Skills¶
A Skill is a Markdown file the model can invoke on demand via the Skill tool. Skills live in ~/.anthrogo/skills/ (home, all projects) or <cwd>/.anthrogo/skills/ (project-local; overrides same-named home skill).
Directory layout¶
~/.anthrogo/skills/<name>/
├── SKILL.md # required — frontmatter + instructions
├── scripts/ # optional — model reads via Read tool
└── references/ # optional
SKILL.md format¶
---
name: git-flow
description: Use when starting a new feature branch off main.
---
# git-flow
When the user asks to start a new branch:
1. Run `git checkout main && git pull`
2. Run `git checkout -b feature/<name>`
3. Confirm the branch was created with `git branch --show-current`
name must match the directory name. description is required — it is shown to the model in the system prompt so it can decide when to invoke the skill.
How skills work¶
anthrogo lists every loaded skill (name + description) in the system prompt at startup. The model picks one, calls the Skill tool with {"skill": "git-flow"}, and receives the full SKILL.md back as a tool result. It then follows the instructions using Read / Bash / Write / etc. — all still gated by the existing permission rules.
Slash commands¶
/skills # list all loaded skills
/skills show <name> # print one skill's SKILL.md
/skills reload # re-scan ~/.anthrogo/skills/ and <cwd>/.anthrogo/skills/
Precedence¶
Project-level <cwd>/.anthrogo/skills/<name>/SKILL.md overrides a same-named home skill. The override is exact-match on the directory name.
Trust¶
The body of a SKILL.md becomes part of the prompt sent to the model when invoked. A malicious skill can instruct the model to leak data, exfiltrate files, or trigger side effects — though every action still flows through anthrogo's tool permission gate. Only install skills from sources you trust.
Example: code-review skill¶
---
name: code-review
description: Use when asked to review code changes or a pull request.
---
# code-review
1. Run `git diff HEAD~1` to see recent changes.
2. Check for: unhandled errors, missing tests, hardcoded secrets, logic bugs.
3. For each issue found: cite the file and line number, explain the problem, suggest a fix.
4. Output a summary section at the end.