pm-dispatch: a SessionStart hook prints the lane's seat-post stub and the SKILL.md pointer, because a multi-repo project's working directory does not surface the repo's .claude/skills (maintainer's word 2026-10-23) #25380

Activity

1. objectstack-fleet

   added

   tooling

   pm:queue

   domain:skills

   priority:p3

   on Oct 23, 2026
2. objectstack-fleet

   assigned

   os-elon-musk

   on Oct 23, 2026
3. objectstack-fleet

   added

   pm:dispatched

   and removed

   pm:queue

   on Oct 23, 2026
4. objectstack-fleet commented on Oct 23, 2026

   

   objectstack-fleetBot

   on Oct 23, 2026

   ContributorAuthor

   More actions

   Claim: PM loop round 1 (card thread) · 2026-10-23T04:56Z · p3, tooling, domain:skills. Adds a SessionStart hook that prints the pm-dispatch pointer (and the lane's seat-post stub when OS_FLEET_LANE is set) so a multi-repo project session finds the skill, on the maintainer's 2026-10-23 word on the card.
   Session: session_01SSWEKCxzY8HYr8EsFpR8VG (card thread of domain:skills seat 2, seat session session_01REMuzhz6MWRjg8uohedMnM)
   Account: os-elon-musk (the seat's linked user; always the card's assignee)
   Branch: claude/project-thread-06hewe (the card thread's designated push branch, platform-assigned at thread creation; the claude/issue-<n>-<slug> convention cannot be used by a project thread, a measured platform exception)
   Worktree: the card thread's own checkout
   Domain: domain:skills
   Seat: domain:skills#2 (seat post #19287)
   File surface: .claude/settings.json (one SessionStart registration beside the existing guards); new .claude/hooks/session-start-fleet.sh and .claude/hooks/session-start-fleet.selftest.sh; one line in .claude/skills/pm-dispatch/references/seat-lifecycle.md (away from L69 and 〈座位贴协议〉); scripts/pm/check-skill-line-ratchet.mjs only if that line needs a ceiling row at its landed count. (stop on breach; explain in the report)
   ⛔ Not in scope: any Tier H file; SKILL.md and every other reference; uploading the skill as an account skill; any network or GitHub read in the hook.
   Container & model: M, mode:thread (the maintainer's word: one project thread per M-or-larger or p1 card), model: CONTRACT_REVIEW_TIER (reason: no path mandate from dispatch-gates.mjs --tier at d12e650d for these paths; the maintainer ruled in this project 「改技能的卡用 fable」 and this hook exists only to load the skill), effort: high (M). Governed surface .claude/** is Tier S: a CONTRACT_REVIEW_TIER PASS record on the PR's head plus all-green checks; landing waits for the maintainer.
   Clause-②: no
   Metadata-protocol: none
   Responsibility: n/a — not a defect card (a maintainer word not yet written into the repo)
   Thread-read: none
   Serial constraints cleared: seat-lifecycle.md is also edited by PR #25348 (L69) and will be by #25379 (thread claude/project-thread-hxtlri, 〈座位贴协议〉); one line, placed apart, merge main when told. .claude/settings.json and .claude/hooks/**: no open PR found at claim time.

   ---

   Generated by Claude Code
5. objectstack-fleet commented on Oct 23, 2026

   

