skill-creator — Setup Guide¶

Source: anthropics/skills (317,700 installs) Category: Skill Development Languages: Python

Anthropic's official skill creation framework. Guides agents through the complete skill authoring lifecycle: scoping the workflow, writing SKILL.md with correct YAML frontmatter, testing with dry runs, handling error paths, and publishing to marketplaces. The same framework used internally by Anthropic for their 60+ published skills.

Installation¶

npx skills add anthropics/skills@skill-creator

Verify:

npx skills list | grep skill-creator

Prerequisites¶

Requirement	Details
Hermes Agent	Any version
Text editor	Any — skill files are Markdown
Git	For version control and publishing
GitHub account	For publishing to skills.sh marketplace

Skill Creation Workflow¶

The framework guides agents through 7 stages:

Stage 1: Scope the Workflow¶

skill-creator scope "Automatically check all social media platforms for new comments and reply helpfully"

The scoping phase identifies: - Inputs: What triggers the skill, what data it receives - Outputs: What the skill produces, success criteria - Tools needed: MCP connectors, terminal access, browser, APIs - Error surface: What can fail, how to handle each failure - Granularity check: Is this one skill or should it be split?

Stage 2: Generate SKILL.md Skeleton¶

skill-creator generate --name "social-comment-monitor" --category "social-media"

Generates a complete SKILL.md template with all required sections:

---
name: social-comment-monitor
description: Monitor and respond to social media comments across platforms
trigger: When checking for new social media engagement
category: social-media
---

# Social Comment Monitor

## Prerequisites
...

## Execution Steps
1. ...
2. ...

## Verification Gates
...

## Error Recovery
...

Stage 3: Define Tool Manifest¶

The tool manifest declares every tool, API, and connector the skill requires:

# In SKILL.md frontmatter
tools:
  - name: terminal
    required: true
  - name: web_extract
    required: false
    fallback: curl
  - name: xurl
    required: true
    version: ">=2.0"

Stage 4: Write Error Recovery Patterns¶

The framework includes a pattern library for common failures:

Failure Type	Pattern	Example
Timeout	Exponential backoff, max 3 retries	API call >30s
Rate limit (429)	Wait for `Retry-After` header, then retry once	Instagram feedback_required
Auth expired (401)	Refresh token, retry once; if still fails, escalate	Gmail OAuth
Missing data	Continue with partial results, flag gaps	API returns empty array
Network error	Retry 3x with 2s delay, then skip and report	Connection reset

Stage 5: Add Verification Gates¶

Every major step gets a verification gate:

## Verification Gates

### After Step 2: Platform Connection
- [ ] xurl whoami returns valid user ID
- [ ] Instagram session loads without LoginRequired
- [ ] Postiz posts:list returns non-empty response

### After Step 4: Comment Collection
- [ ] Comment count > 0 for at least one platform
- [ ] No platform returned error on all comment calls
- [ ] Spam filter processed all comments

Stage 6: Dry Run Testing¶

skill-creator test social-comment-monitor --dry-run

The dry run: 1. Validates YAML frontmatter 2. Checks all declared tools are available 3. Simulates execution with mock tool responses 4. Verifies error paths are handled 5. Reports missing verification gates

Stage 7: Publish to Marketplace¶

skill-creator publish social-comment-monitor --marketplace skills.sh

Publishing handles: - Git tag and version bump - Marketplace metadata formatting - GitHub release creation - skills.sh registry submission

Hermes/CorpusIQ Relevance¶

Quality Standardization: Replaces ad-hoc skill authoring with a structured framework. Every CorpusIQ skill now follows the same 7-stage creation process, ensuring consistency across 133+ skills.

Error Handling: The pattern library alone prevents the "skill breaks silently in production" failures. Instead of discovering error paths through production incidents, skills are tested against every failure type before deployment.

Autonomous Skill Creation: Hermes agents can create new skills autonomously — scope the workflow, generate the skeleton, write error handling, run dry tests, and publish — all without human intervention. The verification gates ensure quality.

Integration with find-skills: After publishing, use find-skills to verify the skill appears correctly in marketplaces and track install counts.