agentleFS
mthinesmthines/agent-skills/skills/workflow/aw-create-walkthrough/SKILL.md

aw-create-walkthrough

Skill20 starsChanged 13 days agoMITView on GitHub

Writes `.agent/{branch}/walkthrough.md`, a one-page post-hoc review of finished work — what changed, what was assumed, and the evidence — for the reviewer to read after the fact. Use at Phase 6 before the draft PR; triggers on create walkthrough, generate walkthrough, write walkthrough artifact.

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 1 is distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Write .agent/{branch}/walkthrough.md: the page a reviewer reads instead of approving a plan up front.

    line 18
  2. Very specific:

    Never guess the branch name or the base.

    line 49
  3. Very specific:

    Write none only when every choice traces to the plan or the user.

    line 56
  4. Very specific:

    Every file outside File Changes goes in the summary's Out of bounds: line with the reason.

    line 69
  5. Distinctive:

    Write the file from Template, then print its Review summary inline in the conversation together with the PR URL.

    line 73

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Create Walkthrough Artifact

Write .agent/{branch}/walkthrough.md: the page a reviewer reads instead of approving a plan up front. It leads with one screen — what changed, what was assumed, and the evidence — and keeps the detail below it.

Contents

  • Prerequisites
  • Procedure
  • Template
  • Validation

Prerequisites

  1. run-checks.mjs has printed its sentinel line on the current head (CHECKS n/n PASS (<sha>), or the FAIL line with the user's approval to ship).
  2. Documentation is updated.
  3. You are inside the worktree, and .agent/{branch}/plan.md exists for a Full-tier run.

Procedure

Step 1: Gather the facts

BRANCH=$(git branch --show-current)
BASE=$(git merge-base HEAD origin/main 2>/dev/null || git merge-base HEAD main)
mkdir -p ".agent/${BRANCH}"
git diff --stat "$BASE"...HEAD
git diff --name-only "$BASE"...HEAD
git log --oneline "$BASE"...HEAD
cat ".agent/${BRANCH}/plan.md" 2>/dev/null
tail -n 1 ".agent/${BRANCH}/checks.run.log" 2>/dev/null   # last sentinel, when the executor teed it

Never guess the branch name or the base. Take Task, Guardrails, File Changes, and any Decisions from plan.md; take the sentinel line from the last run-checks.mjs output in this session or from checks.run.log.

Step 2: List the assumptions

An assumption is a choice you made that the plan, the user, or the code did not settle. List each with the evidence that would prove it wrong. Write none only when every choice traces to the plan or the user.

<!-- ✗ Wrong — restates the task, names no choice -->
- Assumed the feature should work.

<!-- ✓ Right — a real choice and how to falsify it -->
- Assumed expired tokens return 401, not 403 — the API guide does not say; check `docs/api/errors.md`.

Step 3: Check the guardrails

Compare git diff --name-only with File Changes. Every file outside File Changes goes in the summary's Out of bounds: line with the reason; none when there is none.

Step 4: Write and present

Write the file from Template, then print its Review summary inline in the conversation together with the PR URL. The file alone does not count as delivery: the reviewer reads the inline copy first.

Template

Timestamps are ISO 8601 with time (2026-03-07T14:30:00Z).

---
created: {TIMESTAMP}
branch: {BRANCH}
task: {TASK_DESCRIPTION}
pr: {PR_URL}
---

# Walkthrough: {TASK_DESCRIPTION}

## Review summary

**What changed** — {one sentence}.

```text
{before → after diagram of the change, ≤ 12 lines; a mermaid flowchart is fine when it renders}
request ──▶ authGuard ──▶ handler
                │
                └─ expired token ──▶ 401 (new)
```

**What I assumed**

- {assumption} — {how to falsify it}

**Evidence**

- Checks: `{the sentinel line, verbatim — CHECKS n/n PASS (<sha>)}`
- Suite: `{before-PR command}` → {exit code / summary}
- Out of bounds: {files changed outside File Changes, with reason | none}
- PR: {PR_URL}

## Changes

| File | Change | Purpose |
| ---- | ------ | ------- |

## How to verify

1. {a step a reviewer can run}

## Decisions

1. {decision — rationale} <!-- omit the section when plan.md had none -->

Rules for the template:

PartRule
Review summaryFits one screen (≤ 30 lines); written for a reviewer who has not read the plan
DiagramShows the changed path or structure, not the file list; omit only for a pure text or config edit
Evidence → ChecksThe sentinel line copied verbatim, never paraphrased
ChangesOne row per file in git diff --name-only — no file missing, none invented
How to verifyCommands or clicks a reviewer can repeat

Validation

  • The file is at .agent/{branch}/walkthrough.md inside the worktree.
  • Frontmatter has created, branch, task, pr.
  • The Review summary is ≤ 30 lines and contains What changed, What I assumed, and Evidence.
  • The Checks line equals the latest run-checks.mjs sentinel output.
  • The Changes table lists exactly the files in git diff --name-only "$BASE"...HEAD.
  • The Review summary and PR URL were printed in the conversation.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.