Commit Graph

4 Commits

Author SHA1 Message Date
Bo f7cc505a7a fix(agentopsd): harden AI smell surfaces (soc-0l5z #ai-smell-hardening) (#389)
## Summary
- Replace the plans manifest shape-only stub with projection-backed
daemon state.
- Gate llmwiki placeholder outputs behind explicit opt-in and make fake
daemon execution test/demo-only.
- Refresh stale Olympus-ported docs and tune README/PRODUCT claims to
default-safe runtime behavior.
- Repair strict docs-site and hook-lease inventory drift surfaced by the
merge gate.

## Validation
- `scripts/pre-push-gate.sh --fast`
- `go test ./internal/daemon ./internal/llmwiki ./cmd/ao`
- `bash tests/docs/validate-doc-release.sh`
- `scripts/docs-build.sh --check`
- `bash scripts/validate-sovereignty-proof-citations.sh`
- `bash scripts/check-hook-lease-inventory.sh`

Closes-scenario: soc-0l5z#ai-smell-hardening
Bounded-context: BC5-Runtime
Evidence: validate-sovereignty-proof-citations: scanned 3 pages, 16
citations
2026-05-20 21:31:38 -04:00
Boden Fuller 28da44fc5d feat(daemon): wire skill schedules 2026-05-09 09:34:23 -04:00
Boden Fuller 0b2d979c82 feat(schedules): recipe pack + scheduling guide (epic soc-sx99 W2.1)
Surface ao schedule + ao daemon as primary product feature. The scheduler
exists; recipes don't. This adds the always-on lane that the user actually
runs as flagship usage:

- examples/schedules/dream-nightly.yaml      — dream.run @ 23:00
- examples/schedules/forge-daily.yaml        — wiki.forge @ 22:00
- examples/schedules/compile-weekly.yaml     — compile @ Sun 02:00
- examples/schedules/defrag-weekly.yaml      — compile --defrag-only @ Sun 03:00
- examples/schedules/evolve-ondemand.yaml    — manual fire on GOALS regression
- docs/scheduling.md                         — operator guide (catalog +
  custom-schedule path + run inspection via ao watch / ao schedule list)

All 5 recipes validated through cli/internal/schedule.Load() (the same path
ao schedule add invokes before POSTing to the daemon). The 6th recipe,
feedback-drain-hourly.yaml, lands in the soc-sx99.6 commit since it depends
on the --drain flag introduced there.

Closes soc-sx99.4.
2026-05-07 12:03:22 -04:00
Boden Fuller c84e190c2d feat(cli): ao feedback-loop --drain — clear unfed citation backlog (epic soc-sx99 W3.1)
Substrate audit found 1,508/3,735 (40%) citations with feedback_at zero-sentinel:
the close-loop fires but a large fraction of citation events never produce a
utility update. --drain walks .agents/ao/citations.jsonl, applies a default
reward + α-update via UpdateLearningUtility, and clears the sentinel.

- --drain + --drain-reward + --dry-run flags
- Idempotent: re-runs only drain still-unfed entries
- Real-corpus check on this repo: 2,053/3,828 (53.6%) unfed → confirms bug
- L2 integration tests: seed → drain → assert utility moved + feedback_at
  populated; idempotency; dry-run no-mutation; skip-already-fed
- Health check G in scripts/check-memrl-health.sh fails when sentinel rate >5%
- examples/schedules/feedback-drain-hourly.yaml recipe wired in

Closes soc-sx99.6.
2026-05-07 12:03:22 -04:00