Equipt rates how ready a skill or agent is to equip into your AI. The headline is Readiness (0–100), a weighted blend of four axes (a fifth, Fit, is deferred to v1.1), capped by a safety gate.
Axes
| Axis | What it measures | v1 method |
|---|---|---|
| Craft | How well it is built | Static: frontmatter completeness, description quality, body substance, structure, examples |
| Fit | Triggers at the right time, not the wrong time | Deferred to v1.1 (needs an LLM eval). Shown on cards as “coming”. |
| Guard | Safety to run | Static scan of the asset's code (fenced/inline, not prose): destructive / credential-in-URL hard-fails, secret-exfiltration over the network (curl/wget and fetch/XHR/requests/Invoke-WebRequest/…), shell/eval/spawn risk, tool-scope (least privilege) |
| Proof | Evidence it works | Static, continuous: count of example blocks + usage/how-to + structure depth + adoption (low weight) — no flat baseline, so thin assets score low |
| Upkeep | Alive & current | Git recency + current-spec frontmatter compatibility |
Readiness
Readiness = 0.30·Craft + 0.30·Guard + 0.20·Proof + 0.20·Upkeep
Fit is excluded from the v1 weighting (it needs an LLM eval) and lands in
v1.1. If a scored axis is ever missing, its weight is redistributed across the
rest and the card is marked partial: true.
Guard gate
Guard is also a gate: a gated hard-fail, or any Guard score below 40, caps
Readiness at 40 and flags the asset unsafe: true. Unsafe assets can never be
promoted above Provisional.
Tiers
| Tier | Requirement |
|---|---|
| Provisional | Machine-scored (default). |
| Certified | Human-reviewed + Readiness ≥ 70 (and not partial/unsafe). |
| Field-Ready | Human-reviewed + Readiness ≥ 85, Guard ≥ 70, Proof ≥ 60. |
A tier above Provisional requires an entry in curation.json; the scorer only
honours it when the asset still meets the numeric threshold.
Versioning
Changing any weight, threshold, or hard-fail rule bumps standardVersion
(currently 1.1.1) so historical scorecards remain interpretable.
Note on tiers: Provisional means machine-scored, not yet human-reviewed — it is not a quality verdict. A high Readiness with a Provisional tier just means a human hasn't signed off yet.
Roadmap
- Fit and Guard (adversarial injection) LLM evals — the differentiating signals — land in the follow-on evals plan.