fix(codemode): harden OpenAPI search and sandbox transport#23
Merged
Conversation
Rationale: Reuse acyclic component expansions across processed operations so large OpenAPI specs remain below the data-only graph guard while keeping endpoint-local cyclic expansion semantics.\n\nRisk: Cached schema objects are shared by identity for context-independent refs. Expansions containing circular or max-depth sentinels deliberately remain uncached to avoid reusing context-bound results.\n\nTested: mise exec -- pnpm --filter @robinbraemer/codemode exec vitest run test/spec.test.ts; mise exec -- pnpm --filter @robinbraemer/codemode run test; mise exec -- pnpm run lint
Rationale: A ref expansion cached from a shallow endpoint must not bypass a max-depth sentinel when that ref is reached later with less resolution budget. Risk: Cache reuse is now narrower because entries are partitioned by remaining depth, preserving correctness at the cost of duplicate expansions across different budgets. Tested: mise exec -- pnpm --filter @robinbraemer/codemode exec vitest run test/spec.test.ts; mise exec -- pnpm --filter @robinbraemer/codemode run test; mise exec -- pnpm run lint
Rationale: Cached processed specs now share acyclic ref objects, so every custom data executor must receive an independent graph per search while retaining aliases within that execution. Risk: Search now pays one structured clone per invocation; it deliberately preserves shared schema identity inside the executor input and avoids rebuilding the cached spec. Tested: mise exec -- pnpm --filter @robinbraemer/codemode exec vitest run test/codemode.test.ts; mise exec -- pnpm --filter @robinbraemer/codemode run test; mise exec -- pnpm run lint
Rationale: Keep graph-limit and accessor violations distinct from real function values so executor errors accurately explain why a data-only input was rejected. Risk: Only error classification changes for accessor and oversized graphs; findFunctionPath preserves its existing compatibility strings for internal non-null checks. Tested: mise exec -- pnpm --filter @robinbraemer/codemode exec vitest run test/data-only.test.ts; mise exec -- pnpm --filter @robinbraemer/codemode test; mise exec -- task lint.
Rationale: Preserve the legacy findFunctionPath strings used by executor compatibility checks while rejectDataOnlyFunctions formats classified errors. Risk: Test-only change; it locks the existing wrapper messages for accessor and graph-limit violations. Tested: mise exec -- pnpm --filter @robinbraemer/codemode exec vitest run test/data-only.test.ts; mise exec -- pnpm --filter @robinbraemer/codemode test; mise exec -- task lint.
Rationale: Concurrent first searches must share one async spec-provider resolution and processing pass while retaining isolated per-search inputs. Risk: Failed provider resolutions must not poison later searches; the in-flight memo is identity-cleared on either outcome. Tested: pnpm --dir packages/codemode exec vitest run test/codemode.test.ts; mise exec -- task ci
Rationale: Shared OpenAPI ref expansions intentionally form a DAG, but built-in JSON and QuickJS transports recreate each alias occurrence. Bound the expanded representation before marshalling so a small unique-object graph cannot exhaust host memory while preserving the production-scale pre-resolved spec. Rejected: Preserving aliases inside every transport requires a larger cross-runtime protocol change and would not protect existing JSON backends. Risk: The new 500k-occurrence and 10 MiB ceilings reject unusually large data-only inputs, cycles, bigint, and custom toJSON hooks before execution. The measured production fixture remains below both limits. Tested: mise exec -- task ci (CodeMode 167 passed, 9 skipped; LLRT 27 passed, 22 skipped; lint, typechecks, and builds passed). Not-tested: Native LLRT runtime contracts were skipped because upstream native preparation failed during yarn install; TypeScript/build integration passed.
robinbraemer
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Intent
Preserve a fully pre-resolved OpenAPI schema for AI CodeMode search while safely reusing reference expansions, bounding alias-expanded transport payloads, keeping circular/max-depth behavior correct, accurately classifying violations, and publishing 0.4.1 for CNAP production.
What changed
Risk
Low after automated remediation and re-review. The production-scale 4.35 MiB resolved graph remains accepted; oversized or unsafe inputs fail before transport.
Verification
Luna no-mistakes passed. Native LLRT, IsolatedVM, QuickJS, LLRT-process, production-scale search, adversarial transport checks, lint, documentation, publication metadata, and PR CI are green.