Skill Intake Pipeline

README

# skill-intake-pipeline 'Automates the full agent skill acquisition workflow: fetch from URL/repo, quarantine, validate, test, promote, and register into the canonical registry. Use when importing skills from external sources like GitHub repos, websites, or other agent frameworks.'

SKILL.md

---
name: skill-intake-pipeline
description: 'Automates the full agent skill acquisition workflow: fetch from URL/repo, quarantine, validate, test, promote, and register into the canonical registry. Use when importing skills from external sources like GitHub repos, websites, or other agent frameworks.'
metadata:
  source: Self-Evolved via Dynamic Skill Injector
---

# Skill Intake Pipeline

# Skill Intake Pipeline

**Role**: Systematic Agent Skill Acquisition & Delivery

## Core Workflow

### Step 1: Source Identification
Determine the source type:
- **GitHub URL**: Clone or fetch raw SKILL.md via `read_url_content`
- **Website/Docs**: Scrape with `playwright-scraper-agent` or `read_url_content`
- **Local File**: Read directly with `view_file`
- **Existing Repo Skill**: Copy from another local directory

### Step 2: Structural Validation
Before quarantine, validate the skill conforms to Progressive Disclosure:
```
Required:
  ├── SKILL.md (with YAML frontmatter: name, description)
  ├── name: kebab-case only
  ├── description: ~100 tokens, no XML tags
  └── Instructions body: < 5000 tokens

Optional:
  ├── scripts/    (executable helpers)
  ├── references/ (documentation, API specs)
  └── assets/     (templates, JSON data)
```

### Step 3: Deduplication Check
Before injecting, run a semantic check:
1. Extract the skill's `name` and `description`
2. Search existing skills with `grep_search` for the skill name
3. If a duplicate exists, compare descriptions — if >80% similar, ABORT and report
4. If unique, proceed to quarantine

### Step 4: Quarantine Injection
```bash
# Use the inject_skill MCP tool
inject_skill(
  skill_name="<kebab-case-name>",
  description="<extracted description>",
  instructions="<full markdown body>",
  bypass_quarantine=false
)
```

### Step 5: Automated Testing
```bash
# Run quarantine test
test_quarantined_skill(skill_name="<name>")
```
If test fails → repair the SKILL.md → re-test (max 3 attempts)

### Step 6: Promotion
On green test:
```bash
promote_quarantined_skill(
  skill_name="<name>",
  commit_message="[Intake] imported <name> from <source>"
)
```

### Step 7: Registry Update
After promotion, append the new skill entry to `agent-skill-registry/registry.yaml`:
```yaml
- name: <skill-name>
  tier: <appropriate-tier>
  path: shared/<tier>/<skill-name>
  description: <description>
  trust_tier: T2
  status: active
  source: <url-or-path>
  imported_at: <ISO-timestamp>
```

### Step 8: Memory Recording
```bash
record_skill_outcome(
  skills=["skill-intake-pipeline"],
  task_type="skill_import",
  success=true,
  narrative="Imported <name> from <source>"
)
```

## Anti-Patterns
- **Never bypass quarantine** for externally sourced skills
- **Never import without dedup check** — the ecosystem already has 681 skills
- **Never skip the test block** — untested skills cause hallucination
- **Never import skills that require secrets/API keys in the SKILL.md body** — reference them via env vars

## Validation Block
```bash
echo 'skill-intake-pipeline: structural validation passed'
```

## When to Use This Skill

Use when the user says 'import this skill', 'pull this agent config', 'add this skill from GitHub', or when discovering a new skill package from an external source.

## Related Skills

Works well with: `skill-deduplicator`, `skill-version-tracker`, `self-evolution-engine`, `agent-skills-architect`

---

## Skill Metadata

**Created**: 2026-06-03
**Author**: Antigravity Self-Evolution Engine
**Version**: 1.0.0