[Patrick-Erichsen]

Summary

The unchanged gtrk-ai-drama@1.3.9 validator is left partially inspected because bounded native arrays, record values, function results and template joins lack a closed ownership proof. Add an optional JavaScript value/effect proof, leaving unknown calls, grammar and effects unsupported.

constructed value → prove private storage, primitive inputs and native identity
                 → prove every source effect within the existing budget
                 → admit the bounded value only when both proofs hold

The embedded patch and mirrored Docker adapter suites preserve legacy inspection, findings, parser/type budgets and dependency pins. Native argv.slice requires scalar start/end operands, preventing implicit coercion hooks from invalidating later identities.

Evidence

• Before (production c073): unchanged gtrk original fully inspected2/3 files, three generated AE1 incomplete-reference rows.
  **After (installed Linux package):**3/3 fully inspected, those three generated rows disappear; all six missing-reference caveats remain. Validator SHA-256 f099cc9b6613a7d61758b86e9d571e2934ae602e38243117fa235e4389e921e1 is unchanged. No original code was executed.
• Before (rejected a920 candidate): five offset-hook controls incorrectly completed, including a separate later console argument and argv receiver alias.
  After: all five remain partial; four scalar-bound controls remain complete. All three exact independently reported examples reproduce before and stay partial after the repair.
• 752 focused source controls;1,101 installed adapter controls;300 executable-construction negatives and12 namespace mutations pass. Normal embedded-patch application reconstructs all137 source/dependency hashes; uv.lock is unchanged. Full Go tests, go vet ./... and CLI help pass.
• All 30 retained original inventories match. The other29 full raw finding payloads (except generated IDs) and coverage counts match. Exactly one Feishu file refines static_parse_limit to opaque_content while staying partial: lexical lookup memoization preserves22 unsupported constructions under the same100,000-step per-proof budget. The strict comparator failure is retained; a narrow verifier accepts only that traced diagnostic refinement.
• Two additional xiapan originals remain3/9fully,6partial,45findings with identical findings/completeness before and after. This change has no demonstrated benefit for their unresolved module inspection.
• Independent P0–P3 review: all three cumulative source passes completed with the same argv-coercion cause; all three literal examples are fixed, and the focused a920→826 repair review is scoped-clean. Cumulative consumer c073→6481 transport review and focused6481→ca116 integration review are scoped-clean. The historical a920 review retains its three accepted findings and incorrect verdict. Exact final source/reconstruction identities bind the combined scope.

These are local/static proofs. Production adoption, ordinary original recovery and public/install verification follow separately. This PR does not resolve every case in openclaw/clawhub#4058.

Merge Danger

Door: two-way

The source pin can be rolled back. This does not repack original uploads or alter publication/moderation state.

Blast Radius: scanner

Incorrect ownership could hide an inspection gap; affirmative effect/native-input checks, adversarial controls and independent semantic review protect that boundary.

[clawsweeper]

🦞👀
ClawSweeper picked this up.

Pull request received. I will update this pull request when review starts.