| name | prompt-refine |
| description | Full prompt-improvement workflow — runs prompt-audit to find inconsistencies, presents a remediation plan, then runs prompt-migrate to apply fixes and prompt-align to verify harness consistency. Use for any agent or skill prompt that needs structural repair. |
/prompt-refine
End-to-end prompt improvement: audit → plan → migrate → align.
Invocation
/prompt-refine <path-to-prompt-file>
Step 1 — Audit
Invoke /prompt-audit <path-to-prompt-file>. Read its full output (Q&A, learnings, remediation list).
Step 2 — Confirm scope with user
Present the CRITICAL and HIGH remediation items. Ask:
"These are the changes I'll apply. Anything to add, remove, or defer?"
Wait for confirmation before proceeding. If the user defers any items, note them as OUT-OF-SCOPE for this run.
Step 3 — Migrate
Invoke /prompt-migrate <path-to-prompt-file> --remediation <audit-output> using only the confirmed items.
Step 4 — Align
Invoke /prompt-align <path-to-prompt-file>. If any ✗ conflict rows remain after migration, fix them inline (they were missed by the migrate step) and commit:
git -C "${CLAUDE_PLUGIN_ROOT}/../.." add <files>
git -C "${CLAUDE_PLUGIN_ROOT}/../.." commit -m "fix(<scope>): align harness after prompt migration"
Step 5 — Final verification
cd "${CLAUDE_PLUGIN_ROOT}/../.." && npm test && ./tools/lint-marketplace.sh
Expected: All green.
Step 6 — Summary
Emit:
## Prompt-refine summary
Target: <file>
Audit findings: <N CRITICAL, M HIGH, K MEDIUM, J LOW>
Applied: <N CRITICAL, M HIGH> (<K MEDIUM if any>)
Deferred: <items>
Test result: PASS
Harness alignment: clean | <N conflicts fixed>
