Skip to content

README.md

Latest commit

 

History

History
82 lines (61 loc) · 3.32 KB

README.md

File metadata and controls

82 lines (61 loc) · 3.32 KB
Raw

AgentStation Skills

Reusable Agent Skills for Claude Code, Codex, and compatible hosts.

Skills

Skill What it does Details
goago Runs and remediates the goago restriction-only Go linter. goago/README.md
use-modern-go Applies target-version modern Go guidance and active goago policy to Go changes and reviews. use-modern-go/README.md
autoreview Runs an isolated second-model code review at a pre-PR gate or a configured checkpoint. autoreview/README.md
plans Keeps a durable plan that owns one outcome, with a status ledger, verifiable task criteria, and autonomous execution. plans/README.md
technical-writing Applies controlled, lintable writing rules to developer-facing prose. technical-writing/README.md
starport Operates a self-hosted Starport LLM inference gateway: install, start, catalog questions, and diagnosis. starport/README.md

Install

Each skill installs into the shared ~/.agents/skills directory:

npx skills add agentstation/skills --skill goago -g -a codex -a claude-code -y
npx skills add agentstation/skills --skill use-modern-go -g \
  -a codex -a claude-code -y
npx skills add agentstation/skills --skill autoreview -g -a codex -a claude-code -y
npx skills add agentstation/skills --skill plans -g -a codex -a claude-code -y
npx skills add agentstation/skills --skill technical-writing -g \
  -a codex -a claude-code -a goose -y

The agent flags create the harness links. A manual link for each skill is not required. Compatible harnesses that read ~/.agents/skills use the same canonical copy.

Update every installed global skill with:

npx skills update -g

The global skill lock records source and content metadata. It does not pin a GitHub commit. Review upstream changes before a global update when reproducibility matters.

Install from the GitHub source shown above for normal use. A local-path install helps during development, but it records no updateable GitHub source metadata.

Make a skill unconditional

Installation makes a skill available. It does not make the skill's use unconditional. Add a bootstrap rule to each harness's persistent user instructions:

## Default writing style

Use the installed `technical-writing` skill for all developer-facing prose.
Apply `developer` mode to every response for a technical user. Use `strict`
mode for procedures, safety text, and tightly controlled errors. Preserve
facts, uncertainty, glossary terms, identifiers, commands, code, logs, and
structured data.

Use ~/.codex/AGENTS.md for Codex, ~/.claude/CLAUDE.md for Claude Code, and ~/.config/goose/.goosehints for Goose. Other harnesses need the equivalent persistent user-instruction file.

Repository layout

The repository follows the Agent Skills specification. It uses the <name>/SKILL.md layout at its root. The skills CLI discovers this flat collection. SKILL.md addresses the agent. Each README.md addresses a person. GLOSSARY.md holds the approved terms for this repository.

Validate every skill after a change:

./scripts/validate-skills

See a derivative skill's UPSTREAM.md for its provenance.