paired-council
Use when deliberating on strategic questions, brainstorming, or making decisions that benefit from multiple behavioral perspectives in productive tension. Deploys math-optimized persona pairs (from so
SKILL.mdplugins/agent-otto-founder/skills/paired-council/SKILL.md
GitHubPaired Council
Deploys math-optimized persona pairs from the sovereign balance matrix to deliberate on a question. Each pair debates internally, producing productive tension that individual personas cannot. The pair format catches blind spots that single-perspective councils miss.
Commands
| Command | Default Pairs | Purpose |
|---|---|---|
/paired-council deliberate <question> |
Auto-select by query type | Strategic deliberation |
/paired-council brainstorm <topic> |
Innovation-weighted pairs | Creative exploration |
/paired-council audit <concern> |
Safety-weighted pairs | Risk assessment |
/paired-council custom <pair1,pair2,...> <question> |
User-specified | Custom pair selection |
Math-Optimized Pairs (from sovereign balance rankings)
| Combo Name | Pair | Balance | Energy | Best For |
|---|---|---|---|---|
| Mirror Match | Maverick + Specialist | 0.87 | 31 | Innovation + verification. Bold moves stress-tested. |
| Knowledge Bomb | Analyzer + Persuader | 0.50 | 23 | Research → weapon. Evidence packaged for impact. |
| The Humanitarian | Altruist + Controller | 0.50 | 25 | Human cost + accountability. Ethics meets operations. |
| The Bridge | Collaborator + Controller | 0.87 | 21 | Relationship dynamics. Softens rigid systems. |
| Controlled Burn | Maverick + Guardian | 1.50 | 27 | Bold with a firewall. Speed checked by safety. |
| Recon | Analyzer + Adapter | 3.08 | 15 | Scout + staging. Prep work and terrain mapping. |
Auto-Selection Logic
digraph pair_selection {
"Query type?" [shape=diamond];
"Strategic/vision" [shape=box, label="Mirror Match + Knowledge Bomb\n+ The Humanitarian + The Bridge"];
"Technical/build" [shape=box, label="Mirror Match + Controlled Burn\n+ Knowledge Bomb + Recon"];
"People/culture" [shape=box, label="The Humanitarian + The Bridge\n+ Knowledge Bomb"];
"Risk/safety" [shape=box, label="Controlled Burn + The Bridge\n+ Knowledge Bomb + Recon"];
"Brainstorm" [shape=box, label="Mirror Match + Knowledge Bomb\n+ The Humanitarian"];
"Query type?" -> "Strategic/vision" [label="strategy"];
"Query type?" -> "Technical/build" [label="build"];
"Query type?" -> "People/culture" [label="people"];
"Query type?" -> "Risk/safety" [label="risk"];
"Query type?" -> "Brainstorm" [label="creative"];
}
Execution Protocol
Step 1: Parse Command
Extract query, pair selection (auto or custom), scope.
Step 2: Fetch Context
- Repo relays:
GET http://localhost:8100/constellation/relays/repos - Persona relays:
GET http://localhost:8100/constellation/relays/personas - Read relevant profile YAMLs from
airlock-persona/profiles/
Step 3: Select Pairs
Use auto-selection logic or user-specified pairs. Default: 3-4 pairs.
Step 4: Write Briefing
~200 token briefing with: query, scope, key facts from relays, current context.
Step 5: Dispatch Pairs (Parallel)
For each pair, dispatch ONE subagent that simulates BOTH personas in dialogue. Use model: "sonnet" for cost efficiency.
IMPORTANT: Dispatch ALL pair subagents in a SINGLE message (parallel execution).
Each subagent receives the paired-council-prompt.md template:
You are simulating a PAIRED DISCUSSION between two Otto personas.
## The Pair: {COMBO_NAME} ({PERSONA_A} + {PERSONA_B})
- **{PERSONA_A}** (D:{D}, E:{E}, P:{P}, F:{F}) — {BIO}
- **{PERSONA_B}** (D:{D}, E:{E}, P:{P}, F:{F}) — {BIO}
- **Pair Math:** Balance {BALANCE}, Energy {ENERGY}. {PAIR_DESCRIPTION}
## The Question
{BRIEFING + QUERY}
## Format
Simulate the actual discussion. Show productive tension. Then deliver joint verdict.
{PERSONA_A}: [speaks first — in character]
{PERSONA_B}: [responds — in character, catches what A missed]
{PERSONA_A}: [reacts, pushes back or agrees]
{PERSONA_B}: [final word]
{COMBO_NAME} VERDICT:
- The thing we found: [one clear statement]
- Why it matters: [2-3 sentences]
- Recommended action: [specific, actionable]
- Conviction: [0-10]
Step 6: Collect Verdicts
Parse each pair's verdict. Extract: finding, reasoning, action, conviction.
Step 7: Guardian Gate
Scan all verdicts for safety/sustainability/alignment risks. If any concern has conviction >= 7.5, flag as HARD_DISAGREE. If no Guardian persona in pair set, Knowledge Bomb or Controlled Burn serves as proxy gate.
Step 8: Captain Synthesis
As Captain, synthesize:
- Where pairs converge (consensus)
- Where pairs diverge (productive disagreement)
- Sequenced recommended actions
- Decision status: INFORMATIONAL, AWAITING_ZAC, or BLOCKED
Step 9: Persist Report
Write YAML report to ~/Desktop/Airlock/repos/airlock-coordination/reports/
Follow schema. Prepend to INDEX.yaml. Include format: paired_council and pairs: metadata.
Privacy rule: request_summary is SYNTHESIZED, never verbatim.
Step 10: Render Report
══════════════════════════════════════════════════
PAIRED COUNCIL REPORT
Type: {command} | Query: "{query}"
Date: {date} | Pairs: {count}
══════════════════════════════════════════════════
{For each pair:}
──────────────────────────────────────────────────
{COMBO_NAME} ({PERSONA_A} + {PERSONA_B}) — Balance {N}, Energy {N}
Conviction: {N}
[Abbreviated discussion highlights]
Verdict: {finding}
Action: {recommendation}
──────────────────────────────────────────────────
══════════════════════════════════════════════════
CONSENSUS
Avg Conviction: {N} | Agreement: {X}/{Y}
GUARDIAN GATE: {PASS|HARD_DISAGREE}
══════════════════════════════════════════════════
CAPTAIN SYNTHESIS
{synthesis with sequenced actions}
Decision Status: {status}
══════════════════════════════════════════════════
Why Pairs Beat Individuals
| Individual Council | Paired Council |
|---|---|
| Each persona gives isolated take | Each pair produces internal debate |
| No productive tension within response | Behavioral mirrors generate tension |
| Blind spots remain per-persona | Pair's opposite drives catch blind spots |
| Generic agreement patterns | Math-validated complementary perspectives |
| N perspectives, N responses | N/2 pairs, each with internal dialogue |
Cost
- 3-4 pairs × Sonnet = $0.15-0.30 per council
- ~30 seconds parallel execution
- Significantly richer output than individual councils at similar cost