| name | skill-forge |
| description | Skill creator and optimizer following Agent Skills Open Standard. Triggers on skill create, make skill, skill forge, new skill, skill audit, skill optimize, SKILL.md. |
Skill Forge
Meta-skill for creating and optimizing Agent Skills. Generates production-grade SKILL.md files following the Agent Skills Open Standard.
Trigger
- skill create, make skill, new skill, skill forge, skill builder
- skill audit, skill optimize, skill review
- SKILL.md, slash command, custom command
Instructions
When the user requests skill creation or optimization, execute the pipeline below.
Phase 1: Requirements Gathering
Extract the following from user input. Use AskUserQuestion to fill gaps.
Required:
| Field | Question | Default |
|---|
| Skill name | What name? (lowercase + hyphens, max 64 chars) | Infer from input |
| Purpose | What does this skill do? | — |
| Triggers | When should it activate? | Infer from purpose |
| Scope | Project-local vs global? | Project (.agents/skills/) |
Optional (advanced):
| Field | Description | Default |
|---|
allowed-tools | Restrict which tools the skill can use | Unset (all allowed) |
model | Override model when skill activates | Unset |
disable-model-invocation | Prevent automatic invocation | false |
Phase 2: Architecture Selection
Auto-select the optimal pattern based on requirements.
| Pattern | Use case | Body size |
|---|
| CLI Wrapper | External CLI tool invocation | ~50 lines |
| MCP Tool Wrapper | MCP server tool orchestration | ~80 lines |
| Multi-Phase Pipeline | Sequential multi-step workflow | ~150 lines |
| Research / Analysis | Web search + analysis + report | ~100 lines |
| File Generator | Template-based file output | ~120 lines |
| Code Review / Audit | Code analysis + feedback | ~80 lines |
Selection criteria:
- External tool calls only → CLI Wrapper
- MCP tools only → MCP Tool Wrapper
- 3+ sequential steps → Multi-Phase Pipeline
- Web search / research involved → Research / Analysis
- File artifacts generated → File Generator
- Code / document analysis → Code Review / Audit
Phase 3: SKILL.md Generation
3.1 Frontmatter Spec
---
name: {skill-name}
description: {when to use + what it does}
---
Description writing rules:
- "helps with code" (BAD) → "when tests fail, analyzes failures and suggests fixes" (GOOD)
- Include trigger keywords naturally
- One sentence: when to use + what it does
3.2 Body Structure
# {Skill Display Name}
{One-line summary}
## Trigger
- {trigger keywords / patterns}
## Instructions
{Execution instructions}
Phase 4: Quality Validation
Structural Validation:
name field present, lowercase+hyphens, max 64 chars
description field present, max 1024 chars, includes trigger keywords
## Trigger section present, 3+ keywords
## Instructions section present
- Body 200 lines or fewer
- CLI commands use relative names (on PATH) or project-relative paths
Quality Validation:
- Description is specific enough
- Single responsibility — skill doesn't do too much
- Error handling guidance at necessary points
Phase 5: File Creation
- Write file to
.agents/skills/{skill-name}/SKILL.md
- Create scripts/, references/ subfiles if needed
Execution Modes
/skill-forge create → Phase 1–5 (create new)
/skill-forge audit → Audit existing skills
/skill-forge optimize → Optimize existing skills
Best Practices
Do
- Keep body under 200 lines
- Include trigger keywords in description
- Skillify when you repeat the same context 2+ times per week
- Minimize frontmatter — name + description suffice
Don't
- Make one skill do too much
- Write vague descriptions
- Put global rules in skills (belongs in AGENTS.md)
- Write 200+ line monolithic SKILL.md files