curiositech/discourse-coordinator

Discourse Coordinator

When to Use This Skill

Use this skill when:

• Operating as a discourse copilot across a multi-turn conversation
• A user needs holistic discourse support rather than a single specific skill
• Analyzing an ongoing or recorded conversation for quality, integrity, and dynamics
• Facilitating a dialogue between parties with different positions
• The user wants help preparing a persuasive speech, debate position, or written argument
• Conflict level in a conversation is rising and you need to decide how to intervene
• The user asks for "discourse analysis," "conversation coaching," or "dialogue support"

Do NOT use this skill when the user has a narrow, single-skill request (e.g., "find fallacies in this paragraph" → use logical-fallacy-detector directly).

---

Four Cooperating Roles

The eight discourse skills are organized into four functional roles. This coordinator routes between them.

Builder — Structures and Strengthens Arguments

| Skill | Purpose | |---|---| | steel-man-argument | Construct the strongest possible version of any position | | toulmin-argument-analysis | Map claim → grounds → warrant → backing → qualifier → rebuttal | | forensic-speech-structure | Compose and structure a speech or written argument for delivery |

Guide — Shapes the Process of Dialogue

| Skill | Purpose | |---|---| | socratic-questioning | Drive inquiry-based exploration; surface assumptions | | productive-discourse-facilitator | Set norms, manage turns, propose phase transitions |

Critic — Inspects Quality and Integrity

| Skill | Purpose | |---|---| | logical-fallacy-detector | Identify formal and informal fallacies in reasoning | | bad-faith-rhetoric-detector | Flag manipulation, deflection, and bad-faith moves | | discourse-elision-analyzer | Detect what is conspicuously absent or avoided in a response |

Coordinator — This Skill

Routes between roles based on goal, mode, and conversation state. Tracks patterns across turns and manages mode transitions.

---

Three Operating Modes

Mode 1: Silent Analysis (Default)

• Continuously evaluate each turn using Critic skills in the background
• Surface only minimal indicators (e.g., a single-line flag) unless the user explicitly asks for detail
• Track patterns across turns: escalation trajectory, repeated elisions, accumulating fallacy count
• Never interrupt the flow of conversation; log internally and report on request

Activation: Default on load. Also activated by phrases like "watch this conversation," "just observe," "monitor this."

Mode 2: Coaching Overlay

• Provide private feedback to the user only — the other party does not see this
• Examples of coaching prompts:
  • "Your last turn dodged their core question about liability. Consider addressing it directly."
  • "Consider steelmanning their concern about X before pushing Y — it may reduce defensiveness."
  • "Their last three turns show a pattern: they've never engaged your cost argument. Flag it."
• Suggest small, targeted interventions via Guide skills
• Tone: supportive, non-judgmental, specific

Activation: User phrases like "coach me," "give me feedback," "help me respond," "what should I say."

Mode 3: Active Facilitator

• All parties are aware the agent is participating as a third voice
• Uses productive-discourse-facilitator to establish ground rules and manage turn structure
• Proposes Structured Analytical Conversation (SAC) moves: restatements, clarifying questions, phase transitions
• Calls out patterns that undermine dialogue openly (e.g., "I notice this question has gone unanswered three times")
• Can introduce steelmanning rounds, inquiry phases, or synthesis prompts

Activation: User phrases like "facilitate this," "be the moderator," "help us have this conversation," or explicit consent from all parties.

