[finding] 39 scripts/ modules export bindings with no entry guard — the class has now been fixed one file at a time twice, and check:entry-guard accepts "no guard at all" by design #10631

Activity

1. claude

   added

   tooling

   domain:devx

   finding

   on Aug 21, 2026
2. claude

   assigned

   os-zhuang

   on Aug 21, 2026
3. claude

   added

   pm:dispatched

   on Aug 21, 2026
4. claude commented on Aug 21, 2026

   

   claudebot

   on Aug 21, 2026 – with Claude

   ContributorAuthor

   More actions

   Claim: /pm-dispatch domain:devx execution seat.
   Session: session_01DdCnBGcHeufjrq7drTD3wt
   Branch: claude/issue-10631-entry-guard-class
   Worktree: objectstack-issue-10631
   Domain: domain:devx
   File surface: scripts/check-entry-guard.mjs preferred; a 39-file sweep only if the gate-level answer is ruled out with evidence

   ⭐ PM steer, on the record before work starts: this class has now been fixed one file at a time twice (#9757, then #10610/PR #10628), and the card's own framing is that check:entry-guard accepts "no guard at all" as an explicit self-test assertion. Three instances of a shape is the argument for asking whether the fourth belongs in the gate rather than in a 40th ticket. ⛔ Do not open with a 39-file sweep; price the gate-level answer first and argue whichever way the measurement points.

   ⛔ Governed surface out of scope (docs/adr/**, .claude/**, skills/**, AGENTS.md, CLAUDE.md, content/docs/releases/**) and the #9465 epic fence (.changeset/ tooling, cut-rc.yml, release.yml, root package.json).
   ⚠️ No formatter of record (no format/prettier script, prettier not in devDeps, no .prettierrc/.editorconfig) — ⛔ do not run prettier --write (#10622).
   ⭐ Changeset: AGENTS.md:943 — "Pure bug fixes do not require a changeset"; a scripts/**-only diff that publishes nothing takes the skip-changeset label via the additive POST /labels (⛔ never a whole-set PUT — see #5533). Precedent: PR #10502.

   ---

   Generated by Claude Code
5. os-zhuang

   added a sub-issue on Aug 21, 2026

   • Burn down the 35 scripts/ modules that run on import — the KNOWN_IMPORT_UNSAFE ledger from PR #10665 #10667
6. claude commented on Aug 21, 2026

   

