All-dimensions bug-hunt loop state

[REPPL]

This issue is the shared state for the autonomous all-dimensions bug-hunt loop (two sibling scheduled routines: branch prefixes bughunt-a and bughunt-b).

Protocol summary:

• Each scheduled run is exactly one round: baseline (make preflight + gofmt -l .), hunt (code bugs, docs-vs-functionality, infrastructure, internal doc consistency), adversarial refutation of every candidate finding, ledger capture of confirmed findings as iss-N records under .abcd/work/issues/, fix on a bughunt-<a|b>/round-<N> branch, one PR per round, dual independent PR review posted as a PR comment, then the PR is left OPEN for a human to merge.
• Findings are captured in the committed ledger (.abcd/work/issues/), never as GitHub issues.
• Each round ends by posting a round record comment here: Round N — findings: X substantive, Y nitpick, Z refuted; nitpicks-only: yes|no; PR #M merged|open (<reason>)|none.
• Only comments authored by the repo owner or collaborators count as loop state.
• Stop conditions: a trusted comment containing LOOP-STOPPED, or 3 consecutive nitpicks-only: yes rounds (which triggers the LOOP-STOPPED comment).

Consecutive nitpick-only rounds: 0

[REPPL]

Round 1 — findings: 16 substantive, 7 nitpick, 6 refuted; nitpicks-only: no; PR #377 open (CI green, two independent adversarial reviews both SHIP; left for a human to merge — auto-merge not authorised).

Branch: bughunt-b/round-1. Baseline (make preflight + gofmt -l .) green before any change.

Fixed (10 records, resolved this round): iss-291 guard exec-string value-flag cluster bypass (script -Tc out.txt -c '<blocker>' was a silent allow that also disarmed the Tier-2 fail-safe); iss-292 wrapper-probe test wrote junk files into the tracked tree (0/1/PATH/root removed); iss-293 ParseSemver swallowed strconv.Atoi's range error (component overflow clamped to MaxInt64); iss-294 version --check "only network" claim; iss-295 README unshipped brief-onboarding claim; iss-296 ADR-index + ruleset-README count drift; iss-297 adr-40 verb-rename residue in roadmap/brief; iss-298 all 18 broken/stale brief heading anchors and section citations (anchorcheck now 0); iss-299 phantom "in-session dispatch" internals chapter; iss-300 persona glossary role-first contradiction.

Recorded, not fixed (stay open): iss-301/iss-302 — external-review.yml approval-tally (per-page --jq) and error-masking + under-permissioned collaborators endpoint; both genuine and substantive but the file is a required-check workflow that landed the same day and cannot be validated without a real Actions run, so left for a maintainer with proposed fixes attached. iss-303 anchor-lint feature gap (links_resolve never validates fragments — the systemic fix behind the anchor cluster). iss-304 deferred nitpicks (memory-ingest --source size TOCTOU; completion/help omitted from the CLI reference; the CI job id record-lint runs the reviews-charter script).

Refuted / prior art (6): the scripts/ payload over-inclusion (knowingly-deferred open dependency, iss-34), the attribution/reviews gates running from the PR tree (CODEOWNERS contains it), AGENTS.md's CI-job list (ci.yml-scoped, iss-182), the release Go-version float (deliberate), and the 05-personas.md random-picker line (prior art iss-49).

Model routing: orchestration on Claude Fable 5; hunters, per-finding refuters, and one of the two pre-merge reviewers on Claude Opus 5; the second pre-merge reviewer on Claude Fable 5 (dual-model merge gate). Round summary also appended to .abcd/work/DECISIONS.md in the PR.

---

Generated by Claude Code

[REPPL]

Round 2 — findings: 10 substantive, 3 recorded-not-fixed (iss-336 gate_lockstep blank-path disarm [nitpick], iss-337 probe.go inner-descent OpenRoot FIFO race [minor, needs an openat restructure], iss-338 persona pronoun drift [nitpick]), 2 refuted; nitpicks-only: no; PR #380 open (CI green — all 12 checks; two independent adversarial reviews both SHIP; left for a human to merge — auto-merge not authorised).

Branch: bughunt-a/round-2. Baseline (make preflight + gofmt -l .) green before any change.

Fixed (10 records, resolved this round). Code/security: iss-339 the scanner's githubRemoteRe matched the github.com host case-sensitively, so a mixed-case remote (git@GitHub.com:…) left GitRemoteUsername empty and the github_username redaction kind never armed — the caller's handle survived history-capture redaction and the PII scan (added (?i)); iss-340 abcd history show/list printed the untrusted transcript body and metadata fields with no termsafe, replaying ESC/CSI/C1/bidi raw (now SanitizeBlock/Sanitize); iss-341 readLifeboatFile kept an Lstat→root.Open window over an untrusted lifeboat, FIFO-hang/symlink-follow (routed through fsutil.ReadGuardedInRoot); iss-342 SourceContext.ListDir opened directory entries with a blocking open, so a statically-planted FIFO named docs hung disembark probe/plan/pack over an untrusted target repo with no race (live-reproduced; now O_RDONLY|nonBlock). Docs (fix): iss-343 prepare-this-repo.md wrong path + level count; iss-331 lint.md over-scoped abcd-lint:allow waiver. Record (internal): iss-332 development/README.md denied the adr-32 ledger; iss-333 the brief's 03-configuration.md pre-adr-30 record tree (redrawn); iss-334 intents/README.md retired intent review/intent-fidelity-reviewer; iss-335 the issues store contract omitted impact + agent-observation.

Refuted this round: the govulncheck non-gating claim (documented deliberate advisory — commit 61a15b4, ci.yml:362-363; making it required collides with the new-advisory-wedges-the-queue policy), and the CHANGELOG bijection gap for iss-296..300 (the bijection reads the composer's JSON payload, never CHANGELOG.md; the residual is the already-open iss-256).

Id collision (resolved): records were minted as iss-326..338, but the v0.6.1 cut on main concurrently took iss-326..330 (one of them, iss-330, documenting this exact parallel-mint hazard). Re-minted the five colliding records to iss-339..343 on top of the maintainer's Merge branch 'main' (fast-forward, no history rewrite); iss-331..338 unchanged.

Model routing: orchestration on Claude Fable 5; hunters, per-finding refuters, the lifeboat-sibling verifier, and one of the two pre-merge reviewers on Claude Opus 5; the second pre-merge reviewer on Claude Fable 5 (dual-model merge gate). Round summary also appended to .abcd/work/DECISIONS.md in the PR.

