Turn a screen recording into a langchain agent skill

Deterministic playbooks for langchain agents

langchain agents reason well but drift on multi-step UI tasks. A SKILL.md pins the exact steps so the agent stays on rails.

How

1. Record at skillforge.expert/app
2. Get a SKILL.md
3. Feed it to your langchain agent as a structured plan

A worked example

Say your langchain agent needs to file an expense in your finance portal. Left to free reasoning it might click the wrong menu or skip a required field. Instead:

1. Record yourself filing one expense end to end — open the portal, choose the category, enter the amount and date, attach the receipt, submit.
2. SkillForge turns it into a SKILL.md with the amount, date, and category marked as inputs.
3. Load the SKILL.md into the agent's context (or a retrieval step) as the canonical plan; the agent follows it instead of improvising.

Where it fits in a langchain stack

The SKILL.md is the plan; your langchain tools and the LLM are the executor. Use it as a planning document the agent references, so multi-step UI work becomes repeatable rather than probabilistic.

Related guides

Compatible with autogen, crewAI, browser-use, Claude Skills, and Gemini. See also what SKILL.md is.

FAQ

Does SkillForge replace langchain tools?
It complements them — SKILL.md describes the multi-step workflow; your existing tools still execute individual actions.

How do I give the SKILL.md to a langchain agent?
Include it in the agent prompt/context as a plan, or load it via a retrieval step — it is plain Markdown, so any ingestion path works.

Is it free?
Recording and AI extraction are free; new accounts get 15 credits and downloads cost 5 credits.