   claudebot

   on Aug 21, 2026 – with Claude

   ContributorAuthor

   More actions

   {
     "issue": 10631,
     "status": "done",
     "branch": "claude/issue-10631-entry-guard-class",
     "pr": "https://github.com/objectstack-ai/objectstack/pull/10665",
     "premise_still_valid": true,
     "summary": "Re-derived on today's origin/main (9dd192d48b): 123 scanned / 42 guarded / 39 exporting-and-unguarded - the card's numbers reproduce exactly. GATE, not sweep, and the measurement decides it. Every exporting scripts/ file was imported in a child process for its exports alone with a trailing ALIVE marker: of the 42 guarded, 33 loadable ones were all silent+alive (0 dead, 0 noisy); of the 39 unguarded, 8 killed the importer (FIVE of them with exit 0), 14 ran the tool loudly, and 7 were silent. Those 7 - adr-anchors, cli-build-prerequisite, console-spec-probes, eslint-fatal-guard, eslint-stack-headroom, i18n-bundle-surface, regen-artifacts - are pure library modules and are the legitimate exceptions the card asked me to measure before proposing a rule. They cost ZERO allowlist entries because the rule is 'a file that exports must have every top-level statement that RUNS behind the guard', which structurally never reaches a module of declarations. Static rule and behavioural probe agree on all 29 loadable files (7 silent / 22 not), no disagreement either way - that agreement is the licence to make it a rule. The 35 files that violate it go into an inline shrink-only KNOWN_IMPORT_UNSAFE debt list with the expanding remedy REFUSED outright (#8435 route b, so no MAINTAINER-ONLY token and no CONTROL entry in check-ratchet-remedy-authority is owed - that gate is green and still classes this file as handing out no ratchet-expanding remedy); a fixed file fails as STALE and names itself, so the list cannot rot into an allowlist. Burn-down filed as sub-issue #10667 and deliberately excluded: the claim declares scripts/check-entry-guard.mjs, and 35 gate scripts is the surface other seats work in. KEY CORRECTION TO THE CARD: 39 is keyed on guard PRESENCE, which is the wrong property. Keyed on what actually RUNS the population is 35, and its membership is not a subset of the 39: 11 files are guarded through a const alias (const isMain = isEntrypoint(import.meta.url); if (isMain)...), including the inverted 'if (!invokedDirectly) {} else if' idiom - my first cut called all 11 violations - and 3 files the card counts as ALREADY FIXED are leaking: qa/qa-rollup.mjs runs its self-test inside any importer whose argv carries --self-test, objectui-range.mjs prints help for --help, and ts-parse.mjs:462 registers a process.on('exit') hook under an env var. Both measurable ones are silent with a clean argv, so the card's option C (behavioural) would have missed all three. That argv-conditional shape is exactly what pm/check-governed-prose.mjs already hand-works-around by filtering --self-test out of process.argv before importing a sibling.",
     "tests": "All at final commit 4a83b1a3db (union re-derived AFTER the final commit). No build and no dist anywhere in this change - these are plain node scripts - so no ablation rebuild step applies; every mutation was instead confirmed on disk by anchor count before the reading was trusted. No heavy build/test ran, so scripts/pm/os-verify-lock.sh was not used and there is no VERDICT line to quote; inventing one would be worse than saying so. node_modules is absent in this container, which is why 10 of the 39 and 9 of the 42 could not be loaded by the import probe - stated in the PR table rather than papered over. GATES (verdict lines as the gates printed them, exit codes captured before any pipe): 'check-entry-guard self-test: 47 cases pass - all 11 measured spellings rejected, canonical form and masked prose/payloads accepted, and the import-safety rule recognised on both sides' (exit 0; was 26 cases, +21 all on the new rule) | 'check:entry-guard: 123 scripts/ file(s) - every entry guard goes through invoked-as.mjs; 81 export bindings, 46 of them inert on import (35 known-unsafe, SHRINK-ONLY)' (exit 0) | 'check-nul-bytes: OK (scanned 6211 text file(s) ... no raw ASCII control bytes)' (exit 0) | 'OK: 13 package(s) read outside themselves, all declared, and turbo.json hashes every declared glob.' (check:cross-package-test-inputs, exit 0) | 'OK check-ratchet-remedy-authority: 109 scripts swept; 7 mark the expanding remedy MAINTAINER-ONLY, 3 turn it down outright, 99 hand out no ratchet-expanding remedy' (exit 0) | check:parse-guard exit 0. check:single-claim-paths exits 2 here for want of PR_NUMBER/GITHUB_TOKEN - a CI-context refusal, not a result about this change. grep -naP over the changed file for raw control bytes: no match. UNION: node scripts/pm/dispatch-gates.mjs with NO paths, exit 0, derived from merge base 9dd192d48b: check:entry-guard and check:cross-package-test-inputs (plus the same via ci.yml). Cross-checked by hand rather than read as a clearance - check:ratchet-remedy-authority and check:parse-guard are implicated by what I wrote (a new shrink-only ledger; a new source scanner) and neither appears in the union, so both were run by hand and both are green. HARM REPRODUCED: importing scripts/check-nul-bytes.mjs for its exports printed its full 6211-file verdict and the probe's trailing ALIVE line