Assisted-by: Claude:claude-fable-5

---

Generated by Claude Code

[REPPL]

Round 3 — findings: 8 substantive, 5 nitpick, 0 refuted; nitpicks-only: no; PR #391 open (both independent reviews SHIP; CI had not auto-triggered within the 30-min window — appears to await maintainer approval; left for a human to merge — auto-merge not authorised).

Branch: bughunt-a/round-3. Baseline (make preflight + gofmt -l .) green before any change. IDs minted above the sibling bughunt-b/round-2's iss-357 via the cross-ref floor (fetched PR + branch refs before capture).

Fixed (11 records, resolved this round). Code/security: iss-361 the lifeboat coverage renderer (per-repo and aggregate) printed untrusted cross-repo status/tier/tiers_present/section fields to the terminal with no termsafe — only the repo name was sanitised — so a crafted probe report injected raw ANSI/OSC-8/bidi (reproduced); both renders now route every repo-derived string through sanitize, widths from the sanitised strings; iss-366 termsafe.Sanitize masked the deprecated U+FEFF joiner but not its Unicode successor U+2060 WORD JOINER, the U+2061–U+2064 invisible operators, or U+00AD SOFT HYPHEN — all now masked; iss-364 the abcd capture related_specs validator pinned the retired fn-N namespace (reFnID) instead of the live spc-N, making the field unusable — switched to reSpcID, dropped the orphaned reFnID, corrected the store contract and the capture-surface brief. Surface (additive): iss-362 abcd banlist --json omitted the reach caveat its command file tells an agent to relay verbatim — added a reach field to the private JSON, mirroring the ahoy status board. Docs (fix): iss-363 /abcd:lint + /abcd:prepare-this-repo enumerated five lint conventions omitting the shipped identity-positioning rule; iss-367 /abcd:launch documented a nonexistent top-level files count for --dry-run --json (the list is bundle.files). Infra (fix): iss-365 the govulncheck CI job pinned @v1.1.4, whose bundled x/tools caps its type-checker at go1.24 while go.mod declares go 1.25.6, so the scan failed at package loading and analysed nothing — masked as an advisory red since it landed 2026-08-19; bumped to @v1.7.0 (verified locally to load go1.25). Record (internal): iss-358 three ADRs (adr-11/13/20) ended with leaked tool-call scaffolding (</content></invoke>) — deleted; iss-359 the ADR index listed adr-44 proposed vs the record's accepted — corrected; iss-360 adr-7 carried no frontmatter against the ADR contract (sole outlier of 36) — added; iss-368 the core/disembark glossary term linked its counterpart to the interview-context embark (a different bounded context) rather than the /abcd:embark unpack surface — repointed.

Ledger hygiene: resolved the stale iss-338 (round 2's persona-pronoun record already fixed by round 1's iss-321; all ten cited sites are they/them today, repo-wide sweep clean).

Recorded, not fixed (stays open): iss-369 — readTranscript opens the hook-supplied transcript path without O_NOFOLLOW, the last bespoke external-input read not routed through fsutil.ReadGuarded. Refuted as having no reachable exploit (the path is harness-supplied and no path policy exists, so a direct path already reads any file — O_NOFOLLOW closes nothing); defence-in-depth consolidation only, left for a scoped follow-up.

Considered and rejected / downgraded: the readTranscript exfiltration framing (refuted to defence-in-depth only); the termsafe zero-width gap (downgraded substantive→nitpick — visual hiding only, no escape/reordering). A batch of unrefuted infrastructure/record nitpicks (rehearsal verify gating, CODEOWNERS omissions, several stale intent-bucket path references, the AGENTS.md inert-stand-down description) was triaged as nitpick and left for a future round rather than widening the diff.

Model routing: orchestration on Claude Fable 5; five parallel hunters, six per-finding adversarial refuters, and one of the two pre-merge reviewers on Claude Opus 5; the second pre-merge reviewer on Claude Fable 5 (dual-model merge gate). Round summary also appended to .abcd/work/DECISIONS.md in the PR.

Assisted-by: Claude:claude-fable-5

---

Generated by Claude Code

[REPPL]

Round 2 — findings: 12 substantive, 15 nitpick, 8 refuted; nitpicks-only: no; PR #386 open (all 13 CI checks green, dual adversarial review posted; the maintainer enabled auto-merge and the PR entered the merge queue — the loop performed no merge action).

Branch: bughunt-b/round-2. Baseline (make preflight + gofmt -l .) green before any change and on the final merged tree.

Fixed (11 records resolved + iss-201 closed): iss-359 redirect-controlled FinalURL carrying C1/bidi/zero-width runes raw into --json output and the committed citations baseline (byte-wise percent-encoding at the fetch boundary, a Baseline.validate rung, corrected termsafe/coverage doc claims); iss-346 guard "also matched" dropping a matched warn under a synthetic block; iss-347 readTranscript symlink-follow + truncation (now fsutil.ReadGuarded); iss-201 hook stdin over-cap misdiagnosis (cap+1 at readHookInput and the guard hook); iss-348 the root command's hidden record-id dispatch; iss-349 the terminology page calling the fail-open-loud guard fail-closed; iss-350 the README's unkept abcd ahoy PATH-gap promise; iss-351 the three-vs-six disciplines roster (six sites → pointers); iss-352 phantom research/adr/ citations; iss-353 the note-naming rule contradicting adr-30; iss-354 a mutation-verified test coupling ci.yml's go-version pins to the scaffold substitutions; iss-356 the 13-item nitpick batch (urlguard CGNAT/RFC 6890 ranges, repolint not-scanned warn, 64-hex resolve sha, three CLI help truths, release.yml wording + templates, dependabot pip, brief roster/narration/stale-pointer cleanups).

Built, reviewed, reverted (iss-358, wontfix): the scanner hostname trailing-\b fix — the residual of hunt A's iss-307 sweep — was implemented with watched-fail tests, then reverted when this PR's own pre-merge adversarial review proved it flags snake_case selectors (stream.local_addr, args.local_rank) that Stage-1 redaction would rewrite into the irreversible history store. TestSnakeCaseSelectorsStayQuiet pins the decision; the record annotates why the original whole-tree sweep was insufficient evidence.

Recorded, stay open: iss-355 — a batched merge-queue push tags the batch tip, so auto-release + the HEAD^2^ receipt derivation wedge a release roll merged as a non-final queue entry (iss-326's outcome by a route itd-93's pre-merge check cannot close); required-check workflow logic, left with two proposed fixes. iss-357 — the memory-ingest FinalURL sibling of iss-359.

