Ceo Train

Mode 2: Conversational Correction

During a session, the user corrects the CEO's behavior:

User: "No, don't post the review comment yet — check if CI has finished first"

The CEO detects this is a correction and offers to record it:

CEO: "Got it — I'll check CI status before posting comments. Want me to add this as a training rule?"
User: "yes"

Steps:

1. Distill the correction into a concise rule (strip conversational context).
2. Follow Mode 1 steps 2-6.

Mode 3: Playbook Update

If a correction changes a workflow step (not just adding a rule):

User: "Add a CI check step before posting comments in the PR review playbook"

Steps:

1. Read the relevant playbook from $VAULT/CEO/playbooks/.
2. Propose the edit — show the current step and the proposed change.
3. Wait for user approval.
4. Apply the edit.
5. Update last_updated in frontmatter.
6. Confirm: "Updated playbooks/<name>.md: <description of change>"

Mode 4: New Playbook

If the user wants to create a new playbook:

/ceo:train create playbook for deployment checks

Steps:

1. Ask the user to describe the workflow steps.
2. Check $VAULT/CEO/SKILLS.md — if a task type with this name already exists, tell the user and ask if they want to update the existing playbook instead.
3. Write to $VAULT/CEO/playbooks/<name>.md with numbered steps.
4. Add an entry to $VAULT/CEO/SKILLS.md dispatch table (with status: active).
5. Confirm: "Created playbooks/<name>.md and added to SKILLS.md dispatch table."

Constraints

• Always show the user what was written and where.
• Never silently modify training files or playbooks.
• If the target file doesn't exist, create it with proper frontmatter.
• Keep rules concise — one line per rule, strip conversational filler.