dotagents #

Shared configuration for my agentic tools. One source of truth for skills, helper scripts, and instructions used across Claude Code, OpenCode, Codex, Cursor, and friends.

Clone the repo wherever you like, then link the parts you want into each tool's config. A change in the repo is picked up everywhere.

Layout #

dotagents/
├── skills/          # Agent Skills, one directory per skill (each with a SKILL.md)
├── helpers/         # Small scripts and utilities shared by agents
├── INSTRUCTIONS.md  # Shared agent instructions, symlinked as AGENTS.md
├── AGENTS.md        # Repo guide for agents
└── README.md        # Repo docs for people

Using it #

Set the path to your checkout, then link the repo into ~/.agents. Point each tool at the hub.

DOTAGENTS=~/dev/personal/dotagents

# Link the repo into the shared hub.
ln -s "$DOTAGENTS/skills" ~/.agents/skills
ln -s "$DOTAGENTS/INSTRUCTIONS.md" ~/.agents/AGENTS.md

# Then link the hub into each tool.
ln -s ~/.agents/skills ~/.claude/skills
ln -s ~/.agents/AGENTS.md ~/.config/opencode/AGENTS.md

Tools that support the Agent Skills layout can consume skills/ directly; others can reference individual files.

Adding a skill #

1. Create skills/<skill-name>/SKILL.md.
2. Add frontmatter with a name and a short description that tells the agent when to use it.
3. Put supporting scripts or references next to it, or under a files/ subdirectory when there are several.

Skills should be small, self-contained, and safe to share between tools. Avoid tool-specific assumptions where a generic approach will do.

For a broad topic, prefer one skill whose SKILL.md acts as a router. The router maps the request to a playbook under files/, and the agent loads only that part. seo and marketing follow this shape.

Attributions #

Most skills are adapted from open-source collections. Where a skill was renamed, the upstream name is in parentheses.

Source	License	Skills
mattpocock/skills	MIT	code-review, tdd, codebase-design, create-tickets (to-tickets), create-spec (to-spec), implement, research, triage
obra/superpowers	MIT	brainstorming, finish-branch (finishing-a-development-branch), parallel-work (dispatching-parallel-agents)
pbakaus/impeccable	Apache-2.0	impeccable
Leonxlnx/taste-skill	MIT	taste (design-taste-frontend)
getsentry/skills	Apache-2.0	blog-writing (blog-writing-guide)
juliusbrussee/caveman	Apache-2.0	caveman
miqdadbadjuber/anti-slop	MIT	antislop, antislop-ui, antislop-copywriting, antislop-human, antislop-layoutmobile, antislop-code
ChromeDevTools/chrome-devtools-mcp	Apache-2.0	chrome-devtools, chrome-devtools-cli, chrome-a11y-debugging, chrome-cookie-debugging, chrome-debug-optimize-lcp, chrome-memory-leak-debugging
DietrichGebert/ponytail	MIT	ponytail, ponytail-audit, ponytail-debt, ponytail-gain, ponytail-help, ponytail-review

The Ponytail skills were imported unchanged from upstream commit 09c434ab73ca43761c5d8b0c5d341926c2fba710, with a license copy in each skill directory. Plugin features described in ponytail-help, such as automatic activation and the status band, require the upstream plugin. The benchmark sources named in ponytail-gain are in the upstream repository.

Each skill's text is upstream apart from local adaptation: renamed frontmatter, reference docs moved under files/, path fixes, and dropped tool-specific metadata (agents/openai.yaml, authoring-only evals/). Skills not listed here were added before this section existed.

The seo and marketing skills are written for this repo, not imported.