knowledge: 7 insights — lock owner id inherited by a spawned session (1 new page, 6 project-specific plan-gap drops)

[choiyounggi]

Knowledge flush — 7 insight(s) claimed: 1 ingested, 6 dropped

Run id (inherited from hooks/auto-flush.sh): 20260917-224552-43083. Claimed ids:
dbedb1f0f153ea80, a0f287cdc70eac92, 4b3490c5bb95a48a, 62e2909c3f9f7f5e,
e5b99a9fc624776f, bf8a6eee857f7808, 5306e2de6142b95e.

Verified best-practice

1. dbedb1f0f153ea80 — inherited lock owner id in a spawned session → ingested, confidence: verified

Claim. When a session is spawned by a hook/parent that already holds a
run-id-keyed single-flight lock, acquire under the id exported in the
environment rather than a freshly generated one; when an acquire reports held
seconds after session start, compare the holder id with the inherited env before
concluding a foreign run is live.

Sources checked (fetched 2026-09-18):

• https://man7.org/linux/man-pages/man2/flock.2.html — "Locks created by flock() are associated with an open file description … duplicate file descriptors (created by, for example, fork(2) or dup(2)) refer to the same lock"; "If a process uses open(2) … to obtain more than one file descriptor for the same file, these file descriptors are treated independently by flock(). An attempt to lock the file using one of these file descriptors may be denied by a lock that the calling process has already placed via another file descriptor." — the OS-level form of the same failure (a re-opened path is a stranger to its own lock).
• https://docs.oracle.com/en/java/javase/21/docs/api/java.base/java/util/concurrent/locks/ReentrantLock.html — "A ReentrantLock is owned by the thread last successfully locking, but not yet unlocking it"; lock() "will return immediately if the current thread already owns the lock" — re-entrancy is defined by identity equality with the recorded owner.
• https://www.gnu.org/software/make/manual/html_node/Job-Slots.html — the parent make provides jobserver access "through the environment to its children, in the MAKEFLAGS environment variable" via --jobserver-auth=; "Only the last instance is relevant" — a documented case of a parent handing a coordination token to children through the environment.

Reproduction (this session, 2026-09-17 22:46 KST): flush-lock.sh acquire
under a freshly generated RUNID=flush-20260917-224603-43624 →
held 20260917-224552-43083 11s, exit 3. env | grep DEV_LOOP showed
DEV_LOOP_FLUSH_RUN_ID=20260917-224552-43083 exported by auto-flush.sh
(lines 77–79: acquire, export DEV_LOOP_FLUSH_RUN_ID, spawn). Re-running
acquire under the inherited id → already-owned 20260917-224552-43083, exit 0.
flush-lock.sh lines 76–86 implement the re-entrant branch by comparing the
owner-file id to $DEV_LOOP_FLUSH_RUN_ID. Matches the candidate's own evidence
from the 17:16 run (held 20260917-171642-20541 19s → already-owned).

Both the official semantics (ownership = identity match; inheritance via env /
open file description) and a two-way local reproduction (known-bad: fresh id →
held; known-good: inherited id → already-owned) confirm the directive → verified.

2–7. Plan-gap rows from dev-loop-cockle → dropped (project-specific)

id	Decision	Why it is not reusable knowledge
a0f287cdc70eac92 (t7-usage README tree line + log.md entry)	drop	Names one repo's README line and log grammar; no directive applies outside dev-loop
4b3490c5bb95a48a (t9 fail-open sentence identical in three skills)	drop	The directive is one repo's skill wording; the cited sources (near-dup thresholds, MCP token cost) do not support the wording decision
62e2909c3f9f7f5e (t9 wiki-ingest dedupe via wiki_search)	drop	A step-ordering edit to one skill in one repo; the sources cited are about cosine thresholds, unrelated to the decision
e5b99a9fc624776f (t9 neardup subparser spec)	drop	A CLI design that has not shipped — git ls-tree origin/main finds no scripts/wiki-index.py, and the cockle worktree's scripts/ has none either; the generic kernel (dot product on L2-normalised vectors = cosine) is textbook, and the threshold sources say "calibrate", which yields no page-grade directive
bf8a6eee857f7808 (t9 README subsection)	drop	Documentation layout for one repo
5306e2de6142b95e (t9 combined log.md line)	drop	One repo's log convention ("one entry per task")

Existing-layer check

Pages read: backend-common-concurrency-distributed-locks, infrastructure-agent-orchestration-shared-run-state, backend-common-jobs-scheduled-job-overlap, testing-data-test-data-and-isolation, databases-selection-vector-search-engine-selection

Also read: INDEX.md, wiki/infrastructure/index.md (agent-orchestration section, rows 15–26), AGENTS.md lines 90–116, templates/page.md.

