Skills tell the agent what it can do. Agent KBase holds what is true, what you shipped, and what the model must not invent. Export a folder your Cursor agent loads before every task.
STORY
SkillTree interviews and skills are one layer. Agent KBase is the company knowledge base those agents read: offers, constraints, voice, and what must stay [CONFIRM].
That is the difference between an AI org that shares ground truth and a folder of generic prompts.
Run `brain init` or open the web wizard. Same pack from CLI and browser.
Step through Project, Facts, Story, and Gaps. Paste a public URL to draft from the landing page, then edit like an interview transcript.
Validation passes → export `brain.zip`. Facts, stories, AGENTS.md, and open questions become the node at the center of the tree.
Cursor loads `brain/` before tasks. SkillTree skills sit on the SKILLS spoke: how to work, while Agent KBase holds what is true.
AUDIT MOTION
Paste a public URL in the wizard. The extract engine reads headings and body copy, then places draft facts on the FACTS spoke and gaps on GAPS. You review before export.
Full stack audits and revenue sized engagements are SkillTree's audit product. Agent KBase shows the same map logic for open source: scan → plan on the graph → human sign off.
Extract from website →Example map (Auditamia demo leaves)