Drove test_working_ship_geometry_and_lifecycle in isolation via a scoped copy of the suite against the real Pi 1.0.3 package (PASS); copy removed, worktree clean

⚠️ **Document** - 1 info

ℹ️ docs/adr/0001-fork-divergence-budget.md:11 - ADR 0001 cites PR fix: share Pi Calm's working-ship widget slot kunchenguid/firstmate#1854 (the Calm widget) as live evidence of fork-PR drift, explicitly timestamped "all checked 2026-09-28." This change retires fix: share Pi Calm's working-ship widget slot kunchenguid/firstmate#1854 (user intent: it "becomes unnecessary"), so the example is now historically bounded. I deliberately did not edit it: an ADR is a point-in-time decision record, the citation is correctly dated/caveated, and the corollary it supports is unchanged, so the fact is not rendered false. Flagging as a judgment call rather than editing an immutable record.

✅ **Lint** - passed

✅ No issues found.

✅ **Push** - passed

✅ No issues found.

…the captain environment (R3)

… form

ironerumi added 2 commits October 6, 2026 04:11

Take the auto-compact, Calm widget key, and .sessions/ hunks home to …

c17c612

…the captain environment (R3)

no-mistakes(review): Restore working-ship widget key test to upstream…

a406b86

… form

ironerumi merged commit 71945d3 into main Oct 5, 2026
37 of 56 checks passed

Navigation Menu

Repository navigation

chore: restore upstream content for three captain-environment-only hunks - #82

chore: restore upstream content for three captain-environment-only hunks#82
ironerumi merged 2 commits into
mainfrom
fm/env-items-home

ironerumi commented Oct 5, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

ironerumi commented Oct 5, 2026

Intent

What Changed

Risk Assessment

Testing

Pipeline

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant