council

Council

Multi-agent debate for founder decisions. 4 personas argue a topic across structured rounds, respond to each other's actual points, then synthesize. Produces a visible transcript and a recommended path.

When to Use

Invoke when:

• Conflicting signals from conversations and you need to decide which to follow
• About to change positioning in canonical files (/q-calibrate)
• Design partner asks for something that might pull you off-wedge
• Investor feedback contradicts buyer feedback
• Competitive move detected and you need a response strategy
• Any decision where "it depends" means you need structured tension
• User says "council", "debate this", "weigh options", "get perspectives"

Workflow Routing

Trigger	Workflow	Time
Full structured debate (3 rounds)	workflows/debate.md	30-90 sec
Quick sanity check (1 round)	workflows/quick.md	10-20 sec

Default to Quick. Use Debate only for positioning changes, pre-investor prep, or decisions that affect canonical files.

The Personas

4 personas, each anchored to real project files. They read the files before speaking -- grounded opinions, not generic advice.

Persona	Perspective	Reads	Core Question
Operator	What we can actually do today	q-system/my-project/current-state.md	"Can we ship this with what we have?"
Buyer	What a CISO/security leader would pay for	q-system/canonical/discovery.md, q-system/my-project/relationships.md	"Would a buyer pay for this? Does it match what they asked for?"
Investor	The fundable narrative	q-system/canonical/talk-tracks.md	"Does this sharpen or dilute the story?"
Contrarian	What's wrong, what we're not seeing	q-system/canonical/objections.md, q-system/my-project/competitive-landscape.md	"What breaks? What are we ignoring?"

Persona Rules

• Every persona speaks from their anchored files, not from general knowledge
• If a canonical file is empty/template, the persona says "I don't have data to ground this" -- no guessing
• Personas reference specific entries from their files when making points
• Contrarian must raise at least one uncomfortable point per round

Output

Every council session produces:

1. Visible transcript -- the full debate, readable
2. Recommended path -- synthesized from convergence
3. Decision log entry -- appended to q-system/canonical/decisions.md with [COUNCIL-DEBATED] origin tag

Decision Log Format

### [Decision summary]
- **Date:** YYYY-MM-DD
- **Origin:** [COUNCIL-DEBATED]
- **Mode:** Quick / Debate
- **Trigger:** [What prompted this]
- **Convergence:** [Where personas agreed]
- **Dissent:** [Where they disagreed]
- **Decision:** [What the founder chose]
- **Founder action:** [APPROVED / MODIFIED / REJECTED] the recommendation

The founder always makes the final call. Council recommends, founder decides.

Integration with Q System

Q Command	Council Role
/q-calibrate	Quick check before any canonical file change
/q-debrief with conflicting signals	Quick check: "signal or noise?"
/q-plan with competing priorities	Debate: which path to take
Pre-investor meeting	Debate: stress-test the narrative
Design partner feature request	Quick check: "build / park / counter-offer"

Token Budget

• Quick: ~20K tokens (4 agents x 1 round). Fine for daily use.
• Debate: ~60K tokens (4 agents x 3 rounds). Use for real decisions only.

Before running a Debate, ask: "Is this decision worth 60K tokens?" If the answer is "probably not," use Quick.