Skip to content

host-panel

Facilitate research-grounded panels in roundtable, Oxford, and Socratic formats. Use when exploring contested topics from multiple angles. NOT for Q&A, code review, or real human opinion simulation.

host-panel 6035 words MIT v1.0.0 wyattowalsh

Facilitate research-grounded panels in roundtable, Oxford, and Socratic formats. Use when exploring contested topics from multiple angles. NOT for Q&A, code review, or real human opinion simulation.

Install:

Terminal window
npx skills add github:wyattowalsh/agents --skill host-panel -y -g --agent claude-code --agent codex --agent gemini-cli --agent antigravity --agent github-copilot --agent opencode

Use: /host-panel "topic" [format] [num-experts]

Works with Claude Code, Gemini CLI, and other agentskills.io-compatible agents.

host-panel stages a source-aware deliberation across simulated intellectual positions. It is designed to expose cruxes, evidence gaps, and trade-offs without pretending that synthetic panelists represent real human opinion, polling, prediction, or truth by consensus.

The skill includes references/debate-research.md, a research-to-rule matrix mapping AI debate and role-play findings to panel mechanics. It draws on multi-agent debate, STORM-style perspective discovery, role-playing evaluation, sycophancy research, and adversarial-persuasion failure modes.

MechanicWhy it exists
Independent first positionsReduces first-answer anchoring before cross-talk
Bounded critiqueAvoids endless, circular debate rounds
Anti-conformity checksPrevents majority pressure from masquerading as correctness
Source ledgerSeparates verified claims from thin or unverified synthesis
Methodology cardsMakes panelists differ by evidence standard, not theatrical voice
Trajectory synthesisSynthesizes by cruxes and evidence, not final-round vote
FormatPurposeBest for
roundtableOpen multi-perspective mappingPolicy trade-offs, decision-critical topics, exploratory synthesis
oxfordTesting a genuinely contestable propositionBinary motions that pass the false-balance gate
socraticProbing definitions and assumptionsConceptual disputes, ambiguous terms, hidden premises

Defaults: roundtable, 4 experts. Expert range: 2-6.

  1. Parse — Preserve /host-panel "topic" [format] [num-experts]; reject invalid formats, counts outside 2-6, and malformed topics.

  2. Diagnose — Classify the topic as settled-factual, open-value, controversial-factual, speculative, thin-evidence, decision-critical, or casual-experiential.

  3. Ground — Build a source ledger when tools are available; otherwise label the panel as unverified-training-knowledge.

  4. Map perspectives — Choose intellectual traditions from the terrain map and source ledger before creating personas.

  5. Run the panel — Require independent first positions, bounded critique, moderator challenge, convergence checks, and format-specific phases.

  6. Synthesize — Separate factual claims, interpretive disagreements, cruxes, evidence gaps, decision implications, and model limits.

InputMode
Empty $ARGUMENTSShow gallery, defaults, supported formats, and ask for a topic
"topic" onlyParse topic, auto-select format, default to 4 experts
"topic" roundtable|oxford|socraticUse explicit format
"topic" ... <2-6>Use explicit expert count
Invalid format or countReject before research or personas
Loaded premiseChallenge the framing and offer a neutral motion
Settled factual issueReframe or decline false-balance debate
Q&A, code review, real opinion simulationRedirect out of scope

Complete panels contain:

  • Topic Gate
  • Research Status
  • Source Ledger
  • Terrain Map
  • Panelist Roster
  • Panel Phases
  • Final Synthesis
  • Follow-up Options
FilePurpose
references/topic-diagnostic.mdTopic classification, false-balance gate, and format selection
references/research-integrity.mdSource ledger, citation integrity, confidence labels, and provenance
references/debate-research.mdAI debate, role-play, sycophancy, and synthesis research mapped to rules
references/archetypes.mdMethodology-card construction and anti-clustering guidance
references/moderator-rules.mdIndependent first positions, anti-conformity, turn-taking, and provocation
references/formats.mdRoundtable, Oxford, and Socratic phase structures
references/synthesis.mdTrajectory-aware final synthesis requirements

The current eval suite covers weak-topic reframing, false-balance handling, loaded premises, invalid arguments, thin evidence, citation traps, human-simulation traps, majority pressure, persuasive falsehoods, cultural/normative scope, and out-of-scope code-review redirects.


View source on GitHub