Refuted (8): external-review pull_request_target check-sha wedge (empirically: check runs attach to the PR head); the docs-site-ungated claim (docs-lint gates docs/; iss-234); empty Diátaxis indexes (owned by iss-216/the facilitator plan); identity-pin email case-folding (byte-exact sh-gate parity is the contract); the guard backtick parser gap as a fix target (iss-148's reverted attempt; only its help-text disclosure was new); resolvedSHA 64-hex (iss-206 + no SHA-256 on that path); the "pin policy contradiction" reading of the tool pins; the .PHONY claim (inert).

Dedup boundary: iss-307/311/317/319/321/325 (hunt A) and iss-326/327 (the v0.6.0 wedge, dispositioned into itd-93) were held by the sibling loop or the maintainer and not re-reported. Two concurrent-mint id collisions (iss-344/345) were re-minted as iss-358/359 per protocol.

Model routing: orchestration on Claude Fable 5; hunters, refuters, and one pre-merge reviewer on Claude Opus 5; the second pre-merge reviewer on Claude Fable 5 (dual-model merge gate); a third independent verifier confirmed the NO-SHIP blockers closed. Round summary appended to .abcd/work/DECISIONS.md in the PR.

Assisted-by: Claude:claude-fable-5

---

Generated by Claude Code

[REPPL]

Round 3 (hunt B) — findings: 10 substantive, 5 nitpick, 10 refuted; nitpicks-only: no; PR #399 open (all 12 CI checks green; two independent adversarial reviews — Opus 5 SHIP, Fable 5 caught one N−1 manifest blocker now fixed in the final push; left for a human to merge, auto-merge not authorised).

Branch: bughunt-b/round-3. Baseline (make preflight + gofmt -l .) green before any change and on the final tree. IDs minted iss-382..iss-392 above the sibling bughunt-a/round-3's iss-374 (cross-ref floor fetched before capture; no collision).

Fixed (9 records, resolved this round). Security: iss-382 disembark plan's RenderManifest printed source-repo file paths to the terminal raw while the same view sanitised its neighbours (C1/bidi injection over an untrusted source repo, reproduced) — routed through the package sanitiser; iss-383 the history transcript store's read path used raw os.ReadFile while its write path was fully hardened, so a planted FIFO wedged history capture under the store flock and a symlink was followed (both reads now fsutil.ReadGuarded); iss-384 Baseline.validate sanitised final_url alone while echoing the entry key and quoted fields raw to the terminal through the CLI error surface (incomplete-fix residue of iss-359, reproduced end-to-end through the blocking docs-lint gate). Infra: iss-386 launch scaffold stripped the go.mod patch and refused one back, shipping every adopter a floating go-version (the adopter-side residue of iss-289). Docs/record: iss-385 commands/version.md still named only two network verbs after two rewrites (docs cite refresh and memory ingest also fetch); iss-387 the release-gate manifest pinned 17 briefDocs while five shipped surface chapters (guard/ideate/identity/banlist/update) sat outside it, so every full-tier crosscheck attested coverage it never ran — checkerCount → 27 (the fifth, 21-update.md, was the N−1 miss the Fable 5 pre-merge review caught in my first fix); iss-388 the phases README called the shipped itd-88 planned; iss-389 the reserved-identifiers principle described its shipped iss-154 lint as not-yet-built; iss-391 research/notes/README routed to the adr-30-rejected research/phase and research/adr layouts. Nitpick batch iss-392: canonical id-uniqueness key closing the zero-padded-filename hole (watched-fail), .yaml coverage in the go-version lockstep sweep, scripts/*.sh eol=lf, the lifeboat-reviewer 0.1.1 reconciliation, and the commands/abcd/ path repoint across five live records.

Recorded, not fixed (stays open): iss-390 — the pre-decided promotion of the reserved-identifiers principle to a discipline-kind intent is a maintainer ceremony, so it stays open as the tracker.

Refuted / prior art (10): scanner genericHomeRe case-fold (iss-308 already adjudicated against folding the POSIX /users/ arm — real API-route false positives; home_path_other is warn-tier); the ahoy refounding-candidate case compare (documented mutable labels; designed decline path; iss-221 owns the repair); the memory .coverage_index.json raw read (already named by open #365); govulncheck non-gating (documented deliberate advisory, DECISIONS 2026-08-19 / commit 61a15b4); the CODEOWNERS publish-surface omissions (.abcd/* reaches no installed user, the Go-tree reading proves too much, sole-maintainer bypass moots it); isHexSHA 40-hex (display-only, sha256 repos unreachable, iss-206 owns the sibling); commands/docs.md's scoped "on behalf of documentation" heading; the agents host-agnosticism claim (docs-lint bans harness names, not the sanctioned model-id disclosure token); the make smoke vacuous-gate claim; the runbook self-parity claim.

Model routing: orchestration on Claude Fable 5; five parallel hunters and seven per-finding adversarial refuters on Claude Opus 5; the dual pre-merge review ran one reviewer as Claude Fable 5 and one as Claude Opus 5 (dual-model merge gate). Round summary also appended to .abcd/work/DECISIONS.md in the PR.

Assisted-by: Claude:claude-fable-5

---

Generated by Claude Code

[REPPL]

Round 4 — findings: 8 substantive, 7 nitpick, 2 refuted; nitpicks-only: no; PR #409 open (all 12 CI checks green; two independent adversarial reviews — one Claude Opus 5, one Claude Fable 5 — both SHIP; left for a human to merge, auto-merge not authorised).

Branch: bughunt-b/round-4. Baseline (make preflight + gofmt -l .) green before any change and on the final tree. Fresh round: round 3's PR #399 was merged this morning, so this started from the latest main (c439cf4). Ids minted through the adr-45 timestamp-numeric scheme (iss-<yymmddHHMMSS><4 digits>), no collision protocol needed.

Fixed (12 records, resolved this round). Code, each with a test watched fail before the fix and pass after:

• iss-2608211132061930 (major) — lint.LoadConfig read the docs-lint/record-lint config with a raw os.ReadFile while its .abcd/*.json siblings guard.Load/rules.Load/positioning.LoadConfig all guard the read as a trust boundary. Hook-reachable (ahoy.Detect) and cross-repo-clonable, so a committed config symlink to a FIFO wedged docs lint/lint/ahoy/hook/cite refresh and the record-lint CI leg, /dev/zero OOMed the CLI, and an out-of-repo symlink target ran the whole ruleset from a file the repo does not own (reproduced in a fresh clone). Now fsutil.ReadGuarded + an Lstat dir-symlink refusal.
• iss-2608211134281912 — four stdin operand readers read exactly the cap and silently truncated an over-cap payload; worst on history capture -, which stored an 8 MiB prefix under a sha256 idempotency key over the prefix (breaking spc-4's refuse-whole invariant). Added readCappedStdin (cap+1 probe).
• iss-2608211139076040 — the iss-392 canonRecordID fix left four sibling lint keyspaces raw (delivery_state bucket + citation, spec intent-existence, superseded_by, spec_id_unique), so a zero-padded id slipped a gate open or false-blocked a link; routed every key and lookup through canonRecordID.
• iss-2608211140410761 — abcd adr-N confirmed the frontmatter id byte-exact while record-lint and the citation resolver treat padded/quoted/cased ids as one handle, reporting a present ADR absent (and abcd adr-0003 routing then refusing); now a parsed-handle compare rendering the canonical id.

