Skill Creator

Skill Creator — Create, Lint & Package Skills

DITEMPA BUKAN DIBERI — Skills are forged through intent, lint, and seal.

What This Skill Is

A unified skill covering three concerns:

1. Create — Bootstrap, design, and package new skills from user intent
2. Lint — Check trigger clauses for collisions, missing negatives, and vague verbs
3. Scaffold — Interactively create SKILL.md files with proper frontmatter

When to Use

• Capturing user intent for a brand new skill from conversation history
• Conducting interviews to identify edge cases, input/output requirements, and dependencies
• Drafting the initial structure, frontmatter, and instructions for a new SKILL.md
• Validating frontmatter descriptions and target boundaries of any skill
• Trigger drift is observed (e.g. wrangler activated for generic coding)
• Linting a newly forged skill prior to production staging
• Packaging finished skill folders into distributables
• User wants to create a skill, scaffold a skill, or runs /create-skill

When NOT to Use

• Auditing the active skill portfolio for collisions or stale documentation (use skill-inventory)
• Running quantitative benchmarks or evaluating prompt traces
• Performing programmatic runtime execution benchmark testing

§1. CREATE — Bootstrap & Design

Procedure

1. Capture Intent: Analyze the user request and extract target tools, sequence steps, and expected outputs
2. Define Trigger Boundaries: Design preliminary "Use when" and "Do not use when" rules
3. Draft SKILL.md: Compose the frontmatter, purpose, triggers, inputs, procedure, postconditions, failure modes, and telemetry structure
4. Verification Setup: Propose 2-3 realistic test prompts for verification
5. Floor-Tier SEAL Gate (MANDATORY): Classify the drafted skill by primary floor touch
6. Package Skill: Run scripts/package_skill.py to compile the folder structure

Floor-Tier SEAL Gate

Grader output requirement: pass_rate ≥ 0.95 for any SEAL path. Skills with pass_rate < 0.95 → return to draft.

§2. LINT — Trigger Validation

Lint Level Classification

• L1: Stylistic (Wording Clarity):

  • Descriptions slightly wordy or triggers phrased passively.
  • Result: Warning issued; suggestion provided.

• L2: Behavioral Risk (Missing Negatives):

  • Skill lacks a robust "Do not use when" block with at least 3 exclusion boundaries.
  • Result: Block validation; requires prompt adjustments.

• L3: Safety Risk (Constitutional Breaches):

  • Skill performs irreversible state writes, manages sensitive secrets, or coordinates destructive operations without declaring an integration path to arifos-governance and 888_HOLD gates.
  • Result: HARD BLOCK. Requires arifos-governance human release (888 HOLD) to override.

Lint Procedure

1. Strict Trigger Parsing: Extract the Use When and Do Not Use When blocks
2. Vague Verb Scanning: Scan for banned words: "help", "assist", "improve", "manage", "optimize", "support". Require explicit action verbs.
3. Trigger Count Validation: Use when must have 3-7 concrete triggers; Do not use when must have ≥3 exclusion rules
4. Lint Level Classification: Assess for L1, L2, or L3 violations
5. 888 HOLD Enforcement: If L3 violation found, lock the build pipeline, generate ERR_LINT_L3_VIOLATION ticket
6. Lint Status Output: Output failures, warnings, and recommended revisions with exact line references

§3. SCAFFOLD — Interactive Creation

Step 1: Gather Information

Ask the user one at a time:

1. Skill name — lowercase letters, digits, hyphens only. 2-64 characters.
2. Scope — Project (<repo-root>/.grok/skills/<name>/SKILL.md) or User (~/.grok/skills/<name>/SKILL.md)
3. What it should do — describe the workflow, paste an example prompt, or explain the task

Step 2: Draft Description

Write a description frontmatter value that includes:

• What the skill does (1-2 sentences)
• Trigger phrases and keywords for auto-invocation
• The slash command name

Step 3: Create Directory & Write SKILL.md

mkdir -p <SKILL_DIR>

Write SKILL.md following the exact format:

---
name: <skill-name>
description: <the description>
---

<markdown body with instructions, steps, code blocks>

Step 4: Verify and Confirm

1. Verify the file was written correctly
2. Tell the user how to use it: slash command, TUI menu, or automatic invocation

Guidelines

• Keep the SKILL.md body focused and actionable — it is a prompt for the agent, not documentation
• The description field is critical — it controls auto-invocation
• Prefer referencing existing CLI tools over writing custom scripts
• Always use absolute paths when creating files
• Do NOT skip creating the directory

Postconditions

1. A valid skill structure (including SKILL.md) is staged in the target folder
2. The skill complies with baseline structural layout standards
3. The Floor-Tier SEAL Gate was observed
4. All trigger clauses pass L1-L3 compliance

Failure Modes & Escalation

• Intent Ambiguity: User request too vague. Action: Pause and present clarifying questions.
• Missing Exclusion Block: Skill lacks "Do Not Use When" section. Action: Flag L2, suggest 3 templates, block merging.
• Verbal Overload: Triggers in vague prose. Action: Auto-format into short, bulleted imperative clauses.

Telemetry per Run

{
  "skill_name": "skill-creator",
  "version": "2.0.0",
  "floor_tier_reached": "<F1|F2|F5>",
  "sealing_authority": "<888|human|autonomous>",
  "grader_pass_rate": 0.0,
  "lint_level": "<L1|L2|L3|PASS>",
  "human_approval_required": false,
  "hold_code": "<if applicable>"
}