Minimal engineering loop scaffold for daily triage, verification, and controlled auto-fix — built around a readiness ladder (L0–L3) that gates how much autonomy the loop is allowed before a human has to sign off.
npx degit your-org/loop-engineering-starter my-loop
cd my-loop
git initIf you applied the minimal-loop starter manually instead, confirm you have these files before continuing:
STATE.md— current project spine (name, owner, last run frequency, open items)LOOP.md— loop config for the team.grok/skills/loop-triage/SKILL.md— triage skill
The scaffold ships 7 loops, gated into 4 readiness levels. You don't enable all of them at once — each level unlocks the next loop only after the prior one has run clean for a defined period.
| Level | Unlocks | Gate to advance |
|---|---|---|
| L0 — Observe | loop-triage only |
N/A — always on from day one |
| L1 — Verify | loop-verifier |
[N] consecutive clean triage runs with no false positives |
| L2 — Assist | minimal-fix (proposes diffs, doesn't merge) |
[N] verified runs, fix proposals reviewed by a human |
| L3 — Auto-fix | Auto-merge for [whitelisted fix categories] | [N] consecutive human-approved fixes of that category, zero rollbacks |
Fill in the bracketed thresholds (
[N], fix categories) with your team's actual gate values — these are config, not fixed rules, and should live inLOOP.mdso they're versioned alongside the loops themselves.
The other 3 loops in the set of 7 — [loop-name-5], [loop-name-6], [loop-name-7] — slot into this ladder at [L_]; document what each does and where it sits before publishing, since the ladder is the whole pitch of this repo.
- Triage each morning using the
loop-triageskill. - Review findings appended to
STATE.md. - Tune
loop-triage/SKILL.mdfor 1–2 weeks before trusting its output. - Once you've cleared the L1 gate above, add
loop-verifier. - Once you've cleared the L2 gate, add
minimal-fixand start reviewing (not merging) its proposals. - Only after the L3 gate is cleared for a given fix category should auto-merge be turned on for that category — never globally.
No automated fixes. Triage and observe only. This is L0, and it's mandatory regardless of how confident the audit score looks.
git(any reasonably current version — no sparse-checkout, worktrees, or other 2.x-specific features are used)npx @cobusgreyling/loop-audit(for readiness score, see below)
npx @cobusgreyling/loop-audit .Example output:
loop-audit v[x.y.z] — scanning .
✔ STATE.md found, last updated [N] days ago
✔ loop-triage/SKILL.md present
✔ 14 consecutive clean triage runs
✘ loop-verifier not yet enabled
Readiness: L1 (Verify) — eligible for L2
Next gate: enable loop-verifier, accumulate [N] verified runs
Replace this block with real output from a run against this repo. A reviewer should be able to see what "ready" looks like without running the tool themselves.
.
├── STATE.md — project spine: owner, run cadence, open items, triage log
├── LOOP.md — loop config: which loops are enabled, gate thresholds per level
├── README.md — this file
├── SECURITY.md — vulnerability reporting process
├── CHANGELOG.md — version history of the scaffold itself
├── ARCHITECTURE.md — loop state machine and fix-gating logic
├── CODEOWNERS — review routing (no extension — GitHub only recognizes this exact filename)
├── ROADMAP.md — planned loops and readiness-ladder changes
├── CONTRIBUTING.md — how to propose a new loop or skill
└── .grok
└── skills
└── loop-triage
└── SKILL.md
See SECURITY.md for reporting vulnerabilities.
See LICENSE (MIT).