README.md

Randoneering Agent Resources #

Skills, agents, and pi-flavored agent overrides for OpenCode, pi, and Claude Code.

Two parallel agent sets: portable prompts under agents/ (frontmatter subset name + description) and pi-specific overrides under pi-agent/agents/ (extended frontmatter, harness-targeted). Skill format stays the same across both harnesses.

Skills #

Category	Skills
Core Workflow	using-superpowers, brainstorming, writing-plans, executing-plans
Development Workflow	test-driven-development, subagent-driven-development, dispatching-parallel-agents, using-git-worktrees, finishing-a-development-branch
Quality	systematic-debugging, damage-control, verification-before-completion
Code Review	requesting-code-review, receiving-code-review
Writing	documentation-writing, writing-skills, writing-style, resume-tailor
Compression (caveman family)	caveman (mode), cavecrew (delegation guide), caveman-commit, caveman-compress, caveman-help, caveman-review, caveman-stats
Data	dbt, postgres, snowflake, clickhouse, neon
Stacks	python, nix, automation, flox
Cloud & Infrastructure	cloudflare, hashicorp
Security	trail_of_bits, sentry

Caveman family reduces token use while keeping full technical accuracy. caveman is the mode; cavecrew routes subagent output through compressed presenters; the rest are command-shaped skills (commit messages, code review, memory-file compression, help card, session stats).

Agents — Portable (`agents/`) #

Generic prompts with portable frontmatter (name, description).

Category	Contents
Engineering	agents/engineering — implementation, architecture, security, DevOps, review
Marketing	agents/marketing — content, campaign, paid media, SEO, growth
Product	agents/product — research, prioritization, feedback synthesis
Project Management	agents/project-management — planning, tracking, workflow
Specialized	agents/specialized — sales, finance, legal, domain specialists
Support	agents/support — reporting, compliance, infrastructure, finance
Testing	agents/testing — evidence, API, accessibility, performance

Agents — pi (`pi-agent/agents/`) #

Same filenames, pi-harness overrides (extended frontmatter, tool routing, prompt tweaks that depend on pi features). Use when running inside pi.

Category	Contents
Engineering	pi-agent/agents/engineering
Marketing	pi-agent/agents/marketing
Product	pi-agent/agents/product
Project Management	pi-agent/agents/project-management
Specialized	pi-agent/agents/specialized
Support	pi-agent/agents/support
Testing	pi-agent/agents/testing

Claude Code Bundle (`.claude/`) #

Drop-in assets for Claude Code. The skills/ mirror under .claude/skills/ is a curated subset of the canonical skills/ tree.

Path	Contents
`.claude/CLAUDE.md`	Project instructions for Claude Code
`.claude/commands/`	acceptance-criteria, analyze-permissions, analyze-skills, commit, install, optimize-prompt, optimize-ruleset, optimize-skill, prime, prompt-help, sentient
`.claude/skills/`	Curated subset of skills/ — 23 skills covering core workflow, writing, data, and stacks
`.claude/templates/`	cli-tools, data-engineering, nix-packaging, python-api

Structure #

agents/                  # portable agent prompts (name + description frontmatter)
├── {category}/
│   └── <domain>-<role>.md
│
pi-agent/                # pi-harness agent overrides
└── agents/
    └── {category}/
        └── <domain>-<role>.md
│
skills/                  # SKILL.md-format skills (Agent Skills spec)
└── {category}/
    └── {skill}/
        ├── SKILL.md
        ├── references/  # optional
        └── scripts/     # optional (e.g. caveman-compress)
│
.claude/                 # Claude Code drop-in bundle
├── CLAUDE.md
├── commands/            # slash commands
├── skills/              # skill subset
└── templates/           # starter templates

Usage #

OpenCode: copy skills into .claude/skills/ or reference this repo via AGENTS.md
pi: copy skills into .pi/skills/ or .agents/skills/, or point pi at this repo via its skills setting; use pi-agent/agents/ for pi-tuned prompts
Claude Code: the .claude/ directory is a drop-in bundle — copy or symlink the whole tree, or pick commands/ / skills/ / templates/ individually
Portable agents: files in agents/ work as prompt templates, imported agents, or source material for tool-specific packaging

Configuration #

OpenCode reads AGENTS.md
pi reads AGENTS.md and supports CLAUDE.md
Claude Code reads CLAUDE.md (this repo's lives at .claude/CLAUDE.md)
Use repo-local AGENTS.md for cross-harness shared instructions

Compatibility #

Skills use the Agent Skills SKILL.md structure — all three harnesses consume them
Skills may include optional frontmatter (allowed-tools, compatibility, metadata, user-invocable); unknown fields are ignored
Portable agents keep only name + description; pi overrides extend with harness-specific frontmatter
Runtime features still vary by harness. Skills that mention subagents, todo tools, or platform-specific commands may require harness-specific adaptation