project-management

Project Management Skills for AI Agents

A collection of agent skills for task tracking, focus enforcement, debugging, and complex feature planning. Designed to work with any AI coding agent.

Skills

Installation

CLI Installer (Recommended)

Use npx @sirryou/skill-library install for automated installation and subagent provisioning:

# Install to all supported agent runtimes
npx @sirryou/skill-library install

# Install for OpenAI Codex (automatically provisions skills to ~/.codex/skills and subagents to ~/.codex/agents/*.toml)
npx @sirryou/skill-library install --agent codex

# Install for Antigravity (provisions skills and subagents to ~/.gemini/antigravity-cli/)
npx @sirryou/skill-library install --agent antigravity

# Install project-locally in current repository workspace (.codex/ or .agents/)
npx @sirryou/skill-library install --agent codex --workspace
npx @sirryou/skill-library install --agent antigravity --workspace

For Deep Plan on Codex, inspect the effective [agents] configuration before using nested delegation. The agent-facing guidance is in the Codex multi-agent reference.

Native vendor subagent setup

npx @sirryou/skill-library install --claude
npx @sirryou/skill-library install --codex
npx @sirryou/skill-library install --agy

The external npx skills add CLI supports --skill and --agent, but it does not execute this package native subagent provisioning. Use the package installer after npx skills add when custom subagents are required.

Cross-Agent Skills CLI

npx skills add SirRyou/project-management --skill tree-of-work
npx skills add SirRyou/project-management --skill deep-plan
npx skills add SirRyou/project-management --skill investigate

Claude Code (Plugin)

/plugin marketplace add SirRyou/project-management
/plugin install project-management@project-management

Manual

git clone https://github.com/SirRyou/project-management.git
cp -r project-management/skills/* ~/.claude/skills/  # or ~/.cursor/skills/, etc.

Updating

# Check for updates
npx @sirryou/skill-library check

# Update all agents
npx @sirryou/skill-library update

# Update specific agent
npx @sirryou/skill-library update --agent claude

Claude Code (Plugin)

/plugin uninstall project-management
/plugin install project-management@project-management

Manual

cd project-management
git pull origin master
cp -r skills/* ~/.claude/skills/

Publishing Updates (Maintainers)

# Bump version in package.json and plugin.json
npm version patch  # or minor, major
npm publish --access public

Design Philosophy

These are behavioral skills, not tools. They tell agents how to think about problems — they don't impose specific file formats, directory structures, or CLI tools.

• Portable: Work across any agent runtime (Claude Code, Cursor, Copilot, Codex, Gemini CLI)
• Declarative: Define triggers, rules, and workflows in markdown
• Runtime-agnostic: Delegate persistence to platform-native tools
• Composable: Load references on-demand, not in bulk

Each skill declares what it needs from the runtime (requires and capabilities in frontmatter) and degrades gracefully when capabilities are missing.

Skill Details

Tree of Work

Focus enforcement and context recovery for agents.

Core rules:

• One ACTIVE task at all times
• Park before switching tasks
• Scope gate every change outside your current task
• Update progress as you work, not at the end

Key references:

• CLASSIFICATION.md — PARKED vs BLOCKED decision tree, edge cases
• TRAPS.md — Common focus drift traps with concrete mitigations
• ANTI_PATTERNS.md — Substitution Test, silent branching, double-active

→ Full Documentation

Deep Plan

Phased planning with adversarial review for complex features.

Workflow:

1. Understand Scope → 2. Enumerate Gaps (3 lenses) → 3. Draft Roadmap → 4. Adversarial Review → 5. Finalize

Key features:

• Quick Path (3 steps) for simple epics, Full Path (5 phases) for complex ones
• "Outside voice" adversarial review using different model providers
• Problem-fit, resilience, and security analysis in parallel
• Auto-escalation when complexity exceeds thresholds

→ Full Documentation

Investigate

Systematic debugging: trace from symptom to root cause, fix the cause, prove the fix.

Workflow:

1. Root Cause Investigation → 2. Pattern Analysis → 3. Hypothesis Testing → 4. Implementation → 5. Verification

Iron Law: No fixes without root cause investigation first.

Key rules:

• 3-strike rule: if 3 hypotheses fail, escalate
• Minimal diff: fewest files, fewest lines
• Regression test must fail without fix and pass with fix
• Never say "this should fix it" — verify and prove it

→ Full Documentation

Project Structure

project-management/
├── skills/
│   ├── tree-of-work/           # Focus enforcement
│   │   ├── SKILL.md
│   │   ├── README.md
│   │   └── references/         # Classification, traps, anti-patterns
│   ├── deep-plan/              # Feature planning
│   │   ├── SKILL.md
│   │   ├── README.md
│   │   ├── references/         # Phase-specific guides (9 files)
│   │   └── templates/          # Roadmap template
│   └── investigate/            # Debugging methodology
│       ├── SKILL.md
│       ├── README.md
│       └── runtime-bindings.md # Platform-specific glue template
├── .claude-plugin/
│   ├── marketplace.json
│   └── plugin.json
├── README.md
├── CLAUDE.md
└── LICENSE

Documentation

Comprehensive documentation for each skill follows the Diataxis framework — four quadrants serving different reader needs:

Tree of Work

• Tutorial — Walk-through from setup to first working example
• How-To Guide — Task-oriented instructions for common workflows
• Reference — Complete technical description of all concepts
• Explanation — Design rationale and trade-offs

Deep Plan

• Tutorial — Plan a feature from scratch with adversarial review
• How-To Guide — Task-oriented instructions for planning workflows
• Reference — Complete technical description of all phases
• Explanation — Design rationale and trade-offs

Investigate

• Tutorial — Debug a bug from symptom to fix
• How-To Guide — Task-oriented instructions for debugging workflows
• Reference — Complete technical description of all phases
• Explanation — Design rationale and trade-offs

Contributing

See CONTRIBUTING.md for guidelines on how to contribute.

Changelog

See CHANGELOG.md for a list of changes.

License

MIT