[Stephen Belanger (Qard)]

Muse Code supports native tracing across live capture, historical import, managed runs, and attach. Live hooks and exports share deterministic model/tool IDs; detaching leaves native work open, and late completion enriches the same spans without allowing older root snapshots to replace newer answers.

• Support all 17 native hook events, tool success/failure and batch results, permission observations, child lifecycle, compaction, and failed turns. Preserve native timing, response metadata, final output, and counted-once token totals.
• Use muse-code-sdk = 1.4.12 for MSP initialization, stdio framing, request correlation, and concrete session-list parameters/results. The client lives in bt-daemon/src/muse/msp.rs; subprocess creation retains the repository's Windows-safe helper, bounded diagnostics, timeouts, and cancellation cleanup.
• Keep hook/export wire types, messages, outcomes, token DTOs, authored row metadata, and generated configuration in focused modules under bt-daemon/src/muse. Decode hooks once into a typed event enum; stream export records through that same translator. Reducer state stays in translate/muse*.rs.
• Preserve explicit null versus absent completion metadata for replay fingerprints. Unknown native records are skipped; malformed known events have contextual diagnostics. Resource samples never become token counters. JSON remains at transport/row boundaries, open provider/tool leaves, and foreign configuration preservation.
• Preserve enterprise/user hooks, migrate owned legacy configurations, and isolate managed-run routes. muse/config/settings.rs, hooks.rs, and ownership.rs type existing settings, hook collections, command handlers, and strict owned-file recognition. Setup orchestration contains no dynamic JSON navigation. Missing/null distinctions and unknown extension data survive edits; only exact owned files/handlers are eligible for migration/removal. Typed settings use the shared atomic writer under Muse’s native lock.
• Declare Rust 1.97 as the daemon minimum and pin development/CI to 1.97.1, matching the SDK requirement. bt already uses 1.97.1.

Validation:

• cargo test --manifest-path bt-daemon/Cargo.toml --all-features --locked: 622 passed, 9 opt-in tests ignored.
• Strict all-feature/all-target Clippy, formatting, locked metadata, and diff checks pass.
• All four opt-in Muse tests pass on macOS against Muse Code 1.4.11-R6667.1 with deterministic loopback inference/ingest: tool/retry/failure tracing; active attach/detach/reattach plus SDK-backed import muse --all; and both managed-hook coexistence/sanitized-environment tests.
• Native tests use Muse's file credential backend and disable telemetry with temporary homes. Linux native tests and Linux/macOS/Windows Rust checks run in CI; CI for the latest revision is reported separately from local validation.

Native fidelity limits remain explicit: hook messages/output are previews; permission hooks do not report decisions; compaction lacks operation IDs. Observed built-in reminder child hook/export metadata omits parent identity, so ancestry is not guessed across concurrent sessions. Legacy exports without model_step cannot guarantee live/export model deduplication. Muse's official installer supports Linux/macOS; Windows runs Rust fixture tests only.

Distribution: hooks ship through bt. After merge, its existing update-bt-daemon.yml workflow prepares the consumer pin/lockfile PR; publication follows the bt release process.

Review follow-up: consistently resolve BRAINTRUST_MUSE_BIN through the environment helper; prepare fresh explicit daemon endpoints before sanitized hooks run; supply host information from the standalone entry point; and reject empty one-shot imports while allowing empty attach polls and preserving nonempty active sessions. Regressions cover standalone/custom-socket startup and empty imports; native managed-run verification also checks real hook delivery at a fresh explicit endpoint.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-23T14:56:21.790014Z	6be6464	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "Codex (@codex) review" or "Codex (@codex) security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6be646487f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "Codex (@codex) review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "Codex (@codex) address that feedback".