Skills / Departments / Decisions / Skills / Dialectic Cycle
All skills

/ skill

dialectic-cycle

Departments/Decisions/Skills/dialectic-cycle.md

Single-pass orchestration over Steelman → Devils-Advocate → Synthesizer. Optionally grounds each phase in NotebookLM libraries.

Open as raw file

Run this skill

Heads up — this skill doesn’t have a structured form yet, so type the full prompt with all decisions baked in (the subprocess can’t answer follow-up questions). Edit me if needed.

Runs with bypassPermissions — local dev only.

dialectic-cycle

Procedure

1. Steelman phase

  • Pass question, options, context_files, and (if any) libraries to Steelman.
  • Steelman MAY invoke notebooklm-query once per option to find supporting precedents in the libraries — e.g., "What does this corpus say about why option X works?".
  • Collect Steelman output + the citations Steelman cited.

2. Devils-Advocate phase

  • Pass Steelman's output + libraries to Devils-Advocate.
  • Devils-Advocate MAY invoke notebooklm-query once focused on failure modes — e.g., "What does this corpus say about how option X fails or where it's been tried unsuccessfully?".
  • Collect counter-arguments + citations.

3. Synthesizer phase

  • Pass Steelman + Devils-Advocate outputs + the citation list to Synthesizer.
  • Synthesizer MAY invoke notebooklm-query for one final reconciliation question if there are unresolved tensions — e.g., "Which framing in this corpus best resolves the tradeoff between A and B?".
  • Synthesizer produces the recommendation, citing which library evidence was load-bearing.

4. Cache outputs

  • Write all three outputs + the aggregated grounding_used list to Wiki/Notes/_dialectic-{date}-{slug}.md for traceability.

Grounding budget

  • Per phase: at most 1 notebooklm-query call (rate-limit awareness — don't fire 6 queries per cycle).
  • Per cycle: at most 3 total live queries.
  • If a phase needs more depth, escalate to notebooklm-deep-research BEFORE the dialectic kicks off, and use the resulting research note as additional context.

Notes

  • This skill is the engine; decision-memo wraps it with file-writing and logging.
  • Without libraries, the dialectic runs as before — pure LLM reasoning over context_files. Libraries are an enhancement, not a requirement.
  • grounding_used is what gets cited in the final memo's "Grounding" section.