prompt
Skill3 starsChanged 23 days agoMITView on GitHub
Rewrite a rough request into a precise, self-contained Claude Code prompt grounded in the repo (never executes it). Use when the user asks to turn, rewrite or improve a request into a prompt for another session.
23 other projects use a version of this.
Adopted by
Most recently taken up by
- crsmithdev/alephtheir own version, first seen Nov 16, 2026
- MagicJosh/rastheir own version, first seen Nov 16, 2026
- psw7205/skillstheir own version, first seen Nov 16, 2026
- KroxiDev/engineering-skillstheir own version, first seen Nov 15, 2026
- codeboehm/maximtheir own version, first seen Nov 15, 2026
- JetBrains/mcp-steroidtheir own version, first seen Nov 8, 2026
Used with
Projects with it also use brainstorm, init and 3 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
nicolasestremSkillbrainstorm
Structured idea generation and solution exploration with systematic analysis.
Also in 4 projects that use this
MicrosoftSkillinit
Also in 3 projects that use this
MicrosoftSkillhandoff
Compact the current conversation into a handoff document for another agent to pick up.
Also in 5 projects that use this
SentrySkillreview
Fixture review skill.
Also in 3 projects that use this
GoogleSkillcode-review
Adversarial systems, concurrency, memory-model, and architecture code review protocol. Use this skill whenever asked to review code, PRs, diffs, or branches in Filament.
Also in 3 projects that use this
Other versions23 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 23
- treylom/prompt-engineering-skills★ 194same name
- JetBrains/mcp-steroid★ 82same name
- oprogramadorreal/optimus-claude★ 75same name
- treylom/ThisCode★ 40same name
- analystacademy/claude-skills★ 35same name
- mehrad-dm/mastermind★ 23same name
- amatray/claude-skills-public★ 9same name
- KroxiDev/engineering-skills★ 4same name
- alphazede/bearing-lite★ 3same name
- ksprashu/agent-skill-forge★ 3same name
13 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository12 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .cursor/rules/00-consumer-boundary.mdcCursor rule
- .cursor/rules/caveman-mode.mdcCursor rule
- .cursor/rules/cost-routing.mdcCursor rule
- .cursor/rules/execute-operator-intent.mdcCursor rule
- .cursor/rules/ponytail-lite.mdcCursor rule
- debugSkill
- ponytail-liteSkill
- pr-reviewSkill
- shipSkill
- systematic-debuggingSkill
The file
We split this skill into the rules it gives an agent
7 rules. 0 are common, 7 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 20 edited versions we read keep these.
7
Distinctive
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
- Distinctive: line 13
If the raw request is empty, ask for it in one line and stop.
- Distinctive: line 22
Write the prompt: Never change the objective or add requirements.
- Distinctive: line 34
Never pick Execute when the request still needs a plan.
- Distinctive: line 40
Calm wording: No ALL-CAPS, "CRITICAL", or "MUST" stacks.
- Distinctive: line 48
No secrets: Never paste credential values.
- Distinctive: line 49
Short output: Ask for a brief final report (what changed, how verified, open risks), not a narrative.
- Distinctive: line 53
Keep only the lines that carry information, in this order:
“Common” is measured: kept by at least 6 in 10 of the 20 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
prompt
Rewrite the raw request into a prompt that a fresh Claude Code session (Opus 5.5, high effort) can run with no follow-up questions. Never execute the request, never edit files, never write the code asked for. Output only the package below.
Raw request: $ARGUMENTS
If the raw request is empty, ask for it in one line and stop.
Process
- Objective. Find the real objective and the observable "done" state. Several independent objectives → one prompt each, in run order (one session per objective keeps context small).
- Carry the context. The fresh session sees none of this conversation. Turn references ("this", "the error above", "like before") into concrete facts; copy error messages, paths, URLs, IDs and numbers verbatim.
- Ground it, read-only and cheap: up to ~5 Glob/Grep/Read calls to confirm file paths, the test/lint/build command, and patterns to follow. No subagents. In a multi-repo hub, identify the target project folder. Skip when the request is not about files on disk.
- Resolve ambiguity with explicit assumptions. Ask only when a wrong guess would waste the whole run: one AskUserQuestion with 2-4 concrete options, then stop.
- Pick the permission mode (below) and the tier (cost-routing).
- Write the prompt. Never change the objective or add requirements; improve clarity, scope, and verification only. Already a good prompt → change little and say so.
Permission mode
Recommend exactly one:
| Mode | When | How to enter |
| Plan mode | Multi-file refactor, migration, architecture, audit, unclear impact, anything where a wrong first edit is costly | Shift+Tab until "plan mode on" (CLI) or the mode picker (desktop); approve the plan before edits |
| Execute | Scope is clear: fix a bug, implement a bounded feature, add tests, rename, write docs into the repo | Default mode, or accept-edits for low-risk edits |
| Read-only | Explain, research, review, compare options, answer a question | Default mode; the prompt says "do not edit files" |
Never pick Execute when the request still needs a plan.
How to write for Opus 5.5 (high effort)
- Direct and specific. No "You are an expert…" role line unless a role changes behavior (e.g. "review as a security auditor").
- No reasoning boosters. High effort already thinks; skip "think step by step" / "ultrathink".
- Calm wording. No ALL-CAPS, "CRITICAL", or "MUST" stacks; Opus follows plain instructions and over-applies shouted ones. Give the why for non-obvious constraints instead.
- Size to the task. A one-line fix gets a few lines; only complex work uses every section of the skeleton.
- Facts, not guesses. Cite only paths and commands you verified; otherwise say where to look ("the auth middleware under
src/server/"). - Tight scope. Name what to touch and what not to touch; say "smallest change that works, no unrelated refactors" (Opus tends to over-build).
- Files by reference.
@pathfor small key files (loads them into context); plain paths for large ones so they are read only if needed. Bound searches to folders or file types. - Verifiable done. Concrete success criteria plus the exact command to verify (focused test first). No test → say how to check by hand; UI change → which page or route to open and what should appear.
- Stop rules. Task-specific triggers to stop and ask: missing fact, destructive or irreversible step, data or schema migration.
- No repeats. The session already loads
CLAUDE.mdand the always-on rules (commit policy, retry cap, prose style, boundary); do not restate them. Delivery requested → the last step is/ship. - No secrets. Never paste credential values; name the file or env var that holds them.
- Short output. Ask for a brief final report (what changed, how verified, open risks), not a narrative.
Prompt skeleton
Keep only the lines that carry information, in this order:
<Objective in one or two sentences, including why.>
Work in: <folder, only when it is not the session root>
Context:
- <verified facts: @small-file or plain path, current behavior, exact error>
Scope:
- Change: <files or areas>
- Leave alone: <files or areas>. Smallest change that works, no unrelated refactors.
Steps: (only when order matters)
1. ...
Done when:
- <observable result>
Verify: `<exact command>` | <how to check by hand>
Stop and ask if: <task-specific triggers>
Report: what changed, how it was verified, open risks. Brief.Output
Reply in the operator's language; translate the headings, keep this structure:
Recommended mode
Plan mode | Execute | Read-only
Why
One or two lines.
Session
One line: fresh session or /clear if the current context is unrelated; the folder to open when it is not the current root; tier L0-L4 per cost-routing. Mention model or effort only when the tier does not fit Opus 5.5 high (e.g. L0 → /model haiku or /effort low).
Optimized prompt
<self-contained prompt, ready to paste into Claude Code>Fence it with four backticks so code blocks inside the prompt cannot close it.
Assumptions
Assumptions made (omit the section when there are none).
Improvements
Up to 3 short bullets: what became more specific, scoped, or verifiable.
Several prompts → number them in run order, each with its own mode, why, session and prompt; assumptions and improvements once at the end.
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
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.

