Skillforge
Y Combinator Summer 2026 — applying for Tom Blomfield's RFS

Your agents got smart.
Your company didn't.

Skillforge turns the knowledge buried in your Slack, Notion, GitHub, and Intercom into executable skills your AI agents can actually run. The compiler from company history to SKILL.md.

Slack

#cs-policies — refund cap update, Apr 8

Notion

Refund Policy v3 (current)

GitHub

PM-2026-02-14 post-mortem

Intercom

TICK-4488: Lumen Labs refund

compile

skills/handling-refund-requests/SKILL.md

No signup required for the demo. Real Claude API. Real streaming.

0

Source documents

ingested across 4 platforms

0

Structured facts

with provenance and confidence

0

Executable skills

in Claude Agent Skills format

0 days

From idea to live demo

built solo, end to end

The problem

AI agents have learned everything except your company.

Drop Claude or GPT into a real workflow and the wall arrives fast. The model is brilliant. It has no idea how your company works.

Your refund policy lives in a Notion doc. The exception that killed the last deal lives in a Slack thread. The reason you don't restart payment pods lives in a post-mortem nobody re-reads. So agents stay generic and humans stay copy-pasting.

2 years of Slack

The decision matrix nobody wrote down.

100s of Notion pages

Half are out of date. Half contradict the others.

GitHub post-mortems

The hard rules learned the hard way.

One engineer's brain

The dependency that walks out the door at 5pm.

How it works

Knowledge in. Executable skills out.

Three steps from a mess of company history to skills any agent can run.

STEP 01

Connect what you already have

Slack, Notion, Google Drive, GitHub, Intercom, Linear. Read-only, OAuth-scoped, audited. Your IT will say yes.

STEP 02

Extract structured facts

Our pipeline pulls out policies, decisions, hard rules, and tribal knowledge — with provenance. Every fact traces back to its source.

STEP 03

Generate executable skills

Synthesize into Claude Agent Skills format. Each skill is a runnable playbook your agents pick up automatically — no prompt engineering.

live compiler

Or watch the compiler work in front of you.

Paste a Slack thread, a Notion page, a post-mortem. Skillforge extracts structured facts and compiles a SKILL.md file in real time. Real Claude calls. No demo recordings.

forge / output

→ Extracting facts...

[constraint] Deploy freezes apply to prod, not staging.

[constraint] Feature flag flips on net-new features count as deploys.

[procedure] Hotfixes route through Priya or Ravi.

→ Compiling skill...

name: handling-deploy-freeze

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

open source

Or skip the SaaS.
Run it from your terminal.

The Skillforge compiler is open-source. One command from any text to a SKILL.md file your agents can run. MIT-licensed. Bring your own Anthropic API key.

terminal

$ npx @skforge/cli extract slack-thread.txt

✦ Extracting facts…

· [constraint] Deploy freezes apply to prod, not staging.

· [constraint] Feature flag flips count as deploys.

· [procedure] Hotfixes route through Priya or Ravi.

✦ Compiling SKILL.md…

✦ Done · 5 facts → handling-deploy-freezes in 8.7s

---

name: handling-deploy-freezes

description: Decides…

The artifact

Not another chatbot. A skill the agent runs.

Each piece of company knowledge becomes a SKILL.md file in Claude's Agent Skills format. Versioned. Reviewable. Executable across Claude Code, the API, and claude.ai — anywhere your agents run.

  • Provenance, not vibes

    Every fact traces to a Slack message, doc, or PR. Click through to verify.

  • Stays current

    When the source changes, the brain re-extracts and republishes. No quarterly cleanups.

  • Hard rules survive

    The 'we don't restart payments pods' rule outlives the engineer who learned it the hard way.

skills/responding-to-payment-incidents/SKILL.md

---
name: responding-to-payment-incidents
description: Diagnoses payment service incidents and chooses safe mitigations consistent with Northwind's hard rules learned from pas...
---

Responding to Payment Incidents

This skill encodes Northwind's hard-won rules for handling payments-svc incidents. It exists because in April 2025 we caused a customer-facing double-charge incident by doing the obvious-but-wrong thing.

When to use this skill

You are on-call (or assisting on-call) and:

  • payments-svc P95 is elevated.
  • Stripe webhook queue depth is climbing.
  • Charges or refunds are failing or duplicating.
  • PagerDuty fired a payments-related alert.

The hard rule (read first)

Do NOT restart payments-svc pods to clear queue depth or relieve latency. Stripe has already received ack for in-flight webhooks. Restarting causes the new pods to re-process them, leading to duplicate charges.

This caused PM-2025-04-15 (40 customers double-charged, ~$28k in refunds, trust damage). It is the canonical reference for this rule. Every junior engineer should be told this on day one.

Instead: scale out horizontally (bump HPA max). Stripe's retry behavior tolerates delay; it does not tolerate duplicate processing.

Severity decision

  • SEV1: Customer-facing outage of payment processing, data loss, security incident.
  • SEV2: Latency 2x baseline+, partial degradation. Most queue-backup events.
  • SEV3: Cosmetic. No customer impact.

Why we'll win

Skills are the right unit of value.

Glean and Notion AI sell answers. We sell skills. The difference is everything when the buyer is an agent, not a human.

Output is executable

A SKILL.md file drops into ~/.claude/skills/. The agent picks it up automatically. Search products produce text; we produce primitives.

Provenance is first-class

Every fact is traceable to its source — the Slack message, the doc, the PR. Hallucinations aren't a model problem; they're a sourcing problem.

Time is a primary axis

We track when an announcement supersedes a doc, when v3 supersedes v2. Most knowledge tools treat all data as equal; we treat time as load-bearing.

Skills are reviewable

Markdown with frontmatter. Engineering can code-review. Compliance can sign off. Customer Success owns the refund skill. Adoption inside large companies becomes possible.

Why now

The model isn't the bottleneck anymore. The knowledge is.

“The biggest blocker to AI automation of companies is no longer the models — they just got so good so quickly. Now the blocker is the domain knowledge. We need a new primitive: a company brain. A system that pulls knowledge out of fragmented sources, structures it, keeps it current, and turns it into an executable skills file for AI.”
— Tom Blomfield, YC Group Partner. Summer 2026 RFS.

He's right. We're building it →

Get early access

We're onboarding teams that have already shipped agents and hit the knowledge wall. Tell us about yours.

Want to talk to the founder? ahmedtariqcs@gmail.com