Skill Forge

Every instruction your AI reads cites its source: a file and line at a pinned commit, or a documentation URL.

The problem

AI agents hallucinate API calls. They invent function names, guess parameter types, and produce code that doesn’t compile.

The fix

Skill Forge reads a library’s source code and docs and writes a skill: an instruction file (SKILL.md) your AI agent loads before it writes code. Every function signature, parameter type, and usage pattern in it comes with a receipt. The receipt points back to where it came from: a file and line at a pinned commit in the upstream repository, or the documentation page it was read from.

A receipt looks like [AST:cognee/api/v1/search/search.py:L27] ✓

When SKF can’t cite a source, it leaves the instruction out. The main exception is Quick Skill (QS), a fast, best-effort draft with no receipt on each instruction.

Verify any claim in 60 seconds →

How SKF compares

Approach	What it does well	Where it falls short
Skill scaffolding (`npx skills init`)	Creates a SKILL.md template in the standard skill format	The template holds only placeholder text, so you still write every instruction by hand
LLM summarization	Understands context and intent	Generates plausible-sounding content that may not match the actual API
RAG (retrieval) or pasting code into the prompt	Finds relevant code snippets	Hands over loose fragments, not a finished skill your agent can follow
Manual authoring	High initial accuracy	Drifts as the source code changes, doesn’t scale across dependencies
IDE built-in context (Copilot, Cursor)	Convenient, zero setup	Uses generic training data, not your project’s specific integration patterns
Skill Forge	Every instruction cites its source: an upstream `file:line` at a pinned commit, or a doc URL for docs-only skills. You can check any claim in 60 seconds.	How deep it digs depends on which analysis tools you have installed. SKF detects them and picks a tier for you: Quick, Forge, Forge+ or Deep.

Quick install

Requires Node.js >= 22, Python >= 3.11, and uv, the tool SKF uses to run its Python scripts.

npx bmad-module-skill-forge install

Then open your AI coding tool (Claude Code, Cursor, GitHub Copilot and others) and start Ferris, the Skill Forge agent the installer set up there: type /skf-forger ($skf-forger in Codex), or ask your tool to talk to Ferris. Each line below is a request to Ferris, written as @Ferris <code>. Generate your first skill:

@Ferris SF                         # Set up your forge (once per project)
@Ferris forge-auto <repo-or-url>   # One command: a repo or doc URL in, a tested skill out
@Ferris QS <package>               # Or a quick best-effort draft from a package name

forge-auto is the recommended starting point. Once SF has run, it takes one command and asks you nothing along the way. It scopes, compiles, tests, and exports the skill, then adds it to your agent’s context file (such as CLAUDE.md or AGENTS.md). See Getting Started for platform support, the tools that decide your tier, and troubleshooting.