   objectstack-fleetBot

   on Oct 23, 2026

   ContributorAuthor

   More actions

   os-dev-report

   {
     "issue": 25380,
     "status": "done",
     "branch": "claude/project-thread-06hewe",
     "pr": "https://github.com/objectstack-ai/objectstack/pull/25386",
     "session": "session_01SSWEKCxzY8HYr8EsFpR8VG",
     "premise_still_valid": true,
     "summary": "New SessionStart hook .claude/hooks/session-start-fleet.sh prints the pm-dispatch pointer, the SKILL.md path and its latest touch on HEAD (a shallow clone's graft boundary is labelled as one, never presented as a measured touch), an optional note when CLAUDE_PROJECT_DIR is not this checkout, and the lane line (OS_FLEET_LANE => seat-post number read from a `Seat post: #N` line in references/lanes/LANE.md plus the stub reminder; absent => `lane: not set`). No network, no GitHub read, no write, exit 0 always, located from its own path. Registered in .claude/settings.json beside the PreToolUse guards; hermetic selftest with 58 cases; seat-lifecycle.md +1 line in 〈开轮互斥〉 paid by the opening reader-routing sentence the SKILL.md index already carries (96 -> 95, ceiling untouched). Two premises measured against the card: (A1) in this project thread CLAUDE_PROJECT_DIR is unset in the Bash environment and the registered PreToolUse guard did NOT block a known-blocked command, so the nested checkout's settings.json is probably not loaded in this thread shape — whether a project thread ever runs the hook is a platform question the PR cannot settle; (A5) no references/lanes/*.md carries a seat-post number, so the hook prints `not recorded` until one line per lane file is added (outside this card's surface).",
     "tests": "bash .claude/hooks/session-start-fleet.selftest.sh -> `session-start-fleet selftest: 58 passed, 0 failed`, exit 0 (hermetic fixture: git repo + SKILL.md + two lane files; cases cover pointer verbatim, touch sha = git log -1 -- SKILL.md, shallow-boundary labelling both directions, lane set/unset/empty/no file/no number/path chars/shell metachars, cwd and CLAUDE_PROJECT_DIR not inputs, <=15 lines, bash -n, executable bit, settings.json wiring, NON-VACUITY: mutated pointer caught, exit-1 copy seen, set -u abort still exits 0). Hook run on this checkout prints 4 lines, exit 0, with and without OS_FLEET_LANE. No ablation: no tested runtime behaviour changed (hook + selftest + settings + one reference line).",
     "gates": "node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack (no paths; change set off the merge base: 4 files, 390 changed lines) -> 19 commands, each run on head e75a6b53 with the exit captured before any pipe; --ran: `19 derived famil(ies) accounted for — 19 run, 0 NOT-MEASURED`, all exit 0. First sweep had two reds, both resolved before the final sweep: check:bash32-floor exit 1 (the selftest spelled the forbidden tokens into a grep pattern; removed in e75a6b53) and lint check:doc-formula-expressions exit 3 PREREQUISITE NOT MET (@objectstack/formula and @objectstack/lint unbuilt; built, exit 0). check:pm-skill-ratchet: seat-lifecycle.md 95 lines, ceiling 96, headroom 1; new line 115 bytes. CI families (pnpm lint, hook self-test discovery step, type-check lanes): read from the PR's check-runs, not re-run locally.",
     "dogfood_battery": "n/a — no `!` in the PR title and no (narrowing) arm",
     "mcp_calls": "7 — all reads: issue_read get + get_comments x2 (#25380), pull_request_read get_files (#25348), pull_request_read get (#25348); no MCP write tool",
     "api_writes": "3 — git push x2 (branch claude/project-thread-06hewe); fleet-write relay pr_create (POST /repos/objectstack-ai/objectstack/pulls + issues/25386/assignees + issues/25386/labels, run 39905209657); plus this report comment and one verification comment via post-stamped",
     "open_questions": [
       { "question": "A1: the nested checkout's .claude/settings.json does not appear to be loaded in a project thread rooted at the parent directory (PreToolUse guard did not fire; CLAUDE_PROJECT_DIR unset). Should the pointer also be carried by a channel the project thread demonstrably loads (each repo's CLAUDE.md is loaded; project instructions), or is the platform question measured first?", "options": ["A keep the hook as the mechanism and measure one post-merge project thread for the pointer lines before deciding anything else", "B also add one pointer line to objectstack/CLAUDE.md (Tier H, separate card)"], "recommendation": "A, because the card's wanted mechanism is the hook, the measurement is one thread away, and B is a Tier H edit the maintainer must rule on" }
     ],
     "out_of_scope_findings": [
       "carrier: skills seat · references/lanes/*.md carry no `Seat post: #N` line, so the hook prints `not recorded` for every lane until one line per lane file is written (one reference edit, outside this card's surface) · noted, not filed",
       "carrier: skills seat · platform reading for platform-readings.md: in a project thread rooted at the parent of the checkouts, CLAUDE_PROJECT_DIR is unset in the tool environment and the objectstack PreToolUse guards do not fire (measured 2026-10-23 on this card thread with guard-tree-enum's first block case) · noted, not filed"
     ]
   }

   ---

   Generated by Claude Code
6. objectstack-fleet

   closed this as completedin #25386on Oct 23, 2026
7. objectstack-fleet

   removed

   pm:dispatched

   on Oct 23, 2026