# Engage Picard

> Improve the prompt written before (or after) this command with AI - aim at the goal beyond the goal - then carry it out

- Skill: `bodish-coder/engage-picard` (Agent Skill)
- Install (CLI): `npx get-skill add bodish-coder/engage-picard`
- Raw SKILL.md: https://api.skillmd.com/api/skills/bodish-coder/engage-picard/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: bodish-coder (https://skillmd.com/u/bodish-coder)
- Updated: 2026-11-16
- Page: https://skillmd.com/skills/bodish-coder/engage-picard

---

Engage: rewrite a rough prompt into a strong, more ambitious one, then carry it out.
Adapted from be-ambitious (MIT, RoboNuggets), after Felix Rieseberg's article on ambition for humans with models.

The prompt: the words typed after the command
Empty → the prompt is the user's message just before this command (what they wrote
before typing it). Nothing there either → ask for the prompt in one line and stop.

First decide what you were handed. A prompt to improve (a rewrite is asked for, or
it is meant for another agent or tool) → stop after the rewrite. A task to do →
carry it out. Cannot tell → stop after the rewrite.

1. **Aim higher — better, not bigger.** Find the goal beyond the goal: the outcome
   this task serves. Pick the ambition types that fit (usually one or two); skip a
   type that does not fit. Ambition raises the outcome, the standard or the
   perspectives — never the deliverables, features or side quests.
   - **Altitude** — the task serves an outcome: climb the ladder (do this task →
     remove the whole class of it → keep it from coming back, naming the setup
     that needs), and work the highest rung you can finish and check with the
     access you have.
   - **Depth** — the task should be done really well: start from the rawest
     material, not what was digested for you; use the field's best methods and
     instruments, not its vocabulary; set a specific hard target and treat it
     as reasonable.
   - **Range** — the result must land with people, or the obvious field is stuck:
     name the verbs under the nouns, bring in the fields they call for, each
     doing its real job with its own tools (a subagent each, where available) —
     never role-play. A narrow fix with one right answer takes Depth only.
2. **Rewrite it.** Keep every requirement, name, path, number and constraint exactly,
   and the user's intent. Make the goal and the goal beyond it explicit; add what
   "done" means, the scope (what not to touch), the context the repo already gives
   (read the files and conventions it names - never guess them) and the output
   expected. List the data, tools and access the plan needs, each marked have or
   missing (and what a missing one would unlock); never invent a source, tool or
   result. End with the outside judge: a test, benchmark, reproduction,
   measurement or other check that is not you, and what counts as a pass. Settle
   an ambiguity by its most likely reading; never invent a requirement. A gap
   that would change the result and that the repo cannot settle → ask that one
   question first.
3. **Show it.** At most three short lines first: the goal beyond the goal; the
   types picked and skipped; what is missing. Then the rewrite once, under
   `**Engaged prompt:**`, as a quote block. No other commentary.
4. **Do it.** Treat the engaged prompt as the user's request, in this same turn,
   with every rule of this session still in force (skills, CLAUDE.md / AGENTS.md /
   GEMINI.md, permissions). Start at the rung you picked. For the work itself, ask
   first only when that rung replaces what the user asked for instead of raising
   it, spends money, or cannot be undone; otherwise no stop for approval. Report
   honestly against the judge: a missed hard target with real numbers is a result;
   a faked pass is not.

