agentleFS
mikeOnBreezemikeOnBreeze/cc-crossbeam/.claude/skills/long-running-agent/SKILL.md

long-running-agent

Skill297 starsChanged 9 months agoMITView on GitHub

This skill converts planning docs and specs into phase-based task structures and prompts for long-running Claude agents. Use when setting up a multi-session agent workflow or breaking down a spec into phases with verification checkpoints.

7 other projects use a version of this, 1 of them edited.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

4 rules. 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 6 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Before doing anything, fetch and read this blog post for the core patterns:

    line 18
  2. Very specific:

    Create the task file using assets/task-template.json structure:

    line 73
  3. Very specific:

    Use assets/claude-prompt-template.md as the base.

    line 82
  4. Distinctive:

    Complete ALL tasks in phase: - No stopping mid-phase

    line 118

“Common” is measured: kept by at least 6 in 10 of the 6 edited versions of this file we could read. “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

Long-Running Agent Setup

Convert specs and planning documents into phase-based task structures for autonomous multi-session execution.

When to Use This Skill

  • User has a spec or planning doc to execute with a long-running agent
  • User wants to break down a project into phases with verification checkpoints
  • User wants to create a claude-prompt.md for autonomous task execution

Source of Truth

Before doing anything, fetch and read this blog post for the core patterns: https://www.anthropic.com/engineering/effective-harnesses-for-long-running-agents

This blog from Anthropic's engineering team defines the effective patterns. Apply them directly.

Core Workflow

1. Gather Inputs

Request from the user:

  • Spec/Planning Doc: The document to convert (required)
  • Project Name: Short identifier
  • Output Location: Where to create files

2. Read the Blog

Fetch the Anthropic blog post above. The key patterns are:

  • Phase-based work with verification checkpoints
  • Explicit feature/task enumeration (granular, testable items)
  • Task file as progress tracker
  • Git checkpointing after each task
  • Strong constraints (no test deletion, no skipping phases)

3. Decompose into Phases and Tasks

Break the spec into:

Phases (3-8 typically):

  • Logical groupings of related work
  • Each phase has verification steps to test completion
  • Agent completes ALL tasks in a phase, then stops for user verification

Tasks (per phase):

  • Granular, implementable units
  • Each task has specific steps
  • Uses passes: true/false to track completion

Example structure from a real project:

Phase 1: Project Foundation (3 tasks)
  - setup-001: Initialize project structure
  - setup-002: Create environment config
  - types-001: Define core interfaces
  Verification: "Run npx tsc --noEmit, confirm no errors"

Phase 2: Storage & Skills (5 tasks)
  - storage-001: Create storage interface
  - skill-001 to skill-004: Create skill files
  Verification: "Confirm skills load, storage works"

... etc

4. Generate claude-task.json

Create the task file using assets/task-template.json structure:

  • phases array with verification steps
  • tasks array with passes: false initially
  • Tasks reference their parent phase

Output: {project-root}/claude-task.json

5. Generate claude-prompt.md

Use assets/claude-prompt-template.md as the base.

Key sections to customize:

  • Project overview and goal
  • Key files list (SPEC.md, claude-task.json, any API docs)
  • Phases table showing all phases
  • File structure target
  • Technical decisions specific to the project

Output: {project-root}/claude-prompt.md

6. Keep the SPEC

The original spec should remain as SPEC.md for the agent to reference when it needs detailed requirements.

Output Structure

{project-root}/
├── claude-prompt.md      # Agent instructions
├── claude-task.json      # Phases and tasks
└── SPEC.md               # Original planning doc (kept for reference)

Starting the Agent

Instruct the user:

@claude-prompt.md

The agent will read claude-task.json, find the current phase, and work through tasks until phase completion.

Key Patterns (from the blog)

  1. Phase boundaries = verification checkpoints - Agent stops, user verifies, then continues
  2. Complete ALL tasks in phase - No stopping mid-phase
  3. Git commit after each task - task-XXX: description
  4. Never skip phases - Sequential progression
  5. Task file is the source of truth - Agent reads and updates it

Reference Materials

  • assets/claude-prompt-template.md - Template for claude-prompt.md
  • assets/task-template.json - Template for claude-task.json structure

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.