[#279 Phase 1] Pull-based meeting ingestion — sync-and-brief CLI + Granola adapter + SessionStart hook by Knapp-Kevin · Pull Request #318 · BicameralAI/bicameral-mcp

Knapp-Kevin · 2026-05-14T16:21:26Z

Summary

Closes #279 Phase 1 (solo-mode session magic, independent of #277). Delivers the v0 Productization §3 commitment that briefs + drift scans are pre-baked OUTSIDE the agent before Claude sees the first prompt.

Five new surfaces:

`bicameral-mcp sync-and-brief` CLI subcommand — server.py argparse + `cli/sync_and_brief_cli.py`.
SessionStart hook in `setup_wizard.py` (cross-platform POSIX + Windows). Always ends with `exit 0` so it can NEVER block session start. Stderr redirects to `~/.bicameral/hook-errors.log`.
Granola source adapter at `events/sources/granola.py`. Watermark-driven; two-phase commit so a failed ingest doesn't advance the watermark; stdlib `urllib.request` for HTTP (no new dep).
Brief synthesizer (`cli/brief_renderer.py`). Pure function; ≤200 line cap; per-field length caps; respects `signer_email_fallback` from config.
skill + docs — `skills/bicameral-sync-and-brief/SKILL.md` + `docs/policies/sources-config.md` documenting the `sources:` config schema and the `api_key_env` convention.

Prompt-injection isolation

The brief is injected into Claude's pre-session context via the SessionStart hook — that's the explicit goal. So user-sourced content (decision summaries, source refs, drift evidence) becomes a potential prompt-injection vector. Three layers of defense, all pinned by tests:

Block-quote data-framing preamble at the top of every brief.
Triple-backtick code fences around every user-sourced field. A line containing `IGNORE PRIOR INSTRUCTIONS` is presented as fenced data, not as flowing prose.
Fence-break neutralisation: embedded ``` runs in user text are interrupted with a zero-width space so payloads can't break out of their fence.

Audit Pass 1 → Pass 2 deviations (full transparency)

The audit VETOed pass 1 on three findings, all resolved in this PR:

#	Finding	Resolution
1	Prompt-injection on hook envelope	Discipline #6 added; fences + preamble + neutralisation; test pinned
2	Infrastructure: `get_recent_decisions` / `get_unresolved_gaps` do not exist	Use `get_all_decisions("all")` + client-side cap; drop gaps section from Phase 1 (future cycle adds it back)
3	`httpx` dependency unjustified	Use stdlib `urllib.request`; no new dep
(advisory)	`main()` projected >40 LOC	Extracted `_run_source` + `_synthesize_brief` helpers

Drift surface uses `bicameral.preflight` (existing) rather than the `bicameral.scan_branch` referenced in the issue text — `handlers/scan_branch.py` does not exist in main; preflight IS the implemented drift surface today. Flagged in plan Open Question 1.

Test plan

`python -m pytest tests/test_sources_granola_unit.py -v` — 12 tests
`python -m pytest tests/test_brief_renderer.py -v` — 11 tests (incl. prompt-injection canary)
`python -m pytest tests/test_sync_and_brief_cli.py -v` — 11 tests (1 skipped on missing mcp[cli] in test env; passes in CI)
`python -m pytest tests/test_sessionstart_hook_install.py -v` — 8 tests
Manual: `pip install -e .[test]` then `bicameral-mcp sync-and-brief --help`; with a Granola API key + a meeting from today, verify the brief renders correctly and `granola.json` watermark advances only after successful ingest.

Phase 2 (deferred)

Per #279 scope, Phase 2 (team-mode integration via `BackendAdapter.pull_events()` / `push_events()`) was gated on #277. #277 is now CLOSED (merged) so Phase 2 is unblocked — it's a separate follow-up cycle, not part of this PR.

