Small Council

For M-sized tasks and above, the skill convenes a "small council" of perspectives to validate the estimate before output.

Council Members

Member	Role	Perspective
Optimist	Best-case analyst	What if everything goes right? Looks for parallelizable work, reducible scope
Skeptic	Risk analyst	What's missing? Hidden dependencies, integration surprises, underestimated review
Historian	Calibration analyst	How do similar past tasks compare? Reference stories, velocity data

Trigger Rules

Task Size	Council	Rationale
S	None	Overhead exceeds value of validation
M	Skeptic only	Quick risk check
L	Skeptic + Historian	Risk check + historical comparison
XL	Full council	All three perspectives for high-stakes estimates

How It Works

1. The estimate is computed normally (Phase 3)
2. Council members review the estimate from their perspective
3. Each gives 2-3 sentences of analysis
4. A consensus estimate is produced weighing all perspectives
5. Disagreements >20% are flagged for the user

Example Output (XL Task)

Small Council Review:

Optimist: The collaboration features could leverage existing WebSocket
infrastructure, reducing integration time by ~20%. If the team has done
real-time sync before, the learning curve is minimal.

Skeptic: Conflict resolution algorithms are notoriously tricky — the
30% agent effectiveness for XL tasks may be optimistic here. External
API dependencies for document storage add hidden integration risk.

Historian: Similar real-time collaboration tasks in our reference stories
averaged 1.3x the initial PERT estimate. The data-heavy tasks in this
domain tend to underestimate human fix time.

Consensus: Adjusting estimate upward by ~15% based on historical pattern
and conflict resolution complexity. Flagging integration risk.

Implementation

The council uses the Agent tool to spawn subagents when available (e.g., in Claude Code). In environments without subagent support, the perspectives are simulated inline.