Harden for production: surface failures, schema versioning, integration tests, MCP security#2
Merged
Merged
Conversation
…suite Toward production-readiness (P0): - miner: stop silently swallowing embed/store failures. MineStats gains an `errors` count; failed chunks log which file/chunk failed and `mine` reports "N chunk(s) were NOT remembered". You now know when content didn't persist. - db: add SCHEMA_VERSION + `PRAGMA user_version` with a migrateSchema() shim, so future schema changes can migrate instead of silently breaking a palace, and a palace from a newer build is flagged rather than corrupted. - scripts/integration-test.sh: end-to-end suite exercising the real binary — directory mining, semantic ranking (paraphrase queries), the MCP protocol (store -> recall), the SessionStart/PreCompact hooks, and schema versioning. Wired into build.yml CI so the hand-verified behavior can't silently regress. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A security audit found that an LLM agent's own tools/call arguments could trigger silent memory corruption in the ReleaseFast production build: - limit: clamp to [1,100] — out-of-i32 / negative values were UB via @intcast (and overflowed the searcher's `limit * 2`). - name: type-switch instead of blind `.string` — a non-string name was an inactive-union access (UB / out-of-bounds read). - getInt float path: guard @intFromFloat against NaN / out-of-range (UB). - initialize protocolVersion: validate it's a safe version token before reflecting it into hand-built JSON (was a response-injection vector). Added an adversarial-input regression test to the integration suite. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
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.
P0 hardening toward professional use, on top of a security audit.
Robustness: miner surfaces embed/store failures (counts + per-chunk logs + a 'NOT remembered' warning) instead of swallowing them; schema versioning via
PRAGMA user_version+ a migration shim so future schema changes don't silently break existing palaces.Tests:
scripts/integration-test.sh(14 assertions) exercises the real binary — directory mining, semantic ranking, MCP store→recall, SessionStart/PreCompact hooks, schema version, and adversarial-input survival — wired into CI.Security (audit-driven): an LLM agent's own tools/call args could cause silent memory corruption in the ReleaseFast build. Fixed: clamp
limitto [1,100], type-switchnameinstead of blind.string, guard@intFromFloat, and validateprotocolVersionbefore reflecting it (response-injection).🤖 Generated with Claude Code