Loop Presets

Eight recurring scans you enable per project in ai/loop.yaml. Each runs on its own schedule, carries a budget cap the runner enforces, and emits strict JSON so the verdict parser can reject malformed output. Read-only by default — only custom may apply changes, and only when allow_actions: true is set explicitly.

Enable a preset

# ai/loop.yaml
loops:
  ci-watcher:
    enabled: true      # picks up the built-in schedule + budget
  drift-scanner:
    enabled: true
    notify: tg         # override: tg, slack, email, stdout
  custom:
    enabled: true
    schedule: "*/45 * * * *"
    budget_usd: 0.10
    prompt: |
      Check npm audit once; emit a JSON summary of high-severity findings.

The eight presets

pr-health

pro

Flag stuck PRs, missing tests, or CI blocks. Read-only — never comments or merges on your behalf.

schedule
every 30 min
cron
*/30 * * * *
budget
$0.20
pr_health.md →

ci-watcher

pro

Detect main-branch CI failures. Auto-retry flakes (single step, network signature), flag real breaks.

schedule
every 15 min
cron
*/15 * * * *
budget
$0.10
ci_watcher.md →

standup-digest

free

Morning briefing: shipped, stuck, cost, reviews waiting, next action. Included on every tier.

schedule
08:00 daily
cron
0 8 * * *
budget
$0.15
standup_digest.md →

dep-slopsquat

team

Scan new deps for typo-squat edit distance, LLM-hallucinated names, or owner-change signals.

schedule
every 6h
cron
17 */6 * * *
budget
$0.30
dep_slopsquat.md →

drift-scanner

pro

Detect silent drift between codebase-map manifest and the live tree — missing entry points, broken commands.

schedule
every 2h
cron
23 */2 * * *
budget
$0.15
drift_scanner.md →

nerf-check

pro

Wraps nerf_detector against rolling baseline. Surface silent model degradation to your notify channel.

schedule
09:00 daily
cron
0 9 * * *
budget
$0.05
nerf_check.md →

boneyard-watch

pro

Top-5 worst agent moves this week with extracted lessons. Flags repeat offenders across weeks.

schedule
weekly, Fri 09:00
cron
0 9 * * *
budget
$0.20
boneyard_watch.md →

custom

pro

Escape-hatch preset. Your prompt in ai/loop.yaml, sandboxed unless allow_actions: true is set explicitly.

schedule
every 45 min (override)
cron
*/45 * * * *
budget
$0.10 (cap $0.50)
custom.md →

Tier access

  • free — included on every plan. Today that is standup-digest.
  • pro — Pro tier unlocks the majority of presets.
  • team — Team tier unlocks security-differentiator scans (dep-slopsquat).

Write your own

If a scan proves its value over two weeks of advisory-only fires, open a PR that adds a named preset: a .md file under scripts/loop_presets/ and an entry in DEFAULT_PRESETS. The contract is short — ROLE, INPUTS, CRITERIA, OUTPUT (strict JSON), CONSTRAINTS. See pr_health.md for the template.