skillforge

Compile scattered company knowledge — Slack threads, Notion pages, post-mortems — into Claude Agent Skills (SKILL.md files).

The company brain for AI agents, as a CLI.

$ npx @skforge/cli extract slack-thread.txt
✦ Extracting facts…
  · [constraint] Deploy freezes apply to prod, not staging. (95%)
  · [constraint] Feature flag flips on net-new features count as deploys. (92%)
  · [procedure] Hotfixes route through Priya or Ravi during freezes. (90%)
✦ Compiling SKILL.md…
✦ Done · 5 facts → handling-deploy-freeze in 8742ms

---
name: handling-deploy-freeze
description: Decides whether a change can ship during an active deploy freeze...
---

# Handling Deploy Freeze
...

Why this exists

AI agents are smart. Your company isn't documented. The result: agents stay generic and humans stay copy-pasting.

Skillforge takes a Slack thread, a Notion page, a post-mortem — anything with company-specific knowledge — and compiles it into a SKILL.md file in Claude's Agent Skills format. Drop the output into ~/.claude/skills/ and Claude Code uses it automatically. Or upload it via the Skills API.

Install

Run instantly with npx:

npx @skforge/cli extract <file>

Or install globally:

npm install -g @skforge/cli
skforge extract <file>

You'll need an ANTHROPIC_API_KEY:

export ANTHROPIC_API_KEY=sk-ant-...

Get a key at console.anthropic.com. A typical extraction costs ~$0.03.

Usage

Single file → SKILL.md to stdout

skforge extract path/to/notion-doc.md

To a Claude Code skills folder

skforge extract slack-thread.txt -d ~/.claude/skills
# Creates ~/.claude/skills/<skill-name>/SKILL.md

Batch a directory

skforge batch ./company-docs -o ./skills --ext .md,.txt

Walks the input directory, extracts a skill from each file, writes them all to ./skills/<skill-name>/SKILL.md.

Hint the extractor

skforge extract incident.md --hint "A SEV2 incident post-mortem"

Hints help the model orient. They're optional — the extractor figures it out without one most of the time.

Just the facts

skforge extract policy.md --facts-only
# Outputs JSON: { topic, facts: [{ type, statement, confidence, ... }] }

Pipe from stdin

cat slack-thread.txt | skforge extract -

Inspect an input before paying for it

skforge describe my-doc.md
# Prints char/word/line counts and an estimated cost

Commands

| Command | Description | |---------|-------------| | extract <file> | Extract a single SKILL.md (or - for stdin). | | batch <dir> | Extract every supported file in a directory. | | describe <file> | Show input stats and estimated cost. |

Options

-o, --output <path>        Write to a file instead of stdout
-d, --output-dir <dir>     Write to <dir>/<skill-name>/SKILL.md
--hint <text>              Short context hint
--api-key <key>            Override ANTHROPIC_API_KEY
--facts-only               Output extracted facts as JSON
--extract-model <model>    Override extraction model
--synthesis-model <model>  Override synthesis model
--quiet                    Suppress progress logging

How it works

1. Extract. Claude Haiku reads the input and produces structured facts: policies, procedures, decisions, constraints (hard rules), people, systems. Each fact is typed and confidence-scored.
2. Synthesize. Claude Opus takes the facts and writes a SKILL.md file following Anthropic's Agent Skills best practices — gerund-form name, third-person description, ≤500-line body, hard rules called out, examples included.
3. Done. The output is a real Claude Agent Skill. Drop it in ~/.claude/skills/<skill-name>/SKILL.md and Claude Code picks it up.

Cost

A typical single-file extraction uses ~1.5K tokens of Haiku input, ~200 tokens of Haiku output, ~1K tokens of Opus input, and ~1.5K tokens of Opus output. About $0.12 per skill. Run skforge describe for a more precise estimate before spending.

Programmatic API

import { forge } from "@skforge/cli";

const result = await forge(text, {
  apiKey: process.env.ANTHROPIC_API_KEY,
  hint: "A Slack thread about pricing exceptions",
  onProgress: (event) => {
    if (event.type === "fact") {
      console.log(`extracted: ${event.fact.statement}`);
    }
  },
});

console.log(result.skill.name);
console.log(result.skill.body);

What's the hosted product?

This CLI is the open-source compiler. The hosted product at skillforge-provibecodes-projects.vercel.app connects directly to Slack, Notion, Drive, GitHub, Intercom, and Linear; runs the compiler continuously across all your sources; tracks supersession and conflicts across versions; and provides review workflows for cross-functional ownership. Same primitive, hosted plumbing.

License

MIT. Built for Y Combinator Summer 2026 against Tom Blomfield's "Company Brain" RFS.

Links

• Hosted demo: skillforge-provibecodes-projects.vercel.app/demo
• Live extraction: skillforge-provibecodes-projects.vercel.app/forge
• About: skillforge-provibecodes-projects.vercel.app/about
• Issues / PRs: GitHub repo