workflow-doctor
Checks if your workflow setup is working right and finds problems without fixing them.
Installation
Paste this into Claude Code, Cursor, or any agent that can run commands.
SKILL.mdShow the author's original SKILL.md
--- name: workflow-doctor description: "Diagnose workflow-axa installation, global routing, rule coverage, duplicate copies and local kit/report assets. Read-only; no automatic repair or installation." --- # Workflow doctor Inspect without changing the environment. Check package/version, host profile/registrations, global/project routing, duplicate copies, local assets and selected Matt specialists. Return PASS / WARNING / BLOCKED / NOT VERIFIED with evidence and the next step in chat, not a new artifact folder. Separate **installed on disk**, **discoverable/enabled**, **actually loaded**, **followed** and **enforced**. Validators cannot prove the later claims. Do not install, enable, delete, overwrite or repair anything automatically. Verify `python` starts; otherwise warn that hooks cannot run. Repeated approval denials: workflow/SETUP.md "Updating sessions safely". Repeated waiting: inspect goal_status and Stop feedback using workflow/TASK-FOLDER.md "Approval waits and host goals". ## Global artifact policy Use workflow/scripts/global_policy.py --check with the actual profile paths. Respect CLAUDE_CONFIG_DIR/CODEX_HOME and active Codex overrides. Missing/changed markers, stale policy, newer deployments, duplicates or linked configuration need review; never repair them automatically. Test an ordinary request without /workflow in the real host to establish loading; remote environments need their own setup. Pass project protections to workers. Policy version 2.5.0 can legitimately differ from the package version; compare bytes, not version equality. ## Shared rules Use rule-maker/INSTALLATION.md and scripts/audit_installation.py --check-global with explicit known candidate paths. activate_rules.py --check remains available. Byte comparisons detect mixed copies; names do not. readiness.py inspects bounded inputs but cannot infer adequate rules from file presence. version_facts.py reports static versions, not runtime compatibility. Adequate existing project rules need no migration. Report `rule-maker/scripts/rules_stamp.py <project-root>` (CURRENT, BEHIND with its changes, NONE or AHEAD). Report host-loading and enforcement gaps separately; missing capabilities block dependent work, not diagnosis. ## UI and report route Check local ui-design, shared design.md, CONTRACT.md, VERIFY.md, LIVE-KIT.md and the diff/index helpers. The UI rule forbids new CSS, allows only existing classes (kit or code) and plain HTML only on explicit instruction. Behaviour-only repairs preserving presentation need no complete kit; new UI needs selected approved components/patterns. A sample, filename or tag is not approval. Check codebase-map/kit_bindings.py and its ui-design sibling. Distinguish current story-source observations, declared approval, real capture provenance and actual renderer use. Missing/deleted/stale sources block their selection, not unrelated work. Do not build a kit, refresh approval hashes or create substitute templates. No UI framework or tool-level enforcement is installed. ## Existing records Report conflicts with preservation of presented reviews or unnecessary bulk migration. Legacy STATUS.md layouts and old review filenames are not setup defects. New approval-ready UI inputs need current kit references; historical reports stay unchanged. Test ordinary prompts, scope changes, resume and delegation on the installed hosts before claiming reliable operation.
Ships with 1 supporting file:
- agents/openai.yaml
Mirrored from the author's public source. Install counts from the open skills registry.