flow-health
Diagnose repository and PWDEV Flow health without changing the project. Use when the user asks for a health check, workspace consistency check, dependency assessment, readiness report, or evidence-bas
SKILL.mdplugins/pwdev-flow/skills/flow-health/SKILL.md
GitHubDiagnose PWDEV Flow Health
Read health, artifacts, audit, and safety before collecting evidence.
Route
Choose exactly one mode from the request:
fullfor repository, Flow workspace, local quality checks, documentation, Git, dependencies, and locally available security checks;workspacefor Flow structure, state, gates, links, memory, audit, and migration consistency;depsfor manifests, lockfiles, locally available dependency checks, and freshness evidence.
Default to full only when the user asks for a general health check. State the selected mode before running commands.
Procedure
- Read applicable
AGENTS.md, then inspect Flow config, state, manifests, and relevant artifacts without opening secret files. - Discover verification commands from repository governance and manifests. Do not invent package-manager commands.
- Run only read-only or non-mutating checks already available locally. Keep network-backed checks separate and request authorization before using them.
- For
workspace, validate links, phase/state agreement, gate evidence, memory index targets, migration records, and audit integrity with flow_audit.py when a trail exists. - Classify findings by severity and grade each applicable area
AthroughFusing the health contract. UseN/Afor unavailable evidence, never as a passing grade. - Recommend bounded next actions but do not implement them.
Constraints
- Remain read-only: do not install packages, update lockfiles, repair audit data, edit state, or apply findings.
- Do not treat a missing optional tool, unavailable network, or stale external database as a product failure; report the limitation and confidence.
- Persist a report only when the user requests it or an active Flow contract already authorizes it.
Output
Return MODE, urgent findings first, a scorecard with evidence and confidence, limitations, OVERALL, and NEXT. Link persisted reports with absolute local paths.