make-skill

Make Skill: $ARGUMENTS

Create a new skill from existing prompt file(s).

Step 1: Parse Arguments

Extract from $ARGUMENTS:

• paths: Prompt file paths or glob patterns (e.g., ./prompts/research-*.md, ./prompts/paper-reader.md)
• --name NAME: Optional explicit skill name (otherwise derived from content)
• --global: If present, install to the user-level skills directory for the active platform instead of the repo-local skills directory

Step 2: Read Source Prompts

# If glob pattern, expand it first
ls $PATHS 2>/dev/null || echo "No matches"

Read all matching prompt files.

Step 3: Analyze Patterns

For each prompt, identify:

Fixed Elements (Boilerplate)

• Output format templates
• Section headers
• Standard instructions
• Safety boilerplate (file error workarounds, parallel swarm awareness)

Variable Elements

• Topic/subject (what $ARGUMENTS would replace)
• Input file paths
• Output file paths/names
• Domain-specific content

Workflow Characteristics

• Research/reading: Searches web, reads many files, investigates
• Implementation: Modifies code, creates/edits files
• Audit/review: Reads code, produces reports
• Commit/deploy: Git operations, side effects

Step 4: Determine Frontmatter

Based on analysis, select appropriate settings:

Context & Agent

# If the skill reads many files or does web research:
context: fork
agent: general-purpose

# If the skill is simple/linear with few file reads:
# (omit context - defaults to inline)

Model Invocation

# If skill has side effects (commits, edits code, deploys):
disable-model-invocation: true

# If skill is read-only or research:
# (omit - allows external model calls)

Tool Restrictions

# Read-only research/analysis:
allowed-tools: Read, WebSearch, WebFetch, Glob, Grep

# File creation without code modification:
allowed-tools: Read, Write, Bash(mkdir:*), Bash(mv:*), Bash(rm:*), Bash(ls:*)

# Full implementation work:
# (omit - allows all tools)

Name & Description

name: [derived-from-content-or-explicit]
description: [One sentence describing what skill does and when to use it]
argument-hint: [what arguments look like]

Step 5: Generate Skill Content

Create the SKILL.md with:

1. YAML frontmatter (determined above)

2. Title: # [Skill Name]: $ARGUMENTS

3. Objective section: What the skill accomplishes

4. Steps: Numbered steps extracted from prompts, generalized:

   • Replace hardcoded values with $ARGUMENTS or parameters
   • Keep structural instructions
   • Include decision trees where appropriate

5. Output format: Template for results

6. Safety boilerplate (ALWAYS include):

---

## CRITICAL: File Modified Error Workaround

If Edit/Write fails with "file unexpectedly modified":
1. Read the file again
2. Retry the edit
3. Try path formats: `./relative`, `C:/forward/slashes`, `C:\back\slashes`
4. Prefer your file editing tools over shell text manipulation (cat, sed, echo)
5. If all formats fail, STOP and report

## CRITICAL: Parallel Swarm Awareness

You may be running alongside other agents. NEVER use git restore/checkout/reset/clean.

---

7. Completion instructions:

## Completion

When done, reply ONLY:
\`\`\`
Done - [brief description of output location]
\`\`\`

Do NOT:
- Output findings to conversation
- Modify files outside scope
- Leave temporary files behind

Step 6: Determine Location

# Project-specific (default)
mkdir -p "./.agents/skills/[skill-name]"
# Output: ./.agents/skills/[skill-name]/SKILL.md

# Global (if --global flag)
# Codex: ~/.codex/skills/[skill-name]
# Claude/Gemini: use that platform's user-level skills directory

Step 7: Write Skill File

Write the generated SKILL.md to the appropriate location.

Step 8: Show Summary

Present to user:

## Skill Created

**Name:** [skill-name]
**Location:** [path to SKILL.md]
**Description:** [description]

### Frontmatter Settings
- context: [value or "default (inline)"]
- agent: [value or "default"]
- disable-model-invocation: [true/false]
- allowed-tools: [list or "all"]

### Pattern Analysis
**Fixed elements:** [what stays constant]
**Variable elements:** [what $ARGUMENTS replaces]
**Source prompts:** [list of files analyzed]

### Usage

$[skill-name] [example-arguments]

**Confirm creation? [Y/n]**

Wait for user confirmation before finalizing. If user says no or requests changes, iterate.

---

CRITICAL: File Modified Error Workaround

If Edit/Write fails with "file unexpectedly modified":

1. Read the file again
2. Retry the edit
3. Try path formats: ./relative, C:/forward/slashes, C:\back\slashes
4. Prefer your file editing tools over shell text manipulation (cat, sed, echo)
5. If all formats fail, STOP and report

CRITICAL: Parallel Swarm Awareness

You may be running alongside other agents. NEVER use git restore/checkout/reset/clean.

---

Completion

When user confirms, reply:

Done - created [path to SKILL.md]

Do NOT:

• Create skills without showing summary first
• Overwrite existing skills without warning
• Create documentation files beyond SKILL.md