Skip to content
Agentmart
Submit a tool

executing-plans

Use when you have an approved plan to execute step by step with TDD, review gates, and atomic commits.

Skill by POPFD

IndexedMIT★ 1
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/popfd-executing-plans

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Use when you have an approved plan to execute step by step with TDD, review gates, and atomic commits.

Use cases: testing

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/executing-plans

What's inside

Skill name
executing-plans
Description
Use when you have an approved plan to execute step by step with TDD, review gates, and atomic commits.
License
MIT
Instructions
238 lines in SKILL.md · view on GitHub

Files (1)

  • SKILL.md

SKILL.md (first lines)

# Executing Plans

Walk through an approved plan one step at a time. Each step
follows a test-first workflow, passes review, and lands as a
coherent series of small, human-readable commits.

## Plan directory layout

Plans live in their own directory:

```
.sweatshop/plans/<plan-name>/
  plan.md           # the plan file
  step-1.md         # step 1 notes (produced as step 1 lands)
  step-2.md         # step 2 notes
  ...
```

`<plan-name>` is the slug chosen when the plan was written
(e.g. `2026-04-15-add-auth`). The plan file is always named
`plan.md` inside that directory; step notes are siblings
named `step-<N>.md`.

## Preparation

1. **Locate the plan directory** — find
   `.sweatshop/plans/<plan-name>/` and read `plan.md`.
2. **Confirm step order** — plans are executed strictly in
   the order steps are listed. No skipping, no reordering.

## Process per step

For each step in the plan, follow this exact sequence:

1. **Load prior context — only when needed.** Re-read
   `plan.md` and every existing `step-*.md` ONLY if one of
   these is true:
   - This is the first step you are executing in the current
     session.
   - Auto-compaction has fired since the last step (prior
     conversation no longer visible).
   - You are unsure whether a fact from an earlier step is
     still in your context.

   Otherwise skip this step — re-reading the same files every
   iteration is the largest token leak in this workflow. The
   notes are the authoritative source after compaction; before
   compaction, the conversation already has them.
2. **Gather additional context** — if this step touches
   unfamiliar code not covered by prior notes, invoke the
   `research` skill.
3. **Write tests first** — tests that verify the step's
   acceptance criteria. They should fail at this point.
4. **Implement in chunks** — minimum code to make the tests
   pass. Stay scoped to this step only. As you work, plan
   natural commit boundaries: each chunk is one coherent
   idea a reviewer can read in a single sitting (e.g. "add
   helper X", "migrate callers to X", "wire X into the
   endpoint"). Aim for 2–5 chunks per step; one is fine if
   the step is genuinely small. See "Chunking commits"
   below for what makes a good boundary.
5. **Verify** — invoke the `verification` skill once. It
   runs build, test, and lint as a single pass and is
   silent on success. Do NOT invoke `/build`, `/test`, and
   `/lint` separately — that triples the skill-load
   overhead for no extra signal.
6. **Update the plan file** — flip this step's
   acceptance-criteria boxes from `- [ ]` to `- [x]`. Do NOT
   modify any other step's boxes.
7. **Write step notes** — save
   `.sweatshop/plans/<plan-name>/step-<N>.md` using the
   format in the "Per-step notes" section below. These notes
   must survive context compaction and carry forward anything
   later steps will need to know.
8. **Review (risk-gated)** — invoke the `requesting-review`
    skill ONLY when the step is non-trivial. Skip review for:
    - Pure docs/comment changes.
    - Test-only additions where the test follows existing
      patterns.
    - Mechanical renames, formatting, or moves with no logic
      change.
    - Config/tooling edits with no runtime effect.

    Always review for: new logic, API/contract changes,
    security-sensitive code, anything the plan flags as
    high-risk, and any step where domain `paths` match.

    If review requests changes, apply fixes and re-review (max
    3 iterations before escalating).
9. **Commit each chunk** — invoke /commit-changes once per
    chunk identified during implementation. Stage only the
    files belonging to that chunk so each commit reads as a
    single coherent change. The FINAL commit of the step
    must also include the updated plan file AND the step
    notes file. Earlier chunk commits MUST NOT touch the
    plan or notes files.
10. **Report progress** — one line: which step finished and
    what's next. Do NOT prompt the user about compaction.
    Auto-compaction handles context pressure on its own, and
    step notes guarantee state survives whenever it fires.

## Per-step notes

Step notes are a durable handoff to future steps (and to
future-you after compaction). They are NOT a diff summary —
`git show` is authoritative for what changed. Capture only
what is non-obvious from reading the commit.

Path: `.sweatshop/plans/<plan-name>/step-<N>.md`
(e.g. plan at `.sweatshop/plans/2026-04-15-add-auth/plan.md`
→ step 1 notes at
`.sweatshop/plans/2026-04-15-add-auth/step-1.md`).

Format:

```markdown
# Step <N> notes: <step title>

## Decisions

From POPFD/claude-sweatshop, MIT.

Signals

Stars
1
Forks
0
Open issues
0
Last commit (this folder)
Jun 8, 2026 (4 months ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from POPFD

Cite this page

executing-plans skill. Agent Mart. https://agentmart.dev/skills/popfd-executing-plans (last updated 2026-10-24)