---

Routing Logic

Use these rules to decide which skill(s) to invoke for a given turn or request:

| Situation | Primary Route | Secondary Check | |---|---|---| | User's goal is to persuade | Builder skills | Critic (quality check) | | User's goal is to understand | Guide skills | Critic (integrity check) | | User's goal is to analyze/evaluate | Critic skills | — | | Conflict level rising | Guide → productive-discourse-facilitator | Critic (monitoring) | | Long response follows short pointed question | discourse-elision-analyzer | logical-fallacy-detector | | Repeated non-engagement with a core point | bad-faith-rhetoric-detector | discourse-elision-analyzer | | User preparing to speak or write | Builder → Critic | — | | Emotional intensity high | productive-discourse-facilitator first | Builder after de-escalation | | Argument structure unclear | toulmin-argument-analysis | — | | User wants strongest form of opponent's view | steel-man-argument | — | | Formal speech or debate prep | forensic-speech-structure | steel-man-argument for rebuttal prep |

For full decision trees, threshold definitions, and conflict-resolution between skills, see references/routing-rules.md.

---

Orchestration Patterns

Hot Topic Protocol (for active conflict)

Use when two or more parties are in active disagreement with rising stakes.

1. Phase 0 – Setup: Facilitator asks each side for a brief 2–3 sentence position statement. Run toulmin-argument-analysis privately to map each side's structure.
2. Phase 1 – Mutual Steelman: steel-man-argument helps each side restate the other's position at its strongest. The originating party must confirm the restatement is accurate before proceeding.
3. Phase 2 – Inquiry: socratic-questioning drives a round of mutual questions. Each party asks one genuine question; the other must answer directly before asking theirs.
4. Phase 3 – Integrity Check: logical-fallacy-detector and bad-faith-rhetoric-detector scan the full conversation neutrally; findings shared with all parties.
5. Phase 4 – Synthesis: productive-discourse-facilitator guides toward identifying any shared ground, naming remaining disagreements precisely, and proposing next steps.

Quick Analysis (for a single turn or response)

Use when a user pastes a text and asks "what's wrong with this?" or "analyze this response."

1. Run discourse-elision-analyzer on the text — what is conspicuously absent?
2. Run logical-fallacy-detector on any arguments found
3. Summarize findings in 3–5 bullet points, ordered by significance

Speech Prep Pipeline

Use when a user is preparing a speech, essay, or structured argument.

1. toulmin-argument-analysis — structure raw ideas into claim/grounds/warrant/rebuttal scaffold
2. logical-fallacy-detector — clean up reasoning; remove or repair weak inferences
3. forensic-speech-structure — compose the speech with appropriate structure (introduction, body, conclusion, transitions)
4. steel-man-argument — anticipate the strongest counterarguments; build refutation section

---

State Tracking

Maintain this state across turns in a conversation:

• Conversation temperature: calm → warm → heated → crisis
• Unanswered core questions: running list; flag if a question goes unanswered for 3+ turns
• Pattern accumulation:
  • Elision count (how many times key topics have been avoided)
  • Fallacy count (types and frequency)
  • Bad-faith pattern count (types and frequency)
• Mode: current operating mode and what triggered any transitions
• Phase (in Hot Topic Protocol): which phase the conversation is in

Report state summary when asked or when escalating to a new mode.

---

Sensitivity Settings

Adjust behavior based on the declared or inferred context:

relationship-mode (personal relationships, family, romantic partners):

• Extra caution with labeling; avoid "bad faith" as a definitive label — use "pattern that may indicate..."
• Validate emotional content before analytical content
• Emphasize understanding over winning
• Do not recommend confrontational moves

professional-mode (workplace, business negotiations, academic debate):

• More direct analysis; less emotional hedging
• Name patterns clearly and specifically
• Focus on argument structure and productive outcomes

debate-prep-mode (competitive debate, formal argumentation):

• Optimize for competitive strength; find every weakness in the opponent's case
• Less concern for relationship preservation; prioritize logical rigor
• Use forensic-speech-structure with competitive framing
• Aggressive use of steel-man-argument to anticipate and pre-empt

Set by user instruction or infer from context. Default: professional-mode.

---

Reference Files

• references/routing-rules.md — Detailed decision trees, threshold definitions, mode-transition procedures, skill-conflict resolution, priority ordering, full orchestration examples, and guidance on when to refer to human support. Load this file when handling complex routing decisions or unfamiliar orchestration scenarios.