| name | reprompt |
| description | Rewrite a rough request into a sharp, structured prompt before acting on it. Use when the user says "reprompt this", "reprompt", "improve this prompt", "make this prompt better", "structure this request", or hands you a vague/underspecified ask you should sharpen first. Complements the automatic per-turn sharpener (the UserPromptSubmit hook) for explicit, heavier rewrites. Personalizes from ~/.stellar/profile.md when present. |
What this skill does
Turns a rough ask into a tight working spec, then acts on it — the explicit,
on-demand counterpart to the loop's automatic per-turn sharpener. Use it when a
request is vague enough to be worth structuring out loud, or when the user
literally asks you to "reprompt this".
How to run it
1. Capture the raw request
Take the text the user wants sharpened (everything after "reprompt this:", or
the preceding message if they just say "reprompt that").
2. Load any learned profile
If ~/.stellar/profile.md exists (the optimizer distills it from real usage),
read it and treat it as defaults — domain, typical scope, preferred stack,
tone — unless the request contradicts them. Honor STELLAR_HOME if set.
3. Produce the sharpened spec
Rewrite the request into this structure. Keep it tight; don't pad.
Goal: the outcome actually wanted (not the literal words)
Context: relevant assumptions, environment, prior decisions
Task: the smallest precise action(s) that achieve the goal
Constraints: scope limits, must-nots, non-negotiables
Success: how we'll know it's done right
4. Gate on ambiguity
If a genuine ambiguity would change the result, ask 1–3 sharp questions first
(one at a time) and fill the spec from the answers. If it's clear enough,
proceed directly.
5. Act
Show the sharpened spec briefly, then execute on it (or hand off to the right
stellar-build skill — e.g. find-stellar-idea, nicole-pm, tyler-architect).
Don't make the user re-approve unless a clarifying answer is still pending.
Notes
- This is the manual lever. The automatic version runs every turn via the
UserPromptSubmit hook and needs no invocation; turn it off with
STELLAR_BUILD_NO_REPROMPT=1.
- The more you use stellar-build, the better both get:
/optimize-skills
refreshes ~/.stellar/profile.md from your traces, which feeds these rewrites.
