← ClaudeAtlas

controlflow-plan-auditlisted

Use when a saved implementation plan should be adversarially reviewed before coding, especially for small, medium, or large tasks that may contain architecture flaws, scope gaps, destructive risk, weak rollback, weak tests, or dependency conflicts.
Smithbox-ai/ControlFlow · ★ 11 · AI & Automation · score 69
Install: claude install-skill Smithbox-ai/ControlFlow
# ControlFlow Plan Audit ## Overview Adversarially review a saved implementation plan before coding. Look for architecture defects, dependency mistakes, missing rollback, scope gaps, weak tests, and first-phase executability problems. Invoke this skill via `/controlflow-claude-code:controlflow-plan-audit`. ## Local Contract - Read the actual plan artifact and cross-check it against repository files, tests, commands, architecture, and dependencies. - Use [references/audit-checklist.md](references/audit-checklist.md) for audit coverage and [../controlflow-planning/references/llm-behavior-guidelines.md](../controlflow-planning/references/llm-behavior-guidelines.md) for generic evidence discipline. - Save the verdict to `plans/artifacts/<task-slug>/plan-audit.md` using [../../templates/plan-audit-report-template.md](../../templates/plan-audit-report-template.md). - Tie findings to plan sections or repository facts. ## Workflow 1. Read the saved plan artifact first, not a summary. 2. Check file/path reality, available tests and validation commands, architecture/dependency surfaces, rollback, acceptance quality, contract completeness, and first-phase executability. 3. Return one verdict: `APPROVED`, `NEEDS_REVISION`, `REJECTED`, or `ABSTAIN`. 4. For non-approval, classify as `fixable`, `needs_replan`, or `escalate`. 5. Save and summarize evidence-backed findings. ## Spec-Before-Plan Health Check For non-trivial work, confirm the plan is anchored to explicit r