Docs/record: iss-2608211142146585 + iss-2608211142142469 the release-gate runbook claimed the verify job runs macOS+Linux (it is ubuntu-only; ci.yml's check job is the matrix and the merge gate) and framed the semantic-receipt gate as dormant-until-public-flip though it is public and has fail-closed two releases (v0.3.0/iss-108, v0.6.0/iss-326); iss-2608211142496517 cmd/scaffold-sync named a "scaffold-sync workflow" that was built and rejected (iss-209); iss-2608211143184945 + iss-2608211143185943 the README Go badge (1.25→1.26) and the retired "next free id" capture description; iss-2608211144265046 commands/lint.md documented a line: 0 sentinel the omitempty JSON never emits; iss-2608211144555085 the ADR index omitted the accepted adr-45 row (third recurrence; iss-38 wants such indexes gated); iss-2608211145448208 the DCO deferral text adr-43 retired survived in AGENTS.md and scripts/check-attribution.sh; iss-2608211146222734 the banned pre-adr-30 roadmap/intents literal survived in AGENTS.md and .abcd/README.md, outside the lint roots; iss-2608211146562243 RD001 documented as unconditional though check-reviews.sh exempts 7 of 10 live dirs (sha-keyed receipts); iss-2608211147403129 the ADR↔RFC frontmatter pairing contract disagreed across three surfaces (reconciled rfcs/README.md and rfc-2).

Refuted / prior art (kept out of the ledger): the CHANGELOG [Unreleased] Fixed-vs-additive sectioning (the cut discards [Unreleased] prose and re-derives sections from records; owned by open iss-256); the .abcd/work tier-map "omission" (highlight enumerations, not inventories; iss-333 already blessed the table); and the launch.DenyNamespaces case-fold symlink slip (inert — the hardlink alias map rejects via aliasDenied).

Model routing: orchestration on Claude Fable 5; five parallel hunters and ten per-finding adversarial refuters on Claude Opus 5; the dual pre-merge review ran one reviewer as Claude Opus 5 and one as Claude Fable 5. Round summary also appended to .abcd/work/DECISIONS.md in the PR.

Assisted-by: Claude:claude-fable-5

---

Generated by Claude Code

[REPPL]

Round 5 — findings: 4 substantive, 9 nitpick, 9 refuted; nitpicks-only: no; PR #411 open (all checks green — ci, attribution, external-review; two independent adversarial reviews — one Claude Opus 5, one Claude Fable 5 — both SHIP; left for a human to merge, auto-merge not authorised).

Branch: bughunt-b/round-5. Baseline (make preflight + gofmt -l .) green before any change and on the final tree. Started from main at 9054b7e; PR #410 landed mid-round, so origin/main (06c3114) was merged into the branch and a commands/ahoy.md conflict resolved by union (main's reworded pinned install-mode text plus this round's (shadowed on PATH) note). Ids minted through the adr-45 timestamp-numeric scheme, no collision protocol needed.

Fixed — substantive (4 records, each with a test watched fail before the fix and pass after):

• iss-2608211432258689 + iss-2608211432257954 — abcd launch --dry-run output hardened. The refusal-reason lines were printed with %v and no sanitiser, so a committed filename carrying raw terminal escapes (a control-char-rejected path carries its own bytes) reached the terminal and any CI log; each reason now routes through termsafe.Sanitize like the citation line beside it. Separately, an unreadable pinned manifest embedded a raw *os.PathError absolute path in lockstep.detail — and the dry-run report is a success envelope that never passes the error-surface path scrub — leaking the developer-identity root into the report and its --json; stripped at loadJSON via pathFreeError, leaving the named cause.
• iss-2608211432258975 — abcd lint mapped a rule-engine fault (a stat hitting a symlink loop, an unreadable tracked file, a git-index fault) to exit 1 — the code the documented Conftest tri-state reserves for "warnings only" — so a CI gate keying on exit >=2 read a lint that never ran as an advisory pass. An engine fault now exits 2; clean→0 and warnings→1 are untouched.
• iss-2608211432254405 — launch scaffold deriveBranch assumed .git is a directory, so in a linked worktree or submodule (where .git is a gitfile) both ref reads failed and the fallback branch main was stamped into the generated release workflows — leaving the release gate silently inert on a non-main-default repo. Now resolves the gitfile: HEAD from the worktree gitdir, origin/HEAD from the shared commondir.

Fixed — nitpick (9 records): iss-2608211432389181 the intent verdict reader was the last CLI operand on Lstat-then-os.ReadFile (with a false benign-TOCTOU comment) — routed through fsutil.ReadGuarded; iss-2608211432384430 the history rootSHA diagnostic named only 40 chars though the regex accepts 64 (SHA-256) — now a shared const naming both; iss-2608211432384091 / iss-2608211432389477 / iss-2608211432389791 three user-facing doc claims (README SessionEnd self-bootstrap overclaim, memory.md source.class→source_class, ahoy.md missing the (shadowed on PATH) install_mode suffix); iss-2608211432489481 / iss-2608211432489288 / iss-2608211432482363 / iss-2608211432483805 record hygiene (seven link display-text/href mismatches, two bare paths to absent dirs, the rfc-1/itd-26 one-sided pair, the .abcd/work tier roster omitting the ledger/reviews/rulesets).

