feat(overview): add /api/autonomy-score endpoint (#688)

[vivekchand]

Summary

Adds GET /api/autonomy-score — the north-star autonomy metric from issue #688. The endpoint scans the last 7 days of session JSONL files (already on disk) and computes how far apart human nudges are, what fraction of sessions ran without any follow-up intervention, and whether autonomy is trending in the right direction. No new dependencies, no schema changes, no UI touched.

Changes

• routes/overview.py — new _compute_autonomy_data(sessions_dir, cutoff_ts) helper (pure function, testable without a server) + GET /api/autonomy-score endpoint that calls it, following the same pattern as api_main_activity() and api_timeline().
• tests/test_autonomy_score.py — 18 unit tests covering: missing/empty dirs, .deleted./.reset. file exclusion, zero-nudge sessions, gap computation, median correctness, 7-day cutoff window, malformed JSON resilience, custom cutoff, and full return-shape invariants. All pass without a live server.

Response shape

{
  "median_seconds_between_nudges": 1843.5,
  "zero_nudge_ratio": 0.4,
  "trend_slope_7d": 120.3,
  "sessions_analyzed": 10,
  "zero_nudge_sessions": 4,
  "daily_medians": [
    {"day_offset": 0, "median_gap_seconds": 2100.0},
    {"day_offset": 1, "median_gap_seconds": 1800.0}
  ]
}

trend_slope_7d is the OLS slope of daily median nudge gaps over 7 days — positive means the agent is handling more work between human check-ins.

Test plan

• python3 -m pytest tests/test_autonomy_score.py -v → 18/18 pass (no server needed)
• python3 -c 'import ast; ast.parse(open("routes/overview.py").read())' → clean
• curl http://localhost:8900/api/autonomy-score returns the expected shape

Bot meta

Draft PR opened autonomously based on the plan in #688. Marked draft for human review — mark Ready for Review once happy.

Closes #688

---

Generated by Claude Code

[vivekchand]

Heads-up — main now has routes/autonomy.py with GET /api/autonomy that already serves issue #688's north-star autonomy score (Krentsel's nudge-spacing definition: score, trend_slope_7d, trend_direction, samples_7d, series_daily). This PR adds a parallel GET /api/autonomy-score with raw metrics (median_seconds_between_nudges, zero_nudge_ratio, daily_medians). Up to you whether to merge as complementary raw-metric endpoint or close as duplicative — leaving for your call. CI is green, mergeable.