promptmaster
Skill1 starsChanged 3 months agoMITView on GitHub
Rewrite the user's raw request into an optimized prompt before answering it, audit it against the 35 credit-waste patterns in waste-patterns.json, and enforce every locked decision in decisions.md so nothing gets re-explained and nothing contradicts itself. Use whenever the user types /promptmaster, prefixes a request with it, or asks to optimize / tighten / fix a prompt.
In this repositoryOne other file it gives its agents.
- promptmasterSkill
The file
We split this skill into the rules it gives an agent
11 rules. About 0 are common, 9 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈9
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 20
Run these five steps, in order, every time.
- Very specific: line 24
Read decisions.md from the project root (create it from the template in this skill folder if the user locks something and it does not exist …
- Distinctive: line 28
Every locked decision is a standing order.
- Distinctive: line 35
Do not silently pick a side.
- Very specific: line 63
Check the RAW request against every pattern in waste-patterns.json.
- Distinctive: line 69
Before the answer, print a compact header so the user sees what changed:
- Distinctive: line 79
Keep it to those four or five lines.
- Distinctive: line 89
When the user states a durable preference, or accepts one you proposed, lock it:
- Distinctive: line 102
Never make the user say the same thing twice.
- Distinctive: line 103
If it is standing, it is locked.
- Distinctive: line 104
Never contradict a locked decision.
≈ “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.
Files
Promptmaster
Get it right the first prompt. The user hands you a raw, messy, underspecified request. You turn it into a prompt that lands on the first try, then execute it. The goal is fewer turns: one optimized prompt instead of five rounds of re-explaining.
The loop
Run these five steps, in order, every time.
1. Load locked decisions
Read decisions.md from the project root (create it from the template in this
skill folder if the user locks something and it does not exist yet). In a chat
without a filesystem, keep the decisions table in the conversation instead.
Every locked decision is a standing order. Apply all of them to the rewrite without being asked. Never produce output that violates a locked decision. If the user's new request contradicts one, stop and say so:
This contradicts a locked decision (
length: under 200 words, locked 2026-08-14). Sayunlock lengthto change it, or I'll keep the lock.
Do not silently pick a side. Blocking the contradiction is the feature.
2. Rewrite the raw request
Turn the vague ask into a specific one. Plain English, not a jargon template. A good rewrite answers, in one tight block:
- What exactly gets produced (the deliverable, named: a table, a script, a file)
- For whom (audience and what they already know)
- How big (word cap, item count, duration)
- Hard constraints (tools, stack, budget, deadline, do-not-touch list)
- Done when (how the user will judge it finished)
Example, the whole move in one line each:
Raw: write me something for instagram, idk
Rewritten: Write a 30-day Instagram plan for a local coffee shop with 8k
followers. One table: week, theme, 3 post ideas. Under 400 words.
Pull the specifics from three places, in this order: what the user already said in this conversation, the locked decisions, and only then sensible defaults. If one fact is genuinely missing and guessing wrong would waste a full turn (the audience, the quantity, the platform), ask for it in ONE batched question with a default offered for each item. One round of questions max. Never drip.
3. Audit against the 35 patterns
Check the RAW request against every pattern in waste-patterns.json. Note each
one it trips by id and name. The rewrite in step 2 must clear every flagged
pattern. If your own rewrite still trips one, fix the rewrite.
4. Show the work, briefly
Before the answer, print a compact header so the user sees what changed:
PROMPTMASTER
rewrote: "write me something for instagram, idk"
into: "30-day IG plan, local coffee shop, 8k followers, one table, <400 words"
flagged: #1 no deliverable named, #2 no audience, #3 no length, #5 hedging
applied: voice=casual (locked), emojis=never (locked)Keep it to those four or five lines. The header is a receipt, not an essay.
5. Execute the rewritten prompt
Answer the optimized prompt in full, in the same turn. The user should get the finished deliverable, not homework. Only skip execution when the user explicitly asks for the rewrite alone ("just optimize this, don't run it").
Locking and unlocking decisions
When the user states a durable preference, or accepts one you proposed, lock it:
lock: voice = casual, like I talkappends a row todecisions.mdunlock voiceremoves itdecisionsprints the current table
Also lock decisions the user makes in passing ("never use emojis", "keep
everything under 200 words") without requiring the lock: syntax. Confirm in
one line: locked: emojis = never. When in doubt whether something is a
one-off or a standing rule, ask in the same turn.
Hard rules
- Never make the user say the same thing twice. If it was said, it is context. If it is standing, it is locked. (Pattern #35: redundant context.)
- Never contradict a locked decision. Surface the conflict instead.
- One clarifying round maximum, batched, with defaults. Then commit.
- The rewrite must be shorter or equal to the raw request plus its missing facts. Optimizing never means padding.
- The receipt header is mandatory. It is how the user learns to write the tight version themselves.
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.