Refuted / considered-and-rejected (9, kept out of the ledger): the ScrubbedEnv GIT_CONFIG_GLOBAL gap — the two refuters split (one CONFIRMED, one REFUTED); adjudicated refuted, because keeping the developer's global git config readable is deliberate and test-pinned (TestScrubbedEnvStripsHijackKeepsGlobalConfig), the attack needs control of abcd's process env (out of scope under the trusted-env model), and the obvious fix (scrub to /dev/null like IsolatedEnv) would blind the redaction probe for every developer whose identity lives in global config. Also refuted/deferred: folding local_username to case-insensitive (net-negative — broadens a documented hard_fail false-positive class and reopens iss-31); widening hasControlChar (the render sanitiser, not the integrity gate, is the fix — the render site is the one fixed above); launch --no-index dropping a force-added tracked file (dry-run writes no artefact, the exclusion is visible, over-exclusion is the safe direction); the guard check exit-2 enumeration (the governing "could not be evaluated at all" definition already covers a disabled registry); itd-3's spc-1 (a documented, minting-honored reservation); the CHANGELOG dangling-id and two other nitpicks (maskSecret byte-vs-rune snippet on already-safe code, DetectLicence unreachable dead-code read) left unactioned as real-but-low-value.

Model routing: orchestration on Claude Fable 5; five parallel hunters and nine per-finding adversarial refuters on Claude Opus 5; the dual pre-merge review ran one reviewer as Claude Opus 5 and one as Claude Fable 5. Round summary also appended to .abcd/work/DECISIONS.md in the PR. (Note: the harness auto-appends a footer to this comment that this session cannot strip — the direct GitHub API is disabled here and no comment-edit tool is exposed; the intended final line is the trailer below.)

Assisted-by: Claude:claude-fable-5

---

Generated by Claude Code

[REPPL]

Round 6 — findings: 10 substantive, 3 nitpick, 7 refuted/withdrawn; nitpicks-only: no; PR #414 open (all checks green — ci, attribution, external-review; two independent adversarial reviews — one Claude Opus 5, one Claude Fable 5 — both initially NO-SHIP, both SHIP after remediation; left for a human to merge, auto-merge not authorised).

Branch: bughunt-b/round-6. Baseline (make preflight + gofmt -l .) green before any change and on the final tree. Started from main at b67e231; ids minted through the adr-45 timestamp-numeric scheme, no collision protocol needed.

Fixed — substantive (10 records): Security/code (each with a watched-fail test): iss-2608211849467013 the guard tokenizer did not know shell redirection operators, so a glued git push --force>/dev/null mutated the flag token and the blocker missed — a silent allow — and a leading redirection degraded a Tier-1 block to a warn; iss-2608211849463840 the record/docs lint reporting walk (and, after both pre-merge reviews caught the sibling, the glossary_dir walk in loadForbiddenSynonyms) read a cloned-repo-controlled config path with a raw os.ReadFile and no containment, so a committed config could read/lint/follow a file outside the tree (reproduced: escaping root, symlink leaf, /dev/zero OOM); iss-2608211849461061 frontmatter.Fields required --- on line 0, so a comment-led record (13 glossary term files) slipped the no_git_metadata blocker; iss-2608211849466878 citation DaysBetween took each date's local calendar day, making staleness and the release-gate overdue blocker timezone-dependent; iss-2608211849468791 the attribution gate false-red a CRLF pull-request body on the required check. Doc/record: iss-2608211849581263 + iss-2608211849583757 the persona brief said the picker chooses at random and omitted Nia (a residual of resolved iss-300 — by role, roster of fourteen); iss-2608211849583208 the bundled INTENTS default persona rule reached this repo's own intent authors, so .abcd/rules.json now overrides INTENTS toward the registry; iss-2608211849580624 the README described download-on-every-update provisioning and a no-op .binary-meta remedy, rewritten for the spc-35 persistent cache; iss-2608211849582190 CONTRIBUTING granted the fenced-quotation carve-out for commit messages too, now scoped to the pull-request body.

Fixed — nitpick (3 records): iss-2608211850070541 identity init faults exit 2 (was 1); iss-2608211850074600 the lifeboat probe reads cap+1 and refuses a file grown past the cap (size TOCTOU, closed by construction — its boundary test guards the cap, not the race); iss-2608211850070318 history list --json emits [] not null.

Recorded, not fixed (stays open): iss-2608211914592726 — the residual config-derived os.ReadFile sites in internal/core/lint (persona registry, spec docs, changelog, index-drift, subverb snapshots, record stores) share the cloned-repo trust boundary the roots/glossary walks now guard; triage and sweep in a follow-up (both pre-merge reviews flagged these; kept out of this PR's scope, which was narrowed accordingly).

Refuted / considered-and-rejected (7): R3 the release-gate manifest prompt naming ten of twenty commands (an illustrative grounding string, not a gated inventory; editing it ripples into receipt hashes); R4/R5 related_adrs and prose citations of superseded ADRs (resolve via the retired-set mechanism — deliberate historical provenance); R6 "11 adapters" vs "five capability seams" (implementations vs categories); R7 the adr-13 bare path to itd-36's old bucket (a benign ungated breadcrumb; the authoritative related_intents resolves by id); the updater's redirect-host case-fold (fails closed, availability only, impractical to test through a loopback redirect — withdrawn); and I2, the release.yml "dormant until the public flip" comments, which were implemented then reverted — release.yml is a scaffold template rendered for adopters who may be private, so the visibility-flip framing is generic-correct there, unlike the abcd-specific runbook round 4 corrected.

Pre-merge review note: both reviewers independently returned NO-SHIP on the first head over the same two issues — an unswept glossary_dir sibling read path (in the function this round edited) and a vacuous probe cap-boundary test — both fixed in follow-up commits (826b304, 98cc623); both re-verified and returned SHIP, with a fresh symlink-leaf escape probe confirming refusal and the new glossary test confirmed to fail on the pre-fix code.

Model routing: orchestration on Claude Fable 5; five parallel hunters and eight per-finding adversarial refuters on Claude Opus 5; the dual pre-merge review ran one reviewer as Claude Opus 5 and one as Claude Fable 5. Round summary also appended to .abcd/work/DECISIONS.md in the PR.

Assisted-by: Claude:claude-fable-5

---

Generated by Claude Code

[REPPL]

Round 7 — findings: 11 substantive, 4 nitpick, 5 refuted/downgraded (+2 prior art, +1 recorded-open); nitpicks-only: no; PR #439 open (all checks green — ci, attribution, external-review; two independent adversarial reviews — one Claude Opus 5, one Claude Fable 5 — both initially NO-SHIP on the same blocker, both SHIP after remediation; left for a human to merge, auto-merge not authorised).

Branch: bughunt-b/round-7. Baseline (make preflight + gofmt -l .) green before any change and on the final tree. Started from main at a2c1843; ids minted through the adr-45 timestamp-numeric scheme.

