capx-ai/casa/casa-compound

casa-compound

The compounding craft. Each unit of work should make the next easier. This captures what was learned, while it is fresh, in a form the next session can find.

Steps

1. If company-brain/profile.json does not exist, tell the founder to run /casa-start

first and stop.

1. Capture while fresh. Identify the lesson: what problem, what was tried, what

actually worked, what did not, and the guidance for next time (repeat this, avoid that). Pull specifics from the session, decisions/, and experiments.jsonl.

1. Check for an existing entry. Search company-brain/learnings.jsonl for an

overlapping lesson. If one exists, update it rather than adding a near-duplicate.

1. Write the learning. Append a structured record to

company-brain/learnings.jsonl with: a short title, the context, what worked, what did not, the reusable guidance, the tags (domain, level, playbook id), and the date. For a substantial lesson, also write a short markdown note under company-brain/playbooks/<slug>.md.

1. Keep it discoverable. If a new domain of learning appeared, make sure the company

CLAUDE.md points future sessions at the store (edit inside the AUTO markers only). A learning no future agent finds did not compound.

1. Link forward. Note in the relevant decisions/ or experiments.jsonl entry that a

learning was recorded, so the thread is traceable.

Rules

• Capture while the context is fresh; a lesson written a week later loses the detail

that made it useful.

• Update over duplicate. One good entry per lesson, not five overlapping ones.
• Every learning is tagged and dated so casa-learnings can retrieve it.
• No em-dashes, no emojis.