Skip to content

Latest commit

 

History

History
72 lines (58 loc) · 4.06 KB

File metadata and controls

72 lines (58 loc) · 4.06 KB

Workflow: Author Technique (Research-First)

When to use

User wants to create or improve a technique file that encodes how to approach a type of situation. This workflow grounds the technique in researched best practices before layering organizational expertise.

Process

Step 1: Define the technique scope

Ask the user:

  1. "What type of situation does this technique cover?" (e.g., buyer persuasion, competitive displacement, executive outreach, onboarding)
  2. "What outputs will this technique be used to generate?" (e.g., sales decks, emails, one-pagers, talking points)
  3. "What foundation files will this technique reference?" (e.g., ICP, messaging, product, competitors)

Step 2: Generate a research prompt

Based on the technique scope, generate a targeted research brief that covers:

  • Established methodologies and frameworks relevant to the situation type
  • Best practices from experienced practitioners (not academic theory)
  • Common anti-patterns and failure modes
  • How the approach should adapt to different personas or contexts

The research prompt should be specific enough to produce actionable findings, not generic advice. Present the prompt to the user for review before executing.

Step 3: Execute the research

Run the research prompt using web search, retrieving practitioner-written content, methodology documentation, and real-world examples. Prioritize:

  • Named methodologies with proven track records
  • Practitioner content (CROs, founders, domain experts who've done the work)
  • Data-backed findings (e.g., from Gong, Forrester, industry research)
  • Framework documentation from published books

Compile findings into a structured research document organized by topic area.

Step 4: Generate a technique authoring prompt

Using the research findings, generate a detailed prompt for writing the technique file. This prompt should:

  • Specify the required structure (from system-instructions.md): required foundations, required context, reasoning approach, output constraints, anti-patterns
  • Encode the key research findings into specific reasoning steps
  • Identify which foundations the technique should reference
  • Define what context fields are required at generation time
  • Ensure the technique passes the "swap test": if you replaced all foundations with a different organization's knowledge, the reasoning approach should still be structurally valid

Present the prompt to the user for review before executing.

Step 5: Draft the technique

Execute the authoring prompt to produce the technique file. The technique must:

  • Follow the file structure from system-instructions.md exactly
  • Include a sequential reasoning approach (step-by-step logic the AI follows)
  • Reference foundations by filename, never by restating their content
  • Specify required context fields that must be provided at generation time
  • Include anti-patterns with explanations of why they fail
  • Include output constraints (tone, length, format, inclusions, exclusions)

Step 6: User sharpening

Present the draft to the user for review. Their expertise fills the gaps research cannot:

  • Are the reasoning steps in the right order?
  • Does the technique match how your best people actually approach this situation?
  • Are there anti-patterns from your experience that research didn't surface?
  • Are the output constraints right for your standards?

Iterate based on feedback until the user approves.

Step 7: Save and register

Save the technique file to knowledge/techniques/technique-[situation-type].md. Update the knowledge base README with the new technique.

Quality checks

After authoring, verify:

  • Technique references foundations by filename, never restates their content
  • Reasoning approach is sequential — each step builds on the previous
  • Required context fields are explicitly listed
  • Anti-patterns include why they fail, not just what to avoid
  • The swap test passes: structure works with any organization's foundations
  • Output constraints are specific enough to enforce, not vague preferences
  • No specific knowledge that belongs in a foundation file has leaked in