Note on numbering: an earlier bughunt-b/round-7 ran and merged into main earlier today (the guard &> and frontmatter-BOM fixes now in [Unreleased]), then its branch was deleted, but it posted no round record here — so the state issue's last bughunt-b entry was Round 6 and this round took the free round-7 slot on the current main. When this run started, git ls-remote origin 'bughunt-b/*' and the open-PR list were both empty, so it correctly began a fresh round; its findings are entirely distinct from and additive to that earlier merge.

Fixed — substantive (11). Security/code (each with a watched-fail test): the guard tokenizer now recognises $'...' ANSI-C and $"..." locale quoting — a Tier-1 blocker bypass where $'--force' tokenised to $--force while bash handed the child byte-identical argv (iss-2608221456463223; the dual pre-merge review caught that the first fix omitted bash's \u/\U/\c escapes, closed in the final push and re-verified against real bash); the hard-fail local_username redaction matcher folds case, closing a leak of the caller's own login on a case-folding filesystem while keeping iss-31 suppressed (iss-2608221456469938); five site-generator fixes — the SVG prolog/comment/CDATA build-vs-check disagreement, unparsed-page gate masking, an unscreened repository href scheme, a record-id collision, and non-ASCII path dates (iss-2608221456462677, …469558, …466438, …457123924); the launch scaffold's symlink path leak and the ./internal/... race leg that wedged a bare adopter (iss-2608221456599559, …456597173); a capture skipped[].error absolute-path leak (iss-2608221456599229); CODEOWNERS coverage of site-src/ and docs/requirements.txt (iss-2608221456593173). Docs/record: the writing-style escape scope + names/* row (iss-2608221456593814); the site-build "exactly this set" drift (iss-2608221457125132); the shipped abcd site check called shipped in the brief and pinned in the release-gate manifest (iss-2608221457122761); the 03-invariants.md PLACEHOLDER banner over fourteen live invariants and the brief/README.md evidence-chapter "placeholders" (iss-2608221457127152, …457128333).

Fixed — nitpick (4): principle-id collision refusal + core.quotePath=false (iss-2608221457123924), the 00-meta.md open-questions clause, and the --no-color/Windows-planned/dry-run---json doc-string corrections.

Recorded, not fixed (stays open): iss-2608221457227161 — the guard brace-expansion bypass ({--force,} → byte-identical argv), a distinct expansion-not-quoting follow-up larger than this round's scope.

Refuted / downgraded / prior art: the commands/site.md "runs in CI" audit claim (in-flight PR #436); the install-surface transport-pin and -q deltas (defence-in-depth, no reachable attack); the iss-315 guard-warn citation; the CI record-lint job name (prior art iss-304) and the prepare-this-repo template path (prior art iss-87).

Model routing: orchestration on Claude Fable 5; five parallel hunters, the per-finding adversarial refuters, and one of the two pre-merge reviewers on Claude Opus 5; the second pre-merge reviewer on Claude Fable 5 (dual-model merge gate), plus a dual re-verification of the remediation (one Opus 5, one Fable 5). Round summary also appended to .abcd/work/DECISIONS.md in the PR.

Assisted-by: Claude:claude-fable-5

---

Generated by Claude Code

[REPPL]

Round 7 merge-gate resume — PR #439 merged (by the maintainer, c354114). No new hunt this run per protocol (a bughunt-b/* PR was open).

This run found the PR conflicted against a moved main (#440, #441, #432 landed after the first gate pass) and drove it back to green: merged main into bughunt-b/round-7 (0778f56), resolving the two site-brief-chapter conflicts by taking main's newer website-ships prose (which postdates and subsumes the round's corrections of those passages) while keeping the round's 22-site.md inputs-table rows main lacked. make preflight + gofmt -l . green on the merged tree before pushing; all 13 CI checks green on the new head; two fresh independent adversarial re-reviews (one Claude Opus 5, one Claude Fable 5) both returned SHIP, each proving the merge lossless mechanically (two-dot = three-dot diff vs origin/main; merge-tree replay showing exactly the two expected conflicts). The maintainer merged while the re-verification ran; the merged tree is byte-identical to the reviewed head, so both verdicts apply to what shipped. Verdicts + CI status posted on the PR.

Round 7's findings tally stands as previously recorded (11 substantive, 4 nitpick; nitpicks-only: no — streak unchanged). Reviewer follow-up candidates for a future round: commands/site.md writes-list and 04-surfaces/README.md row-22 gate enumeration now narrower than the merged 22-site.md; the site record-id collision guard does not cover two frontmatter-free stores sharing an id.

Assisted-by: Claude:claude-fable-5

---

Generated by Claude Code

[REPPL]

Round 4 — findings: 0 substantive, 2 nitpick, 0 refuted; nitpicks-only: yes; PR none (a nitpick-only round ships no PR — per protocol it skips to REPORT).

Branch: bughunt-a/round-4, cut from main at 3b5f091 (v0.6.3). Baseline green before any change: make preflight exit 0 with the full suite executed (no silent skip — 40 packages reported real test times incl. surface/cli 197s, lifeboat 54s), and gofmt -l . empty.

Hunt. Five parallel Opus 4.8 hunters swept tracked files across all four dimensions — (a) code bugs in the source tree, split core-trust-boundary vs newer/recently-changed packages; (b) docs vs functionality; (c) infrastructure; (d) internal record consistency. Each hunter read the committed playbook (2026-07-17-bug-hunting-principles.md), CONTEXT.md, the DECISIONS.md tail, and the 254 open ledger records to dedup against prior art before reporting.

Result — three honest zeros, two nitpicks:

• Code / core trust boundaries — []. Traced every hardened primitive (fsutil.ReadGuarded(InRoot), gitutil.IsolatedEnv/ScrubbedEnv, termsafe.Sanitize, (?i) identity matchers, guard tokenizer, urlguard) and its unhardened siblings; each remaining os.ReadFile/os.Environ/raw-print instance proved hardened or explicitly out of scope under the trusted-worktree model. Candidates mapped to open prior art (iss-2608231025198888 capture-redaction, iss-369 readTranscript, iss-322 lifeboat atoi).
• Code / newer packages (site, update, release, cite, recordid, rules, banlist, frontmatter, vintage) — []. The site generator (SVG inliner, htmlscan output tokenizer, CSL renderer), cite redirect re-guarding + encodeHiddenRunes, update checksum case-normalisation, and the frontmatter first-key-wins scanner are all hardened or documented-intentional; the one concrete candidate (svgSizeRe integer-only) is a documented constraint at assets.go:79-82, and the loose-StripFrontmatter sibling is already open (iss-2608221126066379).
• Infrastructure — []. Makefile targets match their doc claims; the ci.yml fail-closed classifier stands down only on the exact docs/+.abcd/development+.abcd/work+root-prose allowlist and runs the full matrix on any code-carrying or non-PR path; every uses: across all 7 workflows + both scaffold templates is 40-hex SHA-pinned; release.yml/auto-release.yml check out immutable SHAs with a least-privilege floor. One narrow-assumption candidate (check-reviews.sh:30 SHA-1-only {40} receipt-dir match) self-refuted as unreachable (abcd's repo is SHA-1; the script is not scaffolded to adopters).

