Skill Detail

Prove Agent Skills with SkillForge

Use SkillForge to triage, create, test, review, compile, and maintain Claude Code and Codex-ready skills with baseline failures and regression evals.

Templates & WorkflowsClaude Code
Templates & Workflows Claude Code Security Reviewed
⭐ 802 GitHub stars
INSTALL WITH ANY AGENT
npx skills add agentskillexchange/skills --skill prove-agent-skills-with-skillforge Copy
Works best when you want a reusable capability, not another fragile one-off prompt.
At a glance
Tools required
Claude Code skill runtime, Git, Python 3, local skill directories, optional Claude Code hooks and command support
Install & setup
Clone `https://github.com/tripleyak/SkillForge`, copy the repository contents into the Claude Code skills directory as `skillforge`, and optionally copy `commands/skillforge.md` into the Claude Code commands directory for the `/skillforge` command. Run the included Python scripts from the installed skill directory when triaging, testing, or compiling skills.
Author
Tripleyak
Publisher
Open Source Project
Last updated
Jul 30, 2026
Quick brief

Use SkillForge to decide whether a new skill is needed, capture baseline failures, generate a skill from a grounded spec, run with-skill behavioral checks, lint and adversarially review the result, and ship reusable evals for later regression testing. It also supports skill ecosystem checks for trigger collisions, stale references, duplicate skills, budget drift, and cross-runtime compile targets.

How it works

What this skill actually does

What the agent/operator does

When to invoke

Invoke this skill when a user wants a durable agent skill rather than a one-off prompt, especially when duplicate detection, cross-runtime packaging, behavioral proof, trigger tests, or ongoing skill ecosystem health checks matter.

Scope boundary

This is not a generic skill marketplace, prompt-writing guide, or plugin bundle. Keep the workflow bounded to SkillForge’s evidence-driven skill creation and maintenance pipeline: triage, RED baseline, spec, generation, GREEN testing, lint/review, evals, doctor checks, and optional compile targets. Do not treat it as approval for unrelated skill packs or broad skill hosting claims.