Add Copilot CLI V1 capture with hooks, sync, watch, and status

[duncankmckinnon]

Summary

GitHub Copilot CLI recordings are ingested into a local schema-v1 archive of transcript, SQLite, hook, and metadata source records so later V2 projections can run without recapture. thirdeye add --copilot and remove --copilot install or drop only the owned user-level hook file; thirdeye copilot sync, watch, and status resolve --source-home, then COPILOT_HOME, then ~/.copilot, and never export Copilot content. Captured sessions appear in existing list/show/events/search/tail and generic UI event views as raw envelopes, not reconstructed turns or UsageStore rows.

Changes

• Add capture contracts in src/thirdeye/platforms/copilot/{types,constants,identity}.py: schema version 1 TypedDicts, invocation-time source-home resolution, SHA-256 source_key, and stored IDs copilot-<source_key[:16]>-<native_id> with native-ID and path-traversal rejection.
• Read session-state/**/events.jsonl and workspace.yaml (transcript.py) and allowlisted session-store.db tables (database.py: sessions, turns, assistant_usage_events) as bounded SourceSlices with locators, incomplete-line waiting, and read-only WAL-aware SQLite (busy timeout 100ms).
• Persist composed SourceBatches through a per-session lock/journal (archive.py, state.py) that recovers after crash, deduplicates committed source IDs, and writes copilot_transcript/copilot_database/copilot_hook/copilot_metadata events; SessionWriter.append accepts a validated source timestamp; fsops.sync_directory makes journal publish durable.
• Parse hook stdin into SourceRecords and spool them atomically before taking the capture lock (hook_payload.py, spool.py); install/remove only $COPILOT_HOME/hooks/thirdeye.json (install.py); register the thirdeye-copilot-hook dispatcher.
• Compose discovery and capture (sources.py, capture.py); make hook receipt fail-open and nonblocking with a bounded follow-up worker (hooks.py, followup.py); add foreground poll-and-ingest (watch.py) and health reporting (status.py).
• Register thirdeye copilot sync|watch|status, add/remove --copilot, and Copilot as a setup tracing platform; treat Copilot camelCase hook names as native in provenance.py.
• Document V1 commands and Windows caveats; include Copilot CLI 1.0.83 fixtures plus labeled synthetic V1 cases.
• Add behavioral coverage for contracts, readers, archive recovery, hooks, composition, watch/status, CLI registration, and generic UI views.

Test plan

• uv run python -m pytest tests/test_copilot_*.py tests/test_add_command.py tests/test_setup_command.py tests/test_provenance.py tests/web/test_copilot_capture_views.py
• thirdeye add --copilot writes $COPILOT_HOME/hooks/thirdeye.json (default ~/.copilot/hooks/thirdeye.json) with a resolvable thirdeye-copilot-hook command; thirdeye remove --copilot removes only owned entries and leaves captured sessions and other hooks in place
• Point --source-home at a home populated from tests/fixtures/copilot/cli-1.0.83 and run thirdeye copilot sync: counts print, prompt bodies stay out of diagnostics, a missing --session-id exits nonzero, and empty discovery is a successful no-op
• Confirm hook receipt spools before capture, follow-up is nonblocking, and thirdeye copilot status reports resolved paths, readable capabilities, configured hooks, last observed hook, last successful import, and pending follow-up; missing hooks are informational while source errors affect exit status
• thirdeye copilot watch --source-home PATH --interval 1 performs an initial sync, picks up SQLite-only updates, rejects intervals below 0.1s, and exits cleanly on SIGINT with no remote export
• Captured sessions show in thirdeye list/show/events/search/tail and generic UI event views as copilot_transcript/copilot_database/copilot_hook/copilot_metadata envelopes with schema_version=1, without turn-sliced views or UsageStore rows

🤖 Generated by workbench pr_writer