feat(analytics): durable adoption analytics, diagnosing-analytics skill, hook replay test, Claude schema validation

[ScriptedAlchemy]

Why

Investigating "is TraceDecay actually being used?" surfaced four structural problems:

1. analytics_events looked dead but wasn't. The only writer targets the user-level global.db; the per-project sessions.db carries the same schema with no writer, so anyone inspecting the project store concludes telemetry is broken.
2. Hook telemetry splits across two JSONL files with no attribution. Hooks write to the project store when they can resolve a project root and silently fall back to ~/.tracedecay/hook_analytics.jsonl when they can't (Claude preToolUse and all Kiro hooks hardcoded None). Rows carried no project field, so the split streams could never be re-joined.
3. Dashboard diagnostics read only the project-store file, so the active user-level stream was invisible.
4. No CLI answered adoption questions — tracedecay analytics … didn't exist, and nothing taught agents which surface to use.

What

• Writer attribution — hook analytics rows now carry project_root/event_cwd; Claude preToolUse resolves the root from the event cwd then the process cwd, and the three Kiro hooks resolve from the event cwd.
• Merged reader — dashboard diagnostics reads both hook files (user-level rows filtered by attribution) and reports a hook_sources list with per-file row counts.
• analytics_bridge — idempotent hook-JSONL → analytics_events import using parse_offsets byte cursors; partial trailing lines stay unconsumed for the next run.
• New CLI — tracedecay analytics diagnostics [--all|--no-sync] (imports, then prints the same summary the dashboard serves) and tracedecay analytics sync.
• New skill — diagnosing-analytics teaches agents the analytics/doctor/lcm_status/sessions surface and its guardrails (never raw sqlite); added to CURSOR_PLUGIN_SKILLS.
• Hook replay E2E test — drives all four providers' hook subcommands through the real binary with representative events, asserting attributed JSONL rows land in the project store, nothing spills to the fallback file, and analytics sync bridges every row into analytics_events.
• Claude schema validation — vendored draft-07 schemas for Claude Code hooks.json, plugin.json, and marketplace.json with positive + negative tests, closing the gap where only Cursor/Codex configs were schema-checked. docs/PLUGIN-VALIDATION.md table updated.

Verification

• New tests: hook replay E2E (12 provider events, wiring end to end), bridge import integration test (incremental/idempotent/partial-line), 3 bridge unit tests, 4 Claude schema tests.
• Suites: hooks_lsp_suite 117/117, agent_suite skill/schema subset 43/43, dashboard analytics 6/6, session_suite import test — all green on this branch.
• Live run on real data imported 38,686 backlogged hook rows into analytics_events and the diagnostics command now answers adoption questions in one shot.

🤖 Generated with Claude Code

[changeset-bot]

⚠️ No Changeset found

Latest commit: e9b75f3

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.