Skip to content
Back to skills

Controlflow Planning

ASecurity

Use when a repository task needs a strict ControlFlow-style implementation plan before coding, especially for small, medium, or large scope, cross-file edits, risky migrations, architectural uncertainty, or any work that should produce a saved Markdown plan artifact in plans/.

  • 13 stars
  • 0 votes
  • 0 copies
  • 9 views
  • Added May 27, 2026
data-aigo

Security analysis

A100/100

Pro scans all 7 files and shows the line behind each finding

Scanned September 6, 2026

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

Installs into .claude/skills of the current project.

Are you the author of Controlflow Planning?

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

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

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-planning
description: "Use when a repository task needs a strict ControlFlow-style implementation plan before coding, especially for small, medium, or large scope, cross-file edits, risky migrations, architectural uncertainty, or any work that should produce a saved Markdown plan artifact in plans/."
---

# ControlFlow Planning

## Overview

Create a durable, execution-ready Codex plan. Keep the contract local: the saved artifact defines scope, risks, files, phases, validation, review route, and handoff without depending on VS Code agent mode.

## Local Contract

- Read the repository before phase decomposition; keep verified facts separate from assumptions.
- Use [references/plan-template.md](references/plan-template.md) as the plan contract, not a loose outline.
- Apply [references/llm-behavior-guidelines.md](references/llm-behavior-guidelines.md) for generic assumption, simplicity, scope, and verification discipline.
- Keep Codex portable: ask the user directly when a requirement changes scope, or record a bounded assumption when it does not.

## Workflow

1. Create a saved plan when the user asks for one or when work is `SMALL`, `MEDIUM`, or `LARGE`; skip it only for truly `TRIVIAL` work.
2. Save to `plans/<task-slug>-plan.md` unless the user names another path.
3. Clarify only when the answer changes file scope, user-visible behavior, architecture, or destructive-risk handling.
4. Map likely files, tests, commands, dependencies, and change boundaries before phase decomposition.
5. Read [references/complexity-tiers.md](references/complexity-tiers.md) and [references/semantic-risk-taxonomy.md](references/semantic-risk-taxonomy.md); assign one tier and fill all seven risk categories.
6. Write the artifact using [references/plan-template.md](references/plan-template.md).
7. For non-trivial Codex plans (`SMALL`, `MEDIUM`, `LARGE`), include the five fixed lifecycle sections in this exact order and with these exact headings: `## Progress`, `## Discoveries`, `## Decision Log`, `## Outcomes`, `## Idempotence & Recovery`. These headings are required by [../../scripts/validate-strict-artifacts.ps1](../../scripts/validate-strict-artifacts.ps1) and must match the template and fixtures exactly.
8. Use `ABSTAIN` or `REPLAN_REQUIRED` when confidence is below `0.9` or evidence is insufficient.
9. Add research or spike phases before implementation when a `HIGH`-impact risk is unresolved.
10. Keep phases incremental, testable, and explicit about files, dependencies, quality gates, and failure expectations.
11. Route non-trivial plans through strict review: `SMALL` -> `controlflow-plan-audit`; `MEDIUM`/`LARGE` -> add `controlflow-assumption-verifier`; unresolved `HIGH` risk -> include `controlflow-assumption-verifier` regardless of size.
12. Check [references/controlflow-portability.md](references/controlflow-portability.md) before handoff.

## Spec Capture Handoff

When requirements, acceptance criteria, boundaries, or verification gates are not explicit, use `controlflow-spec` first. Carry its `spec_path`, scope, acceptance criteria, constraints, success metrics, and open risks into the plan. If spec capture is unnecessary, state why the request is already clear enough.

## Planning-Specific Failure Checks

- Do not plan from chat memory when a spec is required; cite the spec artifact or create one.
- Do not skip semantic risk review because the tier feels low; fill every category.
- Do not decompose phases before reading the repository and mapping likely files/tests.
- Do not mark `READY_FOR_EXECUTION` without a strict review route and artifact destination.
- Generic anti-rationalization rules live in [references/llm-behavior-guidelines.md](references/llm-behavior-guidelines.md).

## Mandatory Output Contract

- Prefer structured text over raw JSON.
- Default plan file path: `plans/<task-slug>-plan.md`.
- Use the ControlFlow-style header and section order from [references/plan-template.md](references/plan-template.md).
- Write `Agent: Planner` in the artifact so the document matches the original conceptual role even though the implementation is a Codex skill.
- Record assumptions separately from verified facts.
- For `READY_FOR_EXECUTION` plans, include a clear handoff section that points execution to orchestration.

## References

- `references/plan-template.md`
- `references/complexity-tiers.md`
- `references/semantic-risk-taxonomy.md`
- `references/controlflow-portability.md`
- `references/planner-output-contract.md`
- `references/llm-behavior-guidelines.md`

Files in this skill

  • SKILL.md4.4 KB
  • references/complexity-tiers.md1.2 KB
  • references/controlflow-portability.md2.4 KB
  • references/llm-behavior-guidelines.md3 KB
  • references/plan-template.md3.6 KB
  • references/planner-output-contract.md1.8 KB
  • references/semantic-risk-taxonomy.md2 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.