Skip to content
Back to search

promptmaster-skill

Get it right the first prompt. A free Claude skill that rewrites your prompt into an optimized format, audits 35 credit-waste patterns, and locks your decisions so you never re-explain.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/jakeschincariol--promptmaster-skill/badge.svg)](/listings/jakeschincariol--promptmaster-skill)

HTML

<a href="/listings/jakeschincariol--promptmaster-skill"><img src="/api/listings/jakeschincariol--promptmaster-skill/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

The fast way: **paste this repo link into Claude and ask it to install the skill.** Then type `/promptmaster` before any request.

https://github.com/Jakeschincariol/promptmaster-skill
Install this skill, then confirm /promptmaster works.

Manual, for Claude Code:

git clone https://github.com/Jakeschincariol/promptmaster-skill
cp -R promptmaster-skill/.claude/skills/promptmaster ~/.claude/skills/

Project-local instead: copy the same folder into your repo's `.claude/skills/`. No Claude Code? Paste [SKILL.md](SKILL.md) at the top of any Claude chat and it runs as a mode.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 109 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. ---
  2. name: promptmaster
  3. description: >-
  4. Rewrite the user's raw request into an optimized prompt before answering it,
  5. audit it against the 35 credit-waste patterns in waste-patterns.json, and
  6. enforce every locked decision in decisions.md so nothing gets re-explained
  7. and nothing contradicts itself. Use whenever the user types /promptmaster,
  8. prefixes a request with it, or asks to optimize / tighten / fix a prompt.
  9. ---
  10. # Promptmaster
  11. Get it right the first prompt. The user hands you a raw, messy, underspecified
  12. request. You turn it into a prompt that lands on the first try, then execute it.
  13. The goal is fewer turns: one optimized prompt instead of five rounds of
  14. re-explaining.
  15. ## The loop
  16. Run these five steps, in order, every time.
  17. ### 1. Load locked decisions
  18. Read `decisions.md` from the project root (create it from the template in this
  19. skill folder if the user locks something and it does not exist yet). In a chat
  20. without a filesystem, keep the decisions table in the conversation instead.
  21. Every locked decision is a standing order. Apply all of them to the rewrite
  22. without being asked. Never produce output that violates a locked decision. If
  23. the user's new request contradicts one, stop and say so:
  24. > This contradicts a locked decision (`length: under 200 words`, locked
  25. > 2026-08-14). Say `unlock length` to change it, or I'll keep the lock.
  26. Do not silently pick a side. Blocking the contradiction is the feature.
  27. ### 2. Rewrite the raw request
  28. Turn the vague ask into a specific one. Plain English, not a jargon template.
  29. A good rewrite answers, in one tight block:
  30. - **What exactly** gets produced (the deliverable, named: a table, a script, a file)
  31. - **For whom** (audience and what they already know)
  32. - **How big** (word cap, item count, duration)
  33. - **Hard constraints** (tools, stack, budget, deadline, do-not-touch list)
  34. - **Done when** (how the user will judge it finished)
  35. Example, the whole move in one line each:
  36. Raw: `write me something for instagram, idk`
  37. Rewritten: `Write a 30-day Instagram plan for a local coffee shop with 8k
  38. followers. One table: week, theme, 3 post ideas. Under 400 words.`
  39. Pull the specifics from three places, in this order: what the user already said
  40. in this conversation, the locked decisions, and only then sensible defaults. If
  41. one fact is genuinely missing and guessing wrong would waste a full turn (the
  42. audience, the quantity, the platform), ask for it in ONE batched question with a
  43. default offered for each item. One round of questions max. Never drip.
  44. ### 3. Audit against the 35 patterns
  45. Check the RAW request against every pattern in `waste-patterns.json`. Note each
  46. one it trips by id and name. The rewrite in step 2 must clear every flagged
  47. pattern. If your own rewrite still trips one, fix the rewrite.
  48. ### 4. Show the work, briefly
  49. Before the answer, print a compact header so the user sees what changed:
  50. ```
  51. PROMPTMASTER
  52. rewrote: "write me something for instagram, idk"
  53. into: "30-day IG plan, local coffee shop, 8k followers, one table, <400 words"
