tools-archfit

archfit deterministic evidence

Use archfit as a deterministic measurement and gate engine. It supplies whole-repo facts; the architect still judges intent, domain volatility, runtime/deploy context, false positives, and missing risks.

When to use

Use when a target repo contains .archfit.yaml or archfit.yaml, when archfit is available on PATH, or when the user asks to combine architect review with archfit output. Use it during architecture-review before final triage/scoring so hard facts can reduce sampling bias.

Do not use this skill for:

• archfit analyze --ai-summary or archfit explain <fingerprint> --ai-summary AI narration as primary evidence;
• archfit config init, config update --apply, baseline, config enrich (labels/owner/volatility/subdomain), config init --ai-classify, config enrich ... --apply, or config writes unless the user explicitly asks for config work;
• applying fixes from agent_tasks;
• replacing the architect's independent evidence gathering and Balanced Coupling judgment.

Skill navigation

• Missing or absent config: return to architecture-review; record archfit as missing or skipped. Optional archfit absence is a coverage gap, not a blocker.
• Current skill: run deterministic archfit commands, summarize facts, and build a calibration matrix.
• Next skill: return to architecture-review for independent verification, scoring, and report writing; use methodology-architecture-fitness for checks that should become enforced gates.

Read-only discipline

Prefer existing config only. Write command outputs to a temp directory outside the target repo. Do not create baselines, labels, generated reports, or config patches inside the repo without explicit user approval.

If archfit or a child tool writes caches or indexes under the target repo, record that path as generated tool state. If the output path is inside the scanned root, warn that reports can contaminate measurements and move future outputs outside the repo or exclude them.