| name | workflow-creator-improve |
| description | Use when auditing, repairing, redesigning, migrating, or applying feedback to an existing workflow. Corrective shared-v1 entry; audit-only requests remain read-only. |
| allowed-tools | Read, Write, Edit, Grep, Glob, Bash, Skill, AskUserQuestion, EnterPlanMode, ExitPlanMode, Agent, Workflow |
| hooks | {"PreToolUse":[{"matcher":"Read|Glob|Grep|Bash","hooks":[{"type":"command","command":"bun ${CLAUDE_PLUGIN_ROOT}/hooks/clarify-before-recon-guard.ts --workflow workflow-creator"}]},{"matcher":"Write|Edit|MultiEdit|NotebookEdit|Bash","hooks":[{"type":"command","command":"bun ${CLAUDE_PLUGIN_ROOT}/hooks/orchestrator-mutation-guard.ts --workflow workflow-creator"}]},{"matcher":"Agent|Workflow","hooks":[{"type":"command","command":"bun ${CLAUDE_PLUGIN_ROOT}/hooks/approved-artifact-gate.ts --workflow workflow-creator"}]}],"PostToolUse":[{"matcher":"AskUserQuestion","hooks":[{"type":"command","command":"bun ${CLAUDE_PLUGIN_ROOT}/hooks/episode-phase.ts --workflow workflow-creator"}]},{"matcher":"ExitPlanMode","hooks":[{"type":"command","command":"bun ${CLAUDE_PLUGIN_ROOT}/hooks/approved-artifact-persist.ts --workflow workflow-creator"}]}]} |
Workflow Creator Improve — Corrective Entry
Correct an existing workflow through shared-v1. Audit-only requests never receive mutation authority.
Write surface: main chat does not do the work
You may Write/Edit only under .planning/ and .claude/. Every target workflow file — skills,
hooks, scripts, tests — is written by a dispatched agent. orchestrator-mutation-guard is
registered in this skill's frontmatter, so the attempt is REFUSED, not corrected: a write you try
anyway costs a turn and produces nothing. Reach for Agent first, not after a denial. Bash is held
to the same line — only the named read-only checks and compilers are permitted from main chat.
Two narrow exceptions: the generated plan while you are IN Plan mode, and .claude-workflows.json
when adopting governance.
Entry and compatibility
Reject legacy .planning/wc/** lifecycle state. A resumable episode must have .planning/ACTIVE_WORKFLOW.md with workflow: workflow-creator, lifecycle: shared-v1, and a semantic phase. Otherwise start a fresh corrective episode.
Initialize or update the marker:
---
workflow: workflow-creator
lifecycle: shared-v1
entry: corrective
phase: clarify
target: <workflow slug>
---
Read beat-clarify and ask desired correction, exclusions, whether the request is audit-only or mutation-bearing, required evidence, and human review surface before inspecting the target.
Branch
CLARIFY → read-only workflow-creator-verify diagnosis
├─ audit-only → HUMAN REVIEW
└─ repair/redesign/migrate → native Plan → plan review → compiler → beat-implement
→ independent workflow-creator-verify verification → HUMAN REVIEW
Audit-only
Set phase: diagnosis. Invoke workflow-creator-verify with auditOnly: true, readOnly: true, a deterministic targetFiles manifest, semantic phases, approved criteriaRows, and mechanicalProbes: []. Pre-approval diagnosis dispatches only the workflow-auditor agent profile, whose tools are structurally limited to Read, Grep, and Glob; it never executes caller-supplied commands or project scripts. Hook-contract and mechanical-probe execution are explicitly deferred to an approved verification run. Render evidence-bearing findings into .planning/AUTOMATED_REVIEW.md, then set phase: human-review and load beat-review. Do not create a generated plan or receipt, dispatch an implementation workflow, or mutate target files.
Repair, redesign, or migration
Use the diagnosis as planning evidence. Enter native Plan mode with the canonical Workflow Output Manifest required by the fresh entry. Structural changes, output-set changes, or altered criteria require a replacement plan and fresh approval; immutable plans are never patched.
After independent plan review:
- Compile the exact approved plan with
scripts/wc/workflow-plan-compiler.ts.
- Dispatch the returned ready wave through
beat-implement with workflow: workflow-creator.
- Re-run the deterministic compliance probe and
workflow-creator-verify independently, selectively carrying authenticated prior reviews where valid:
bun ${CLAUDE_SKILL_DIR}/../../scripts/wc/compliance-probe.ts --target "<plugin repo root>"
The probe is safe pre-approval — it only reads files — and its findings are deterministic, so they are evidence rather than a score. Carry them as a mechanical probe result, never as a reviewer dimension.
4. Continue until every approved criterion has evidence and no blocking finding remains, within the active /goal budget.
5. Return to shared human review in .planning/HUMAN_REVIEW.md.
Boundaries
- Main chat may write only administrative
.planning/.claude state.
- Read-only pre-plan agents are allowed; generic Agent and every Workflow require current approved-plan admission.
- Bash is fail closed to named read-only checks; it may not edit, redirect, compile, or generate target artifacts.
- Compiler failure blocks; never hand-enumerate files or fall back to the retired generator.
- Tactical feedback loops here.
REJECT: returns to clarification and replaces intent/criteria.
