Install Skill from GitHub

Fetches, converts, and installs skills from GitHub repositories into Gemini format under configs/users/<username>/_agents/skills/.

Quick Reference

Target path:  configs/users/<username>/_agents/skills/<skill_name>/SKILL.md
              (configs/ is a SIBLING of google3/, not inside it)

Step 1 — Fetch the Skill

Use the GitHub API or raw URLs to retrieve skill content:

# List directory contents
https://api.github.com/repos/{owner}/{repo}/contents/{path}

# Fetch raw file
https://raw.githubusercontent.com/{owner}/{repo}/{branch}/{path}

For standalone repos (skill IS the repo), use the root: https://api.github.com/repos/{owner}/{repo}/contents/

Step 2 — Identify the Source Format

Step 3 — Convert to Gemini Format

Conversion Rules

Frontmatter Validation

Every SKILL.md MUST have:

---
name: kebab-case-name        # Max 64 chars, gerund preferred
description: >               # Max 1024 chars, third-person
  What the skill does and when to use it. Include trigger
  keywords users might say.
---

Directory Structure

configs/users/<username>/_agents/skills/
└── my_skill/                 # snake_case directory name
    ├── SKILL.md              # Required — entry point
    ├── scripts/              # Optional — executable helpers
    ├── references/           # Optional — supplemental docs
    ├── templates/            # Optional — file templates
    └── resources/            # Optional — static assets

Step 4 — Adapt Content for Google/Gemini Context

Conductor Adaptation

When a skill references ADRs (Architecture Decision Records), adapt: - "ADR" → "Conductor track spec / project guidelines" - "architecture decision records" → "Conductor context (conductor/ directory)" - docs/adr/ paths → conductor/ directory references - "decision log" → "Conductor track plan"

Google3 Adaptation

When a skill references external tools, consider Google equivalents: - git bisect → works in Fig/hg too (adapt syntax) - npm test / jest → blaze test - curl → works as-is - File paths → use google3 conventions when relevant

Hook Conversion

When a source skill uses hooks, convert them to explicit instructions:

Before (Claude hook): json {"event": "PreToolUse", "tools": ["write_to_file"], "command": "scripts/lint-check.sh"}

After (Gemini instruction in SKILL.md): ```markdown

Pre-write Checklist

Before writing any file, run scripts/lint-check.sh to validate the change. ```

Step 5 — Write Files

Create the skill directory and files:

configs/users/<username>/_agents/skills/<skill_name>/SKILL.md
configs/users/<username>/_agents/skills/<skill_name>/scripts/...
configs/users/<username>/_agents/skills/<skill_name>/references/...

[!IMPORTANT] configs/ is a sibling of google3/, not inside it. The full path is: /google/src/cloud/<user>/<workspace>/configs/users/...

Step 6 — Verify

• SKILL.md has valid YAML frontmatter with name and description
• Directory name is snake_case
• name field is kebab-case
• description is third-person, includes trigger keywords
• All referenced scripts/files exist
• No Claude-specific features left unconverted (hooks, commands, etc.)
• Content adapted for Google/Gemini context where relevant