Nitpicks (2, confirmed real, cosmetic — left for a future substantive round to batch, not captured to the ledger since nothing was fixed):

1. capture resolve --commit flag help under-specifies. internal/surface/cli/cli.go:2294 (and the generated docs/reference/cli/commands.md:191) reads "the fixing commit sha (7-64 hex chars, shape-checked only)", but the validator is reCommitSha = ^[0-9a-f]{7,64}$ (internal/core/capture/capture.go:248) and the rejection message itself says "7-64 char lowercase hex sha" (workflow.go:239). The help omits "lowercase"; a user is nonetheless told at the point of failure. Fix would be a one-word help edit + go generate regen.
2. Persona they/them discipline (itd-79) drift in planned/ intents. itd-112 (Press Release, lines 25/26/33 — Alice "she already knows"/"greets her", Carol "her unstamped build") and itd-133 (line 32 — Alice "her terminal") carry she/her for the project's own personas. This is a sweep-incompleteness straggler: the resolved sweeps iss-321/iss-338 scoped only to drafts/+shipped/ ("seven outliers") and never reached planned/; the persona_registry lint matches only said <Name>, attributions, so pronouns pass the mechanical gate silently. The itd-27 "Pocock/his" hit is an external author (correctly exempt per the iss-321 convention).

Neither survives upgrade to substantive (refutation only downgrades). Both candidates were adversarially self-refuted inline (they are real, not false positives, but cosmetic); no per-finding refuter fleet or dual-model merge-gate review was spawned because a nitpick-only round ships no fix to gate. This is the first nitpick-only round in the streak (the prior three trusted round records are all nitpicks-only: no), so no LOOP-STOPPED.

Model routing: orchestration on this session (claude-opus-4-8); five parallel hunters on Claude Opus 4.8 via the Agent tool. No .abcd/work/DECISIONS.md append this round — that step is part of a round's PR, and there is no PR.

(Note: the cloud harness auto-appends a _Generated by [Claude Code]_ footer to this comment that this session cannot strip — no comment-edit/PATCH tool is exposed here. The intended final line is the trailer below.)

Assisted-by: Claude:claude-opus-4-8

---

Generated by Claude Code

[REPPL]

Round 5 — findings: 1 substantive, 1 nitpick, 2 refuted; nitpicks-only: no; PR #480 open (all checks green — ci, attribution, external-review; two independent adversarial reviews both SHIP; left for a human to merge, auto-merge not authorised).

Branch: bughunt-a/round-5. Baseline (make preflight exit 0 with the full suite executed — no silent skip — plus gofmt -l . empty) green before any change and on the final merged tree. Started from main at 3b5f091 (v0.6.3); PR #479 landed mid-round, so origin/main (57781fe) was merged into the branch (clean, no conflicts — main touched no file this round touches) so CI evaluates against the current base.

Fresh-round note: the previous bughunt-a round record (Round 3, PR #391) was closed unmerged by the maintainer, and Round 4 already ran afterward as a fresh round — so the leftover bughunt-a/round-3 branch on origin is a stale remnant, not a died-between-push-and-PR case, and was correctly not resumed. This is a fresh Round 5.

Fixed — substantive (1):

• iss-2608240806090544 — .abcd/development/intents/planned/README.md asserted, present-tense and categorically, that "Each intent here carries a bidirectional link to its spec (spec_id: spc-N)". False: spec_id: null is a legal, lint-enforced state for a planned/ resident (the intent_lifecycle rule, transcribed in the surface contract 05-intent.md:333), and 20 of 43 planned intents carry it. The claim contradicted the parent lifecycle table (intents/README.md:54, :261 naming itd-20/24/63/69 as current null residents, :292) and adr-34's lifecycle/scheduling orthogonality — silently reintroducing the exact over-narrow framing adr-34 had already corrected in the parent README (resolved iss-3). Rewritten to the authoritative wording. A documentation-only correction (no Go behaviour changed, so no unit test); captured to the ledger before the fix and resolved with fixing-commit provenance (52988d9). No CHANGELOG entry — internal durable-record correction, not shipped in the binary.

Considered and rejected / refuted (2 + 1 nitpick):

• drafts/README.md (sweep-completeness sibling) — refuter-cleared: it describes the drafts→planned promotion transition (the plan verb's effect), not current contents, and makes no false categorical claim. Left untouched to avoid scope-creep on a defensible line.
• gitutil.scrubGitVar / GIT_CONFIG_GLOBAL (core-code hunter candidate) — prior art: adjudicated REFUTED in round 5's own DECISIONS entry (deliberate, test-pinned TestScrubbedEnvStripsHijackKeepsGlobalConfig; out of scope under the trusted-env model).
• Nitpick (1, not fixed): persona they/them drift in a frozen dated implementation-prompt transcript (research/abcdev-site/abcdev-implementation-prompt.md:100, Alice "she and her"). A historical session artefact, not living persona prose; the shipped intent it seeded (itd-135) already reads they/them. Cosmetic and self-refuted — left for the record.

Genuine nulls: the code (core trust boundaries + newer packages), infrastructure, and docs-vs-functionality hunters returned honest zeros on this tree — consistent with round 4's null on the identical 3b5f091 tree. No findings were invented to fill a quota. This round has a confirmed substantive fix, so it resets the nitpick-only streak.

Model routing: orchestration on this session (claude-opus-4-8); five parallel hunters and one per-finding adversarial refuter on Claude Opus 4.8 via the Agent tool. Dual pre-merge review: one reviewer on Claude Opus 4.8. Claude Fable 5 was unavailable this run (account rate limit), so the second reviewer ran on Claude Sonnet 5 — a genuinely different model from the first, preserving two-model independence; noted for transparency. Round summary also appended to .abcd/work/DECISIONS.md in the PR.

(Note: the cloud harness auto-appends a _Generated by [Claude Code]_ footer to this comment and to the PR merge-gate comment that this session cannot strip — no comment-edit/PATCH tool is exposed here. The intended final line of each is the trailer below.)

