loop-engineering — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited loop-engineering (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Operational skill distilled from 125-source full ingest (93 non-GitHub line-indexed + 33 GitHub code audits). Load REFERENCE.md for doctrine; INSTALL.md for code-backed installs; CORPUS_INDEX.md for source digests.
Before recommending any repo or install path:
python3 scripts/load-digest.py --repo {slug} OR read data/repo-audits/{owner__repo}.mdfile:line) in your answerDo not recommend installs from REFERENCE.md alone.
/loop, /goal, Routines, Ralph, or custom bash enginesNeed recurring agent work?
├─ Yes → Stop condition verifiable?
│ ├─ Yes → Maturity path (below)
│ └─ No → Fix stop condition first; do not schedule
└─ No → Single-shot skill dispatch; not a loop
Maturity path (jpoindexter / corpus consensus):
1. Reliable manual run (one skill, one outcome)
2. Encode as SKILL.md
3. Add state file or kernel write-back
4. Wrap in gated loop (mechanical + human gate)
5. Schedule via /loop or /goalEvery loop design MUST specify:
| Field | Question |
|---|---|
| Trigger | Schedule, event, or human /loop |
| Input state | What the agent reads (files, DB, kernel) |
| Skill/playbook | Which SKILL.md stages run |
| Verifier | Separate checker sub-agent or mechanical gate |
| Stop condition | Boolean test — not vibes |
| Write-back | What persists after each cycle |
| Human gate | When human must approve before next cycle |
| Primitive | Use when |
|---|---|
/loop | Fixed interval re-run of prompt or slash-command |
/goal | Run until separate model grades "done" |
| Routines | Anthropic-documented recurring workflows |
| Sub-agents | Maker builds; checker verifies (never same agent) |
| Context kernel | Cross-session state at .kernel/KERNEL.md |
See portability kit Automations/README.md for heartbeat layer.
Required: load audit before recommending.
python3 ~/.claude/skills/loop-engineering/scripts/load-digest.py --repo cobusgreyling/loop-engineering
python3 ~/.claude/skills/loop-engineering/scripts/load-digest.py --source S02
python3 ~/.claude/skills/loop-engineering/scripts/repo-pick.py --task "loop init"Default layered stack (v2 code audit confirmed):
/loop + /goal (built-in)cobusgreyling/loop-engineering — 201 files read, score 9.62earendil-works/pi (843 files)xpriment626/pi-factory (68 files, SQLite blackboard)loop-engineering-architect agentsuperpowers:verification-before-completion + code-reviewerwrite-a-skill after one reliable manual runRe-run when corpus grows:
python3 research/loop-engineering-agent-hub-2026/scripts/ingest_corpus_full.py
python3 research/loop-engineering-agent-hub-2026/scripts/audit_github_repos.py
python3 research/loop-engineering-agent-hub-2026/scripts/build_evidence_v2.py
python3 research/loop-engineering-agent-hub-2026/scripts/synthesize_doctrine_v2.pyCopy updated synthesis → REFERENCE.md, INSTALL.md, CORPUS_INDEX.md.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.