🤖 Generated with Claude Code

…brief) Closes the v0 Productization §3 solo-mode commitment: a SessionStart hook runs `bicameral-mcp sync-and-brief` synchronously, the CLI pulls from configured sources, ingests anything new, runs preflight for drift, and prints a synthesized brief — all OUTSIDE the agent, before Claude sees the prompt. Phase 1 (solo mode, independent of #277): - `bicameral-mcp sync-and-brief` CLI subcommand (server.py argparse + cli/sync_and_brief_cli.py) - SessionStart hook installed by setup_wizard; cross-platform (POSIX + Windows shapes), always ends with `exit 0` so it can NEVER block session start - Granola source adapter (events/sources/granola.py) — first integration per the issue. Watermark-driven, two-phase-commit semantics so a failed ingest doesn't advance the watermark - Brief synthesis (cli/brief_renderer.py): pure function, ≤200 line cap, per-field length caps, signer_email_fallback policy respected - skills/bicameral-sync-and-brief/SKILL.md + docs/policies/sources-config.md Prompt-injection isolation (audit Pass 1 finding 1): - Brief begins with block-quote data-framing preamble - Every user-sourced value (decision summary, source_ref, drift_evidence) rendered inside triple-backtick code fences - Embedded ``` runs in user text are neutralised so payloads can't break out of their fence - Pinned by tests/test_brief_renderer.py:: test_brief_renderer_wraps_user_text_in_code_fences (and 10 others) Plan deviations from issue text (audit Pass 1 findings 2-3 + Razor advisory): - Drift surface uses `bicameral.preflight` (existing) rather than `bicameral.scan_branch` (no handler in main codebase) - Decisions fetched via `get_all_decisions(filter="all")` + client-side cap; `get_recent_decisions(N)` did not exist - Gaps section dropped from Phase 1; `get_unresolved_gaps()` did not exist. Future cycle adds the query helper + the brief section. - Granola HTTP layer uses stdlib `urllib.request` (no new dependency) via a `GranolaClient` indirection that lets tests mock transport - `cli/sync_and_brief_cli.main()` extracted into `_run_source` + `_synthesize_brief` helpers to keep each function ≤ 30 LOC Security: - API key handling: config holds `api_key_env` name, NEVER the key itself. Adapter reads from `os.environ` at pull time. - Watermarks at `~/.bicameral/source-watermarks/<source>.json` — outside repo, outside git. Cannot be checked in. - Hook command `exit 0` framing AND stderr-to-log-file redirect. 41 tests pass (12 Granola adapter + 11 brief renderer + 11 CLI + 8 hook install). The 1 skipped test is the cli_main smoke test which needs mcp[cli] installed in the test env — passes in CI. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

coderabbitai · 2026-05-14T16:21:34Z

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: b8864940-4778-47c7-82a9-d83f482986c7

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

Create PR with unit tests
Commit unit tests in branch plan/279-phase1-sync-and-brief

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

- Remove unused typing.Any from events/sources/granola.py - Sort imports in tests/test_brief_renderer.py - Apply ruff format to all 7 new/touched files 41 tests still pass. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…+ granola max() type narrowing - cli/sync_and_brief_cli.py: handle_preflight requires positional 'topic'; pass 'session-start-brief' as the sync-and-brief sentinel. - events/sources/granola.py: coerce ended_at_values to list[str] so mypy doesn't see Any|None passed to max(). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Wires the existing BackendAdapter interface (shipped in #277, closed) into the sync-and-brief CLI flow. Phase 2 of #279 — the team-mode complement to Phase 1's solo-mode session magic (PR #318, merged). Order invariants: 1. BEFORE source pull: backend.pull_events() copies every peer's <email>.jsonl into the local .bicameral/events/ cache. The materializer's existing *.jsonl glob (events/materializer.py:73) picks them up alongside the operator's own events. Peer ingest is visible in the brief without manual action. 2. AFTER source ingest succeeds: backend.push_events() uploads each local <email>.jsonl to the shared backend. LocalFolderAdapter's sha-match skip (events/backends/local_folder.py:44-46) keeps re-invocations idempotent. Failure modes are non-blocking — backend pull/push errors log to stderr + ~/.bicameral/cli-errors.log but never block the brief. The hook wrapper's exit 0 framing makes this invisible to SessionStart users on a network outage. Solo mode (no team: config) is completely unaffected — same code path as before. The brief renderer gains an optional ## Team sync section showing peer_files_pulled + my_file_pushed. Omitted entirely when team_sync=None so solo-mode briefs render byte-identically to pre-Phase-2. Per #205 doctrine — the only operator-facing default in this new path (no team backend when team: config is absent) is a deterministic gate at events/backends/__init__.py::get_backend, not a skill-text-only claim. Backend errors return None safely. 12 tests in tests/test_sync_and_brief_team_mode.py — sociable per CLAUDE.md (real LocalFolderAdapter, real BicameralContext, real file system; no backend mocks). The headline two-machine round-trip test spans Alice's repo → shared remote_root → Bob's repo and verifies Alice's events land in Bob's cache after his next sync-and-brief. 3 new tests in tests/test_brief_renderer.py covering the Team sync section (present/absent/zero-counts). Full Phase 1 + Phase 2 sweep: 56 passed, 1 skipped (mcp[cli] env probe unchanged). docs/policies/sources-config.md extended with a "Team backend" section documenting config schema + failure modes. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Knapp-Kevin had a problem deploying to recording-approval May 14, 2026 16:21 — with GitHub Actions Failure