Assisted-by: Claude:claude-opus-4-8

---

Generated by Claude Code

[REPPL]

Round 5 (hunt-a, second concurrent firing) — findings: 2 substantive, 2 nitpick, 1 refuted (+1 deferred); nitpicks-only: no; PR #481 open (all checks green — ci, attribution, external-review; two independent adversarial reviews — one Claude Opus 5, one Claude Sonnet 5 — both SHIP; left for a human to merge, auto-merge not authorised).

Concurrency collision. Two firings of the hunt-a routine ran at once on round 5. The first took bughunt-a/round-5 and opened PR #480 (planned/ bucket README spec-link overclaim, iss-2608240806090544) while this run was still hunting. When this run tried to push bughunt-a/round-5, the branch was already a live peer's — and it kept advancing (b783352 → 0a1bf46 during this run), so pushing onto it would have clobbered or raced live work (force-push is refused by policy anyway). This run's findings are disjoint from #480 (shipped/README.md vs planned/README.md), so they ship on bughunt-a/round-5-2 as PR #481 rather than compete for the branch. The two PRs are independently mergeable.

Branch: bughunt-a/round-5-2, cut from main at 3b5f091 (v0.6.3). Baseline green before any change: make preflight exit 0 with the full suite executed (no silent skip — surface/cli 198s), gofmt -l . empty. Ids minted through the adr-45 timestamp-numeric scheme.

Fixed — substantive (2 records, resolved this round). Internal record consistency:

• iss-2608240815458212 — .abcd/development/intents/shipped/README.md:3-4 claimed the directory "is empty until the Go build ships capabilities into it (Phase 1 onward)", but it holds 18 shipped intents and the parent index (intents/README.md:313-315) already reads populated. A stale pre-lifecycle artefact, self-contradicted by line 6 of the same file. Rewritten to the accurate move-in lead.
• iss-2608240815453131 — the same front-door (shipped/README.md:6) named the retired intent-fidelity-reviewer agent (renamed to intent-auditor, spc-28; landed — agents/intent-auditor.md exists, no live intent-fidelity-reviewer.md, intent review retired). The exact twin of resolved iss-334, which swept the parent index but missed this sibling front-door (it falls between spc-28's two enumerated lists). Repointed.

Refuted (1). The round's one code candidate — frontmatter.Fields accepts duplicate top-level keys (first-wins) while capture's parser rejects them — was refuted by an independent refuter: every lint rule that reads status consumes the same first-wins map (lint.go:1885/2059/2479), so no split exists within record-lint; a single status: superseded already grants the identical, intended, iss-39-audited historical-record content exemption, and the duplicate key buys nothing. First-wins is the documented contract (frontmatter.go:3-6,30); capture's strictness is a deliberate mutable-store divergence; the roots are disjoint (.abcd/development vs .abcd/work/issues). Sibling-parser divergence is prior art (iss-15/ADR-32, iss-285/287, iss-2608221126066379).

Nitpicks (2, real, not fixed). (1) scanner.ScanBundle (scanner/scanner.go:705) reads bundle files with a raw os.ReadFile — the last unhardened sibling — but reachable only over the owner's own repo being packaged (trusted-worktree, out of scope). (2) Persona she/her for Alice in research/abcdev-site/abcdev-implementation-prompt.md:100 — the first firing (PR #480) logged this considered-and-rejected as a frozen dated implementation-prompt artefact (the seeded shipped intent already reads they/them); dropped here too.

Deferred (1). disciplines/itd-37 also names the retired agent (lines 5/41/60/82), but the project has demonstrably not swept discipline bodies (sibling itd-1 keeps 9 old-name references post-sweep iss-320), so touching only itd-37 would create inconsistency — left for a maintainer-scoped sweep.

The other three hunt dimensions returned honest nulls on this tree: code (core trust boundaries + newer packages), infrastructure, and docs-vs-functionality — each traced its hardened-primitive siblings and live command surface and found nothing substantive beyond prior art.

Model routing: orchestration on this session (claude-opus-4-8); five parallel hunters, three per-finding adversarial refuters, and one of the two pre-merge reviewers on Claude Opus 5 (served by Opus 4.8 in this environment; no separate Opus 5 model override is exposed here). The dual merge-gate review's second reviewer was to be Claude Fable 5, but Fable 5 hit a usage limit mid-run, so the second independent model was Claude Sonnet 5 — both reviewers genuinely distinct models, neither the author. Round summary also appended to .abcd/work/DECISIONS.md in PR #481.

This round is nitpicks-only: no (two confirmed substantive record-drift fixes), so the streak resets and no LOOP-STOPPED applies.

(The cloud harness auto-appends a _Generated by …_ footer to this comment that this session has no tool to strip — no comment-edit/PATCH tool is exposed here. The intended final line is the trailer below.)

Assisted-by: Claude:claude-opus-4-8

---

Generated by Claude Code

[REPPL]

Merge-gate resume (hunt-a) — no new hunt this run per protocol (two bughunt-a/* PRs are open). Both prior round-5 PRs re-verified and both remain ready for a human merge; the loop is blocked until they land or are closed.

• PR Bug-hunt round 5: correct planned/ bucket README spec-link overclaim (iss-2608240806090544) #480 (bughunt-a/round-5, iss-2608240806090544, planned/README.md spec-link overclaim) — mergeable_state: clean; all required checks green on head 0a1bf46 (ci, attribution, external-review success; heavy legs stood down by the inert-path classifier as designed); dual-review SHIP/SHIP already posted.
• PR Bug-hunt round 5 (concurrent firing): correct shipped/ bucket README drift #481 (bughunt-a/round-5-2, iss-2608240815458212 + iss-2608240815453131, shipped/README.md drift) — mergeable_state: clean; all required checks green on head 6067887 (the earlier attribution red was the harness-footer on the first push, superseded by a passing re-run); dual-review SHIP/SHIP already posted.

No regression to fix on either PR (no merge conflict, no red CI), so no push was made. Auto-merge is not authorised, so both stay open for a human to merge — the loop cannot start a new hunt while a bughunt-a/* PR is open. Owner action needed: merge or close #480 and #481 to unblock the next round.

Round-3 stale branch: unchanged — already adjudicated in the Round 5 record above (PR #391 closed unmerged, round 4 ran fresh, the leftover bughunt-a/round-3 branch is a stale remnant, not a died-between-push-and-PR case; correctly not resumed).

(The cloud harness auto-appends a _Generated by …_ footer to this comment that this session has no tool to strip; the intended final line is the trailer below.)

Assisted-by: Claude:claude-opus-4-8

---

Generated by Claude Code