Skip to content
Back to skills

Controlflow Verify

ASecurity

Use after a ControlFlow plan is saved and before implementation. Runs tier-gated adversarial verification inline: structural audit, assumption/mirage detection, and cold-start executability simulation.

  • 13 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added September 6, 2026
ai-agentsapisecurity

Works with

  • api

Security analysis

A100/100

Pro scans all 4 files and shows the line behind each finding

Scanned September 6, 2026

npx -y skills add Smithbox-ai/ControlFlow --skill controlflow-verify --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Controlflow Verify?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Controlflow Verify
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/smithbox-ai-controlflow-verify-controlflow/badge)](https://www.skillsdirectory.com/skills/smithbox-ai-controlflow-verify-controlflow)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: controlflow-verify
description: "Use after a ControlFlow plan is saved and before implementation. Runs tier-gated adversarial verification inline: structural audit, assumption/mirage detection, and cold-start executability simulation."
---

# ControlFlow Verify

## Overview

Verify a saved plan before implementation. The checks run inline in the current host
context with zero subagents shipped by the plugin. This skill combines the useful parts of the former
plan-audit, assumption-verifier, and executability-verifier workflows without reproducing
Codex subagent orchestration.

Select the skill explicitly. In Codex, invoke it with `$controlflow-verify`.

## Input and Contract

- Read the plan from disk; do not verify a chat copy.
- If the active repository provides `schemas/planner.plan.schema.json` and
  `plans/templates/plan-document-template.md`, use them.
- Otherwise use `../controlflow-plan/references/plan-format.md`.

## Tier Gating

| Tier | Verification |
| --- | --- |
| `TRIVIAL` | Skip unless explicitly requested |
| `SMALL` | Phase 1 |
| `MEDIUM` | Phases 1–2 |
| `LARGE` | Phases 1–3 |

Any unresolved applicable `HIGH` semantic risk requires all three phases.

## Phase 1 — Structural Audit

- Check the header, section order, lifecycle headings, all seven semantic-risk rows, phase
  shape, diagrams, exact commands, measurable acceptance criteria, and rollback guidance.
- Verify referenced paths and tests against the repository.
- Apply the Minimum Viable Change Ladder before accepting new abstractions, dependencies,
  or generated surfaces.

## Phase 2 — Assumption and Mirage Check

- Try to refute each factual claim that names a file, symbol, dependency, API, version,
  convention, integration, security boundary, or concurrency assumption.
- Classify claims as `confirmed`, `uncertain`, or `refuted`.
- Use [references/mirage-patterns.md](references/mirage-patterns.md).

## Phase 3 — Cold-Start Executability

- Simulate a fresh executor starting the first phases with only the plan and repository.
- Stop at the first real blocker.
- Require concrete inputs, outputs, commands, test specifics, inter-phase contracts, and
  recovery for destructive work.

## Verdict

- `APPROVED`: all required phases pass and implementation may start.
- `NEEDS_REVISION`: the design is viable but specific plan defects must be fixed.
- `REJECTED`: the scope or architecture is not safely deliverable as written.
- Include `failure_classification: fixable`, `needs_replan`, or `escalate` for non-approval.
- Write a compact verdict to `plans/artifacts/<task-slug>/verify-verdict.md`.
- Do not start implementation until the verdict is `APPROVED`.

## Native Host Boundary

- Do not spawn plugin verifier agents.
- If the user explicitly wants an isolated second opinion, native host review or an
  explicitly requested native subagent can supplement this inline pass. In Codex, `/review`
  is the built-in review command.
- The host owns approvals, sandboxing, retries, and subagent lifecycle.

## References

- `references/adversarial-framing.md`
- `references/verify-phases.md`
- `references/mirage-patterns.md`

Files in this skill

  • SKILL.md3.1 KB
  • references/adversarial-framing.md1 KB
  • references/mirage-patterns.md622 B
  • references/verify-phases.md1.4 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.