loop-readiness-score

Loop Readiness Score

Evaluate a completed goal contract. Do not select a methodology, rewrite the goal, run the task, or turn a score into permission.

Installed dependencies

The checker imports the sibling goal-engineering renderer, which imports methodology-selector. Confirm that all three skills are installed under the same skills root. If either sibling is missing, stop and report the required installation set.

Input

Read references/readiness-contract.md. Use the goal renderer's goal_result as authoritative character evidence and the structured goal_input contract for control checks. The checker rerenders goal_input and requires an exact match with the supplied goal; metadata that never reached the executable goal cannot satisfy readiness.

The eleven answers are no, partial, or yes. Each earns zero, half, or all of its criterion weight. Weights total 100. The score communicates maturity; it cannot average away a hard safety failure.

Use the live four-band model: 0–39 Not ready, 40–59 Manual supervision required, 60–79 Semi-autonomous loop candidate, and 80–100 Strong loop candidate. Round the weighted total once with Math.round before choosing the band or applying the 80-point threshold. The band describes maturity; the separate verdict still applies hard and supervision gates.

Hard gates

Return blocked when any of these is missing or false:

• a specific task outcome;
• validation checks and evidence;
• explicit boundaries;
• at least one stop condition;
• fallback behavior;
• a generated goal within its effective character limit;
• an exact rerender match between the canonical input and executable goal;
• explicit risk and applicability declarations;
• human approval when the contract says approval is required.

Numeric coverage can never compensate for a failed hard gate.

Supervision gates

When hard gates pass, return supervised if an applicable control is absent:

• independent checker for medium/high-risk work;
• smallest useful run for a broad or high-risk task;
• execution handoff for broad or mutating work;
• budget for long-running, repeated, or scheduled work;
• isolation for mutating work;
• persistence for repeated, resumable, or scheduled work;
• schedule policy and the canonical activate or change a schedule human approval action for scheduled work;
• explicit method-fit evidence.

Return ready only when hard and applicable supervision gates pass.

Command

node /absolute/path/to/loop-readiness-score/scripts/check-readiness-score.mjs /absolute/path/to/readiness-input.json

Resolve the first path from the directory containing this SKILL.md. The agent prepares the JSON input; do not ask the human to hand-write it. The result contains score, max_score, score_band, the eleven criterion results, verdict, hard-gate evidence, supervision-gate evidence, a deterministic assessment_id, and the single recommended next correction. blocked exits 2, supervised exits 3, and ready exits 0.

Boundaries

• Do not modify the goal.
• Do not infer that a missing gate probably exists elsewhere.
• Do not average away a hard failure.
• Do not authorize execution, merge, deploy, deletion, purchase, or external communication.

Stop

Stop after one verdict. Route corrections back to goal-engineering or the human who owns the missing scope decision.