  54. flagged: #1 no deliverable named, #2 no audience, #3 no length, #5 hedging
  55. applied: voice=casual (locked), emojis=never (locked)
  56. ```
  57. Keep it to those four or five lines. The header is a receipt, not an essay.
  58. ### 5. Execute the rewritten prompt
  59. Answer the optimized prompt in full, in the same turn. The user should get the
  60. finished deliverable, not homework. Only skip execution when the user explicitly
  61. asks for the rewrite alone ("just optimize this, don't run it").
  62. ## Locking and unlocking decisions
  63. When the user states a durable preference, or accepts one you proposed, lock it:
  64. - `lock: voice = casual, like I talk` appends a row to `decisions.md`
  65. - `unlock voice` removes it
  66. - `decisions` prints the current table
  67. Also lock decisions the user makes in passing ("never use emojis", "keep
  68. everything under 200 words") without requiring the `lock:` syntax. Confirm in
  69. one line: `locked: emojis = never`. When in doubt whether something is a
  70. one-off or a standing rule, ask in the same turn.
  71. ## Hard rules
  72. - Never make the user say the same thing twice. If it was said, it is context.
  73. If it is standing, it is locked. (Pattern #35: redundant context.)
  74. - Never contradict a locked decision. Surface the conflict instead.
  75. - One clarifying round maximum, batched, with defaults. Then commit.
  76. - The rewrite must be shorter or equal to the raw request plus its missing
  77. facts. Optimizing never means padding.
  78. - The receipt header is mandatory. It is how the user learns to write the
  79. tight version themselves.
README.md · README · 75 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # promptmaster — get it right the first prompt
  2. A free Claude skill that rewrites your prompt into an optimized format Claude
  3. understands the first time, so you are not stuck re-explaining yourself and
  4. burning through usage.
  5. Every retry resends your whole conversation. Five rounds of "no, I meant..."
  6. means you paid for the same context five times. Promptmaster kills the retries
  7. at the source:
  8. - **Rewrites your raw ask** into a specific one before answering it. Vague in,
  9. exact out, then it executes the optimized prompt in the same turn.
  10. - **Audits against 35 credit-waste patterns** ([waste-patterns.json](waste-patterns.json)):
  11. the specific ways prompts force retries, from "no deliverable named" to
  12. pattern #35, redundant context.
  13. - **Remembers the decisions you already made** ([decisions.md](decisions.md)):
  14. lock a preference once (voice, length, emojis, stack) and it is applied to
  15. every future request. Contradictions get blocked, not silently absorbed.
  16. The whole move, one line each:
  17. ```
  18. raw: write me something for instagram, idk
  19. rewritten: Write a 30-day Instagram plan for a local coffee shop with 8k
  20. followers. One table: week, theme, 3 post ideas. Under 400 words.
  21. ```
  22. ## Install
  23. The fast way: **paste this repo link into Claude and ask it to install the
  24. skill.** Then type `/promptmaster` before any request.
  25. ```
  26. https://github.com/Jakeschincariol/promptmaster-skill
  27. Install this skill, then confirm /promptmaster works.
  28. ```
  29. Manual, for Claude Code:
  30. ```bash
  31. git clone https://github.com/Jakeschincariol/promptmaster-skill
  32. cp -R promptmaster-skill/.claude/skills/promptmaster ~/.claude/skills/
  33. ```
  34. Project-local instead: copy the same folder into your repo's
  35. `.claude/skills/`. No Claude Code? Paste [SKILL.md](SKILL.md) at the top of any
  36. Claude chat and it runs as a mode.
  37. ## Use
  38. ```
  39. /promptmaster write me something for instagram, idk
  40. ```
  41. You get a short receipt (what was rewritten, which waste patterns were flagged,
  42. which locked decisions applied), then the finished answer. Standing preferences:
  43. ```
  44. lock: voice = casual, like I talk
  45. unlock voice
  46. decisions
  47. ```
  48. ## Files
  49. | file | what it is |
  50. | --- | --- |
  51. | [SKILL.md](SKILL.md) | the rewrite loop + examples |
  52. | [waste-patterns.json](waste-patterns.json) | all 35 credit-waste patterns |
  53. | [decisions.md](decisions.md) | persist locked decisions |
  54. | `.claude/skills/promptmaster/` | the same skill in standard installable layout |
  55. More free skills and resources: [opusjake.ai](https://opusjake.ai)
  56. MIT licensed.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.