Agent skill

do

Manifest executor. Iterates through Deliverables satisfying Acceptance Criteria, then verifies all ACs and Global Invariants pass. Use when you have a manifest from /define.

Stars 163
Forks 31

Install this agent skill to your Project

npx add-skill https://github.com/majiayu000/claude-skill-registry/tree/main/skills/other/do-doodledood-claude-code-plugins

SKILL.md

/do - Manifest Executor

Goal

Execute a Manifest: satisfy all Deliverables' Acceptance Criteria, then verify everything passes (including Global Invariants).

Why quality execution matters: The manifest front-loaded the thinking—criteria are already defined. Your job is implementation that passes verification on first attempt. Every verification failure is rework.

Input

$ARGUMENTS = manifest file path (REQUIRED)

If no arguments: Output error "Usage: /do "

Principles

  1. ACs define success, not the path - Work toward acceptance criteria however makes sense. The manifest says WHAT, you decide HOW.

  2. Target failures specifically - On verification failure, fix the specific failing criterion. Don't restart from scratch. Don't touch passing criteria.

  3. Respect tradeoffs - When values conflict, check the manifest's "Tradeoffs & Preferences" section and apply.

Constraints

Must call /verify - Can't declare done without verification. When all deliverables addressed:

Skill("vibe-experimental:verify", "/tmp/manifest-{ts}.md /tmp/do-log-{ts}.md")

Todo Discipline

Create todo list immediately based on manifest structure—deliverables and their ACs. Use D{N}: prefix. Adapt to THIS manifest's complexity.

Required elements:

  • Log file creation (/tmp/do-log-{timestamp}.md)
  • →log after implementation steps (externalizes progress)
  • (expand: ...) when sub-tasks will emerge during implementation
  • Refresh: read full log before calling /verify
  • Acceptance criteria on each todo ("; done when X")

Update todos after every substantive action—no batching completions.

Log Discipline

Create execution log - Write to /tmp/do-log-{timestamp}.md. This is your working memory.

Write as you go - After each significant action, update the log. Don't wait until the end.

Refresh before /verify - Read the full log to restore context before verification.

What to Do

Read the manifest - Extract intent, global invariants, deliverables with their ACs, tradeoffs.

Work through deliverables - For each, satisfy its acceptance criteria. Log your work.

Call /verify - When all deliverables addressed, verify everything passes.

Handle failures - Fix specific failing criteria, call /verify again. Loop until pass.

Escalate when stuck - If you've tried 3+ approaches on a criterion and can't satisfy it:

Skill("vibe-experimental:escalate", "[criterion ID] blocking after 3 attempts")

Log Structure

markdown
# Execution Log

Manifest: [path]
Started: [timestamp]

## Intent
**Goal:** [from manifest]

## Deliverable 1: [Name]

### AC-1.1: [description]
- Approach: [what you tried]
- Result: [outcome]

### Status: [COMPLETE/IN PROGRESS]

## Verification Attempts

### Attempt 1
- Results: [summary]
- Action: [what to fix]

Flow

1. Read manifest
2. Create todos + log
3. For each deliverable: work toward ACs, log progress
4. Call /verify
5. If failures: fix specific criteria, /verify again
6. All pass: /verify calls /done

Expand your agent's capabilities with these related and highly-rated skills.

Didn't find tool you were looking for?

Be as detailed as possible for better results