Prompt matcher

Prompt Matcher

Reads your message, silently finds the best matching prompt template from your library, fills in every variable it can infer, asks only what it can't, and executes. Every message gets a better prompt than you'd write manually.

---

When to activate

Always-on auto mode. On every task message, before executing:

1. Score the message against the prompt library (0-10 per template)
2. If top match scores 6+: use it — show Using: [Prompt Name] in one line, fill known variables silently, ask only for genuinely unknown variables (1 batch, never one at a time)
3. If all variables are inferable from context: execute entirely silently
4. If no match above 4: execute the message directly without the skill

Direct invoke: /pm [description] — forces the match flow even for low-scoring messages.

Never activate for: casual chat, yes/no questions, messages under 8 words with no task, messages already structured as a complete prompt.

---

Prompt Library Location

Primary library: {PROMPTS_DIR}

Categories:

• writing.md — Email, editing, explaining, summarizing, style rewriting
• research.md — Deep research, compare options, devil's advocate, market research, fact-checking
• coding.md — Bug fixing, code explanation, code review, feature building, script writing
• thinking.md — Decisions, brainstorming, first principles, pre-mortem, action plan, mental models
• learning.md — Concept explainer, study guide, teach me by doing, Socratic method
• business.md — Positioning, pricing, sales email, competitor analysis, launch checklist
• creative.md — Story generator, naming, taglines, dialogue, reframe
• productivity.md — Task breakdown, weekly review, habit system, meeting prep, delegation
• anti-sycophancy.md — Challenge this, steelman opposition, adversarial review, red team
• context-discipline.md — Session start, compact timing, model selection, token hygiene
• ai-engineer.md — Architecture, RAG audit, agent debugger, eval suite, cost killer, production review
• trading.md — Strategy auditor, prop firm rule checker, prediction market designer, risk framework
• content-creation.md — Faceless content system, viral hooks, short-form scripts, Apify pipeline
• career-business.md — ATS resume, cold outreach, AI service niche validator, solopreneur designer

---

Matching logic

Score each prompt on:

• Intent match (0-4): does the prompt's use-when match what the user is trying to do?
• Context fit (0-3): are most variables already answerable from the conversation?
• Specificity (0-3): is this prompt more specific than a generic response would be?

Pick the highest scorer. If two are within 1 point, mention both and let the user choose.

Examples:

• "debug why my agent failed" → Agent Debugger (9/10)
• "write an email to a recruiter" → Email Writer (9/10)
• "should I do X or Y" → Decision Maker (8/10)
• "explain recursion simply" → Concept Explainer (9/10)
• "make my code faster" → Cost Killer or Code Reviewer (7/10 each — ask which)
• "go deeper on the memecoin system" → Deep Research (8/10)
• "audit my prop firm trade" → Prop Firm Rule Checker (9/10)
• "help me write a hook for a faceless Reel" → Viral Hook Generator (9/10)
• "what do you think?" → no match, respond directly

---

Execution flow

User message received
        ↓
Score all prompts in Brain\_prompts\
        ↓
Score 6+?
   YES → Show "Using: [Name]" + fill variables silently
           ↓
         Unknown variables? → Ask all in one message
           ↓
         Execute with completed template
   NO  → Execute directly, no skill

---

Adding prompts

When a pattern repeats 3+ times in a session, or user says "save this as a prompt":

1. Extract the template with [VARIABLE] placeholders
2. Write to Brain\_prompts\[best-category].md
3. Confirm: "Added [Name] to _prompts/[category].md"

---

Slash command

/pm [description] — force-triggers the match flow.