create-todo

name	create-todo
description	Triggers on /create-todo only. Writes a single ai_todo mid-session without waiting for /close - for a "note this for later" moment, a deferred fix, or a skill-improvement observation.
argument-hint	<what to defer, in your own words>

/create-todo

Write one ai_todo right now, instead of hoping /close catches it later.

When to use

Any time something surfaces that's worth deferring but not worth stopping the current thread for:

• Joe says "note this for later" / "we should fix this eventually" / "not now, but flag it".
• Claude notices a repeated manual step, a skill that got violated or proved wrong, or a "this deserves its own skill" moment - mid-session, not just at /close time.
• An offer gets made ("want me to do X?") and declined for now but shouldn't be forgotten.

If there's no project (.for_bepy/ has nowhere to live), say so and stop - do not create .for_bepy/ outside a project.

Step 1 - Determine Type

• task - something Claude can execute later (code, config, analysis).
• skill-improvement - a skill gap, a "had to do this differently than the skill said" note, or a "this project keeps needing X, maybe a local skill" observation.

Infer from context; ask only if genuinely ambiguous.

Step 2 - Write the file

Follow ~/.claude/skills/close/ai-todos-format.md exactly: filename (zero-padded id + kebab-case slug, scan .for_bepy/ai_todos/ and done/ for max id), required sections (# title, **Type:**, ## Goal, ## Context, ## Approach, ## Acceptance), and the off-limits rule (never include git instructions).

The bar is the same as /close's: a future cold AI session must be able to execute the task from the file alone. Don't write a placeholder - if there isn't enough to fill Context/Approach meaningfully yet, ask one clarifying question before writing rather than write a thin file.

Step 3 - Confirm

Print the filename and a one-line summary. Do not run /batch-todos or execute the todo - this skill only files it.

Anti-patterns

• Writing a todo for something that requires Joe's physical action (credentials, hardware, browser login) - there's no persistent home for those right now, just say it directly instead of filing a file nobody will read.
• Batching multiple unrelated asks into one file. One todo, one file, per invocation - call /create-todo again for a second item.
• Re-filing something that's already an open todo - skim .for_bepy/ai_todos/*.md titles first; if a near-duplicate exists, say so and skip writing (dedup across the whole set is /batch-todos's job, not this skill's).