Knapp-Kevin temporarily deployed to ci-test May 14, 2026 16:21 — with GitHub Actions Inactive

Knapp-Kevin had a problem deploying to production May 14, 2026 16:21 — with GitHub Actions Failure

Knapp-Kevin temporarily deployed to ci-test May 14, 2026 16:21 — with GitHub Actions Inactive

Knapp-Kevin had a problem deploying to production May 14, 2026 16:36 — with GitHub Actions Failure

Knapp-Kevin temporarily deployed to ci-test May 14, 2026 16:36 — with GitHub Actions Inactive

Knapp-Kevin had a problem deploying to recording-approval May 14, 2026 16:36 — with GitHub Actions Failure

Knapp-Kevin temporarily deployed to ci-test May 14, 2026 16:36 — with GitHub Actions Inactive

Knapp-Kevin had a problem deploying to recording-approval May 14, 2026 16:42 — with GitHub Actions Failure

Knapp-Kevin temporarily deployed to ci-test May 14, 2026 16:42 — with GitHub Actions Inactive

Knapp-Kevin had a problem deploying to production May 14, 2026 16:42 — with GitHub Actions Failure

style: ruff format events/sources/granola.py (#279 follow-up)

edb4128

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Knapp-Kevin temporarily deployed to production May 14, 2026 16:48 — with GitHub Actions Inactive

Knapp-Kevin had a problem deploying to recording-approval May 14, 2026 16:48 — with GitHub Actions Failure

Knapp-Kevin temporarily deployed to ci-test May 14, 2026 16:48 — with GitHub Actions Inactive

Knapp-Kevin merged commit 0593a30 into dev May 14, 2026
6 of 7 checks passed

Knapp-Kevin mentioned this pull request May 14, 2026

[v0-productization §3] Pull-based meeting ingestion — sync-and-brief CLI + SessionStart hook + source pullers #279

Closed

7 tasks

Knapp-Kevin mentioned this pull request May 14, 2026

feat(team-sync): #279 Phase 2 — team-mode integration for sync-and-brief #321

Merged

4 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[#279 Phase 1] Pull-based meeting ingestion — sync-and-brief CLI + Granola adapter + SessionStart hook#318

[#279 Phase 1] Pull-based meeting ingestion — sync-and-brief CLI + Granola adapter + SessionStart hook#318
Knapp-Kevin merged 4 commits into
devfrom
plan/279-phase1-sync-and-brief

Knapp-Kevin commented May 14, 2026

Uh oh!

coderabbitai Bot commented May 14, 2026 •

edited

Loading

Review skipped

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

Knapp-Kevin commented May 14, 2026

Summary

Prompt-injection isolation

Audit Pass 1 → Pass 2 deviations (full transparency)

Test plan

Phase 2 (deferred)

Uh oh!

coderabbitai Bot commented May 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Review skipped

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

coderabbitai Bot commented May 14, 2026 •

edited

Loading