fix(messages): accumulate temporary tool declarations linearly - #6656

Summary

Temporary inline tool declarations with the same name repeatedly copy the declarations already collected. A chain of temporary additions therefore makes collection work grow quadratically.

Append to checker-owned declaration arrays instead. Permanent replacement still starts a new list; temporary declarations retain denied shadows, and removal, re-offer, anonymous declarations and caller immutability retain their existing behavior.

Add deterministic traversal, frozen-input and production-handler regressions to the existing scope suite, and update the owning protocol contract. Linear collection does not bound Advisor authorization-list lookups, total request memory or elapsed time.

Verification

Base: f92482065217bae891a88e27fd6c00012f212042.
Prepared candidate tree: 60d50c73f42b90314fd5675789b906373fbe4f7b.
Published head: 9389827a03f43a658622dd521a0b9dae936f2d35.

Current-head hosted result — checked 2026-10-06

Cross-platform CI 37412319289 and React Doctor 37412319254 completed successfully for this head. Candidate-head hosted CI is no longer pending. These are workflow conclusions, not a claim that every possible platform, skipped job, live-provider test or native application path ran.

A bounded CodeRabbit review was requested in the author follow-up. Review completion and required independent review remain outstanding. No new local test execution or approval is asserted by this metadata update.

Offline and reused evidence

Fresh offline checks on this candidate:

• Execute the production scope checker and unchanged upstream checker under Node 24.19.0 with TypeScript types stripped: 13 explicit semantic checks and 5,000 deterministic lifecycle comparisons pass with frozen inputs.
• Execute the two pure regression callbacks from the submitted test file unchanged: both pass.
• A fixture with 1,024 temporary additions counts values yielded by array iterators whose first element is the frozen seed declaration. This measures traversal of the seed input and matched collected/active declaration arrays. The original checker yields 526,851 entries; the candidate yields 2,051 against an allowance of 3,075. The original fails that allowance. This measures the repeated-prefix cost, not every checker operation, total authorization work, memory use or runtime. Advisor model/provider-list lookups remain unchanged.
• Run scripts/structure-ssot.ts, scripts/privacy-scan.ts and scripts/file-size-ratchet.ts with Node type stripping and an adapter for their read-only Bun.spawnSync calls: all pass.
• git diff --check <base-tree> <candidate-tree>: passes.

Reused validation on the source tree, with the same patch on an older base, reported 164 passing focused Bun tests across 11 files, plus TypeScript, privacy and structure checks. Inspected hosted CI associated with that source head has successful Linux test shards, typecheck, privacy and structure jobs. These results remain reused source evidence, separate from the current-head hosted results above. Windows/macOS test shards and docs build were skipped in that previously inspected source CI.

Bun and project dependencies are unavailable locally. The production-handler regression, complete Bun/import-connected suite and root TypeScript check were not rerun locally; no downloads, live provider calls, installed configuration or services were used for those offline checks. Current-head hosted execution is recorded separately above and does not erase the local coverage limits. Keep this PR in draft pending review and an accurate review-readiness attestation.

Checklist

• Scope stays focused and avoids unrelated cleanup.
• Docs or release notes were updated when needed.
• Security-sensitive changes were reviewed for secrets, auth, and unsafe defaults.

Review readiness checklist

• Required local validation passed with its scope documented.
• Branch is on the latest dev commit.
• All correct Codex and CodeRabbit findings are resolved.
• Ready for review.

Summary by CodeRabbit

• Performance
  • Temporary tool declarations now accumulate more efficiently, reducing repeated processing as declarations are added. Existing behavior for permanent replacements, denied declarations, and shadow checks remains unchanged.
• Tests
  • Added coverage for long chains of temporary declarations, denied declarations, and permanent replacements, including checks that input data remains unchanged.
• Documentation
  • Clarified how temporary declarations are accumulated and noted that this improvement does not guarantee request-size or processing-time limits.

Maintainer lane verification (sweep lane A, head 8d2c224)

• Merged origin/dev 38455fb into the branch (no force push); no source change.
• Independent gpt-6.1-sol review: PASS. The traversal regression distinguishes implementations (old checker 526,851 visits vs the 3,075 allowance); caller arrays and declaration objects stay untouched.
• bun test tests/claude-integration/messages-native-scope.test.ts tests/ci-workflows/file-size-ratchet.test.ts: 41 pass / 0 fail.
• bun run typecheck: exit 0. bun run structure:check: passed.
• The full local suite was not run because parallel lanes share this machine; remaining coverage is left to exact-head hosted CI.