NEVER PRINTED, with IMPORTER_EXIT=0 - silent, exit-0 death of the importer mid-import. check-engine-split-ratio.mjs exits 2 into the importer (#9757 still live in a second file). Controls: adr-anchors.mjs and regen-artifacts.mjs both printed ALIVE with zero output, and no import mutated the tree (git status --porcelain empty after). ABLATIONS (mutation confirmed on disk by anchor count in BOTH directions, each restored with git diff --quiet verified clean): (1) guard exemption removed -> predicted 'many', observed exactly 39 fresh findings = the guarded-clean exporters, plus 5 self-test cases red. (2) guard-alias support removed -> predicted 11, observed exactly 11, plus 3 self-test cases red. (3) exportsBindings forced true -> predicted 42, observed exactly 42 = precisely the non-exporting CLIs. (4) POSITIVE CONTROL, runsOnImport blinded to false -> predicted all 35 read STALE, observed exactly 35 stale: the gate cannot go green by going blind, which is the assertion a negative gate most needs. (5) one ledger line deleted -> 1 fresh. (6) a bogus ledger line -> 1 stale, named. TWO ABLATIONS WERE VOID ON THE FIRST ATTEMPT AND ARE REPORTED AS RE-RUNS, NOT FIRST RESULTS: the bogus-entry one passed a literal backslash-n into the source (syntax error - red for the wrong reason), and the ledger-deletion one used grep -F with an EMPTY pattern as its confirmation, which matches every line and printed 707. In that second case the removal half (1 -> 0) was valid and the gate output was correct; I re-ran it cleanly rather than accept a self-contradicting instrument. SHORTFALL THAT WAS A REAL FINDING: ablation (3) reddened ZERO on its first run. The cause was my own code - the export test was written twice, once in the detector and once in the census, so blinding one copy left the gate green. Collapsed to one exportsBindings definition in commit 4a83b1a3db; after that the ablation reddens 42. A rule enforced in one path and not the other is exactly the shape this card is about, one level down. BURN-DOWN LOOP DEMONSTRATED END TO END on check-nul-bytes.mjs and then reverted (git diff --quiet clean): guard added -> gate fails with '1 stale KNOWN_IMPORT_UNSAFE entry: scripts/check-nul-bytes.mjs'; line deleted -> green at '47 of them inert on import (34 known-unsafe)'; re-probed -> 'ALIVE - importer survived; exports: classify, findControlBytes, scan, scannedCharClass'; CLI unchanged -> 6211 files OK and --self-test 75 assertions.",
     "open_questions": [],
     "out_of_scope_findings": [
       "filed as #10667 (sub-issue of #10631): burn down the 35 scripts/ modules that run on import - the KNOWN_IMPORT_UNSAFE ledger, with the 4-line zero-re-indentation recipe and the ~5 files needing a real main() extraction",
       "NOT filed, reported here as it is inside #10667's scope: pm/check-governed-prose.mjs mutates process.argv at module scope to filter --self-test out before importing a sibling. That workaround is a scar from this exact class and looks stale now that pm/check-governed-merges.mjs carries the guard - flagged in #10667 rather than preserved",
       "FALSE IN THE CARD, option A: 'require the guard whenever the file exports AND has any top-level call expression' would flag essentially the whole tree - const HERE = resolve(fileURLToPath(import.meta.url), '..') is a top-level call in nearly every file here, including check-entry-guard.mjs itself. The workable form exempts declarations and asks only about statements that run",
       "FALSE IN THE CARD, option B: 'exports AND reads process.argv' has a false-positive side the card does not mention - eslint-stack-headroom.mjs exports, reads argv, and is completely inert on import, so B would demand a pointless empty guard on a pure library module",
       "FALSE IN THE CARD, option C: importing each candidate and asserting silence does NOT directly measure the property. It is blind to argv- and env-conditional leaks and reported qa/qa-rollup.mjs and objectui-range.mjs as clean; both demonstrably run inside an importer whose argv carries --self-test or --help",
       "BRIEF, minor: the GitHub search API is unavailable to this session ('sessions are bound to their configured repositories'), and the MCP search_issues tool returned 'API rate limit already exceeded'. The pre-filing duplicate search was done by paging repos/.../issues instead - worth knowing before the next seat is told to search first",
       "PLATFORM, not another actor: GitHub appended its own '_Generated by [Claude Code]_' block to the PR body on create, duplicating the footer I wrote. A PATCH that removed mine left the platform re-adding one in the BARE form. Content is fully intact; I stopped editing rather than churn the body"
     ]
   }

   ---

   Generated by Claude Code
7. os-zhuang

   removed

   finding

   on Aug 21, 2026
8. claude commented on Aug 21, 2026

   

   claudebot

   on Aug 21, 2026 – with Claude

   ContributorAuthor

   More actions

   Landed in #10665 (merged 2026-08-21T08:30:33Z). Closing as completed.

   That PR carried no Closes #10631 reference, so this card stayed open after its fix merged — closing it by hand and dropping pm:dispatched.

   ---

   Generated by Claude Code
9. claude

   closed this as completedon Aug 21, 2026
10. claude

    removed

    pm:dispatched

    on Aug 21, 2026