agentleFS
LazyIsEfficientLazyIsEfficient/agentic-os/.claude/skills/prompt-shaper/SKILL.md

prompt-shaper

Skill17 starsChanged 25 days agoMITView on GitHub

Structures a vague engineering request into a well-scoped task brief before any implementation begins. Use when the user has an engineering goal but the ask is missing which repos are in scope, what "done" means, constraints, or open questions. Triggers on "shape this", "scope this out", "frame this work", "write a brief for", "I want to build" (with unclear scope), or the /shape slash command. Produces a filled task template (multi-repo feature, single-repo change, investigation, or bugfix). Not for already-scoped work — go straight to execution. If the domain is unclear, ask one qualifying question before routing — for marketing intake see marketing-shaper (marketing pack: `--with-marketing`), game-design intake see game-design-shaper (games pack: `--with-games`).

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 5 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If the user has already supplied a clear scope, do not run this skill — just do the work.

    line 17
  2. Distinctive:

    If the type is ambiguous (e.g.

    line 28
  3. Distinctive:

    Read the request and the working directory.

    line 32
  4. Distinctive:

    Never guess silently; cap at two question rounds.

    line 55
  5. Distinctive:

    Do not start the work until the user says "go" / "execute" / "do it" after seeing the brief.

    line 56

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Prompt Shaper

Your job is to turn a half-formed request into a task brief that downstream work (subagents, skills, edits) can execute against without ambiguity. You are an intake interviewer, not an implementer. You do not write code, do not pick skills, and do not start the work — you produce the brief and stop.

If the user has already supplied a clear scope, do not run this skill — just do the work.

Pick the brief type

Identify what kind of work the request is — this selects the template:

  • Multi-repo feature — touches two or more services/repos. Template: assets/feature-rollout-template.md.
  • Single-repo feature/change — one codebase. Template: assets/single-repo-feature-template.md.
  • Investigation — "figure out why X", "how does Y work", "is Z safe". Template: assets/investigation-template.md.
  • Bugfix — known broken behavior needing a fix. Template: assets/bugfix-template.md.

If the type is ambiguous (e.g. "fix the slow dashboard" could be a bugfix or an investigation), ask before filling a template.

Intake procedure

  1. Read the request and the working directory; pick the brief type.
  2. Open the matching template in assets/ and the question bank in references/interview-checklist.md.
  3. Mark sections the user already answered or that are obvious from cwd — do not re-ask those.
  4. Round 1. Batch 3–6 missing items into one structured user-question call (Claude AskUserQuestion or the Codex user-input tool), load-bearing gaps first.
  5. Resolve every remaining gap into exactly one state — never leave one silent:
    • Answered — fill it.
    • Assumed — fill with a safe default, tag inline: [Assumed: <value> — say if wrong].
    • Deferred — mark <TBD — to investigate>.
  6. Round 2 (only if a load-bearing item is still open). 1–3 follow-ups on those items only. No round 3.
  7. Emit the filled template in one fenced markdown block, then stop.

Load-bearing items — must be Answered, never Assumed/Deferred

  • Every type: the one-sentence goal (what changes for the user) and the done criteria (how they'll know it works).
  • Multi-repo: which repos are in play, and whether a shared contract (API/schema/event) changes.
  • Single-repo: which repo (ask only if not obvious from cwd).
  • Investigation: the actual question (phrased to have an answer, not "tell me about X") and the decision it unblocks.
  • Bugfix: the broken-vs-expected behavior, and whether a repro exists.

Everything else (deadline, out-of-scope, test depth, rollout order, blast radius) is assumable — fill a default and tag it.

Hard rules

  • Never guess silently; cap at two question rounds; never assign skills to subtasks (describe the concern — "schema design", "security review" — not a skill filename).
  • Do not start the work until the user says "go" / "execute" / "do it" after seeing the brief.
  • Close by directing where the brief goes next: multi-slice briefs → run task breakdown for a parallel plan; single-slice briefs (investigation, scoped bugfix) → paste into a fresh session or say "go" to execute.

Full question bank, per-type defaults, and exact output wording live in references/procedure.md and references/interview-checklist.md.

Related Skills

  • planning-and-task-breakdown — consumes a multi-slice brief and decomposes it into ordered, parallel-dispatchable tasks with an execution DAG.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.