Search evidence (whole wiki/, untruncated): grep -rniE 're-?entran|inherited (run|owner|id|env)|RUN_ID|owner id|holder id|already-owned|lock owner|lockfile|lock file|flock' → 46 lines, all in: dependency lockfiles (supply-chain, image-builds, uv), flock -n for cron overlap (scheduled-job-overlap, background-services), ReentrantLock vs synchronized (java threads-and-memory), and the LO_RUN_ID env-leak test incident (test-data-and-isolation). None covers a spawned session inheriting a lock owner id. grep -rliE 'cosine|embedding|near-?dup' → 5 files, all datastore selection — no near-dup page (moot after the drop).

• Overlap / merge candidates: distributed-locks (owner token, atomic release-if-mine) is the parent concept and is linked, not merged into — its trigger is multi-instance services, not a spawned child session; adding a re-entrancy section there would violate one-case-per-page. shared-run-state covers "a repo that may already have a run" (foreign-run detection) — the new page is the complement: the holder that is not foreign. scheduled-job-overlap documents flock -n — the new page's flock edge case links back. test-data-and-isolation holds the LO_RUN_ID env-inheritance incident — same mechanism, opposite direction (env leaking into tests vs. env intentionally carrying the owner id).
• Conflicts: none — no existing directive tells a spawned session to mint its own id.
• Created new: wiki/infrastructure/agent-orchestration/inherited-lock-ownership-in-a-spawned-session.md (69 body lines).
• Related links added both ways: distributed-locks, scheduled-job-overlap, shared-run-state, test-data-and-isolation → new page; new page → all four.
• Lint: node scripts/wiki-lint-prohibitions.js → directives: 75, compliant: 75, violations: 0 (the 1 info item is pre-existing in keys-ahead-of-their-consumer). Vague-qualifier grep on the new page → none. All 4 related: ids resolve to exactly one id: line each.

Open-PR check

gh pr list --repo choiyounggi/dev-loop --state open --search "head:knowledge/" → 17 open heads: #210 (…-171731), #209 (…-160110), #208 (…-150057), #207 (…-135945), #205 (…-100145), #191, #190, #189, #188, #187, #186, #185, #183, #182, #181, #180, #179.

Every head fetched; git diff origin/main origin/<head> -- wiki/ grepped for added lines matching re-?entran|inherited (run|owner|id|env)|RUN_ID|owner id|holder id|already-owned|foreign holder|near-?dup|cosine|embedding: 16 heads → 0 hits; #209 → 3 hits, all "embedding store" in input-manifest-freshness-with-skipped-inputs (derived-index freshness — unrelated). Page names across all heads matching lock|env|inherit|spawn|parent|child|dedup|similar|vector|hook were also inspected: #189's testing-quality-proving-a-critical-section-is-lock-protected (testing a mutex-guarded increment — unrelated) and #208's environment-config.md change (a related: link only).

Candidate	Overlapping open head	Verdict
dbedb1f0f153ea80 inherited lock owner id	none	new — ingested here
a0f287…, 4b3490…, 62e290…, e5b99a…, bf8a6e…, 5306e2…	none	drop (project-specific, see above; not pending-duplicates)

Routing decision

Insight	Domain / category / page	Why here
dbedb1f0f153ea80	infrastructure/agent-orchestration/inherited-lock-ownership-in-a-spawned-session (new page)	The trigger is a spawned agent/hook session coordinating with its parent through a single-flight lock — the agent-orchestration category already holds the sibling cases (shared-run-state, session-completion-gates, worktree-isolated-workers). backend/common/concurrency was rejected because its pages are about service instances contending for a resource, not a parent/child pair sharing one identity. No new category needed. INDEX.md infrastructure route line and wiki/infrastructure/index.md updated; log.md entry appended.
6 plan-gap rows	— (dropped)	No layer: not reusable outside dev-loop-cockle