Agent Toolkit

Single source of truth for the coding-agent skills, subagents, and commands used by:

• Pi (~/.pi/agent/)
• Claude Code — personal (~/.claude/)

Everything in the repo is symlinked into each runtime, so a change made once propagates everywhere.

Layout

agent-toolkit/
├── skills/            6 skills (authored or modified; see Attribution below)
├── agents/
│   ├── claude/        9 subagents (Claude format: model + capitalized tools)
│   └── pi/            11 subagents (Pi format: model + lowercase tools + <task> block)
└── commands/
    ├── claude/        15 slash commands (Claude format: argument-hint + $ARGUMENTS)
    └── pi/            23 prompt templates (Pi format: $@)

Runtime symlinks

Repo path	Pi	Claude (personal)
skills/*	~/.pi/agent/skills/*	~/.claude/skills/*
agents/claude/*	—	~/.claude/agents/*
agents/pi/*	~/.pi/agent/agents/*	—
commands/claude/*	—	~/.claude/commands/*
commands/pi/*	~/.pi/agent/prompts/*	—

Skills are symlinked into both runtimes. Agents and commands are split per platform because their frontmatter differs (model names, tool names, $@ vs $ARGUMENTS).

Skills

Skill	Purpose
adversarial-reviewer	Hostile reviewer personas that break the self-review monoculture
delegation-coach	Task framing and delegation strategy
feature-report	Ephemeral HTML proof-of-work report
github	gh CLI operations (PRs, issues, CI, review)
prompt-crafter	Prompt construction and anti-cargo guidance
review-ready	Human-reviewable handoff packet

Attribution

Modified skills are derived from third-party skills:

• adversarial-reviewer — based on ekreloff/adversarial-reviewer (heavily modified)
• github — based on letta-ai/skills tools/github (modified)

Subagents

Agent	Description	Claude	Pi
designer	Frontend design specialist with strong aesthetic direction	designer	designer
developer	Full-stack: scouts, plans, writes code, runs tests, updates docs, manages PRs	developer	developer
documenter	Technical writing for READMEs, architecture docs, and inline documentation	documenter	documenter
planner	Creates implementation plans from context and requirements	planner	planner
qa	Runs tests and strictly verifies system integrity	qa	qa
reviewer	Code review for quality and security	reviewer	reviewer
scout	Fast codebase recon that returns compressed context for handoff	scout	scout
todo	Appends tasks to the roadmap and optionally creates GitHub issues	todo	todo
worker	General-purpose subagent with full capabilities, isolated context	worker	worker
continuous-learning-compactor	Merges duplicate instincts and tightens instinct files	—	continuous-learning-compactor
continuous-learning-evolver	Evolves instinct files from observed tool-usage patterns	—	continuous-learning-evolver

Commands

Command	Description	Claude	Pi
code	Implements code changes from a plan; manages branches and commits	code	code
code-slm	Same as code, on a small language model	—	code-slm
design	Designs and implements frontend interfaces with strong aesthetic direction	design	design
design-slm	Same as design, on a small language model	—	design-slm
explain-codebase	Analyzes the codebase and generates a detailed, linear walkthrough	explain-codebase	explain-codebase
finish-reviewable	Ends the session with a reviewable PR, feature report, and review-ready packet	finish-reviewable	finish-reviewable
implement	Scout gathers context, planner creates a plan, worker implements	implement	implement
implement-and-review	Worker implements, QA tests, reviewer reviews, worker applies feedback	implement-and-review	implement-and-review
issue-pipeline	End-to-end issue pipeline through dev, QA/review, proof report, and review-ready handoff	issue-pipeline	issue-pipeline
land	Updates the project's "Now" status snapshot for fast next-session re-entry	land	—
plan	Creates an implementation plan from current context	plan	plan
plan-slm	Same as plan, on a small language model	—	plan-slm
pr-review	Reviews code for quality and security; posts findings as a PR comment	pr-review	—
qa	Runs test suites and strictly evaluates system integrity without modifying files	qa	qa
qa-slm	Same as qa, on a small language model	—	qa-slm
review	Reviews code for quality and security	—	review
review-no-gh-comment	Reviews code for quality and security, without posting a GitHub comment	—	review-no-gh-comment
review-slm	Same as review, on a small language model	—	review-slm
run-and-maintain-solo	Fully autonomous orchestrator: fetches issues, implements, tests, reviews, prepares PRs	run-and-maintain-solo	run-and-maintain-solo
run-and-maintain-solo-2-0	Revised autonomous orchestrator workflow	—	run-and-maintain-solo-2-0
scout	Fast codebase recon that returns compressed context	scout	scout
scout-and-plan	Scout gathers context, then planner creates an implementation plan	scout-and-plan	scout-and-plan
scout-slm	Same as scout, on a small language model	—	scout-slm
todo	Appends tasks to the roadmap and optionally creates GitHub issues	todo	todo
todo-slm	Same as todo, on a small language model	—	todo-slm

-slm variants run the same workflow on a small language model.

How to update

1. Edit the file here in the repo (skills, or the pi/claude copy for agents/commands).
2. Symlinks pick the change up automatically in every runtime — no copy step.
3. Commit and push.

Conventions

• Keep agent output formats verdict-first (TL;DR at the top), with detail in collapsible <details> blocks and delta-only re-reviews.
• PR bodies stay short (TL;DR → Summary → Related issues → one-line test evidence); the full review card is a single PR comment.

License

Everything in this repository — skills, agents, commands, and documentation — was created with the help of AI and is released into the public domain under The Unlicense. Free for anyone to use for any purpose, with no attribution required and no warranty.