Skip to content

Ilm-Alan/project-spec

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

19 Commits
 
 
 
 
 
 

Repository files navigation

Project Spec

Create a project spec for AI agents to work from autonomously. Works with Claude Code (Anthropic), Codex (OpenAI), and Gemini CLI (Google).

Features

  • 5-step process — Explore → Ask → Approaches → Derive → Write
  • 7-section structure — Instructions, Mission, Architecture, Components, Constraints, Build Order, Done When
  • State management — PLAN.md for the spec, PROGRESS.md for tracking across sessions
  • Anti-filler guidance — Concrete examples of vague writing to avoid

Installation

Claude Code

mkdir -p ~/.claude/skills/project-spec && curl -o ~/.claude/skills/project-spec/SKILL.md https://raw.githubusercontent.com/ilm-alan/project-spec/main/SKILL.md

Codex

mkdir -p ~/.codex/skills/project-spec && curl -o ~/.codex/skills/project-spec/SKILL.md https://raw.githubusercontent.com/ilm-alan/project-spec/main/SKILL.md

Gemini CLI

First, enable the experimental skills feature in ~/.gemini/settings.json:

{
  "experimental": {
    "skills": true
  }
}

Then install the skill:

mkdir -p ~/.gemini/skills/project-spec && curl -o ~/.gemini/skills/project-spec/SKILL.md https://raw.githubusercontent.com/ilm-alan/project-spec/main/SKILL.md

Run /skills list to verify the skill was loaded.

Usage

Invoke the skill when starting a new project:

/project-spec

Or reference it contextually:

Using the project spec skill, create a spec for a CLI tool that manages AWS credentials

Structure

Every generated spec contains:

Section Purpose
Instructions How the agent operates autonomously—build without asking, make decisions from code
Mission What, who, why in 1-3 lines
Architecture Technical decisions stated, not debated
Components What to build—one line each
Constraints Non-negotiables only
Build Order Sequenced phases by name
Done When Concrete, testable acceptance criteria

Output Files

  • PLAN.md — The project spec (under 100 lines)
  • PROGRESS.md — State tracking updated by the executing agent

Example

Prompt:

/project-spec

Conversation:

What are you building?
> A CLI tool for rotating AWS access keys

Who is this for?
> DevOps engineers managing multiple AWS accounts

What does "done" look like?
> Keys rotate without downtime, old keys are revoked after confirmation

Output (PLAN.md):

## Instructions
Read PROGRESS.md to know what's done. Check git log for context. Build the next incomplete phase—do not ask what to work on. Make decisions from existing code patterns. Update PROGRESS.md when done and continue to next phase. No clarifying questions—inspect code, make reasonable choices, ship working features.

## Mission
Your job is to build a CLI tool for rotating AWS access keys. For DevOps engineers managing multiple AWS accounts. Eliminates manual key rotation and reduces security risk from stale credentials.

## Architecture
Go binary. AWS SDK for credential operations. Viper for config. No database—uses AWS IAM as source of truth.

## Components
- Key rotator: creates new key, updates configs, revokes old key
- Config manager: reads/writes credential locations from ~/.aws and environment
- Account switcher: handles multiple AWS profiles

## Constraints
- Zero downtime during rotation
- Old key revoked only after new key confirmed working
- Must support assumed roles

## Build Order
1. Config manager
2. Key rotator
3. Account switcher

## Done When
- `rotate` command creates new key, updates all configured locations, revokes old key
- `status` command shows key age across all profiles
- Works with 10+ AWS profiles without performance degradation

License

MIT. See LICENSE.txt.

Guidance

Perfection is a continual process.

About

Execution protocol skill for AI coding assistants. Generates structured project specs with mission, architecture, components, and acceptance criteria. Works with Claude Code, Codex, and Gemini CLI.

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors