skill-forge_skill

Overview

This skill is a mandatory gate for building, validating, and improving behavioral skills. It provides a unified six-step workflow with automated validation, a CSO description scoring system, and subagent testing to ensure skills are discoverable and reliable before sharing. Invoke this skill first when starting any new skill work to avoid common discovery and structure failures.

How this skill works

The skill inspects a skill package for structure, naming, description quality, and required assets. It runs linting, CSO scoring for description quality, automated subagent tests, and an optional sharing scan to detect secrets or risky content. Results map to a clear pass/fail checklist and actionable fixes to bring the skill into compliance with discovery best practices.

When to use it

• Before creating or publishing any new skill to ensure naming, description, and structure are correct
• When improving an existing skill's discovery rate or quality of description
• Before sharing a skill publicly to run PII/secrets scans and finalize packaging
• When validating a skill's workflow and subagents to confirm expected triggers and behaviors

Best practices

• Define concrete trigger phrases and expected outcomes before development
• Use a strict naming convention (kebab-case, capability form) and keep names synchronized with directories
• Write the description as an imperative gate: include timing (BEFORE/FIRST/MANDATORY), explicit triggers in quotes, and a short method preview
• Keep content concise and domain-specific; move long guides into reference files
• Run linting, description scoring, and subagent tests as a required pre-release sequence

Example use cases

• Initialize a new skill scaffold, create scripts/assets, and validate discovery readiness before deployment
• Scan a skill bundle for exposed emails, secrets, or company terms prior to public release
• Score and iterate on a skill description to increase invocation frequency in search
• Pressure-test workflows with a subagent harness to verify end-to-end behavior and error handling