multi-lens
Multi-Lens
Meta — Dynamic Multi-Agent. View a problem through multiple perspectives via debate or polling.
Core Question: "What do multiple expert perspectives converge on — and where do they genuinely disagree?"
Critical Gates — Read First
- Choose the right mode — debate for trade-off decisions, poll for filtering hallucinations and finding consensus. If unsure, default to debate (richer output for fewer agents).
- Problem must be specific — N agents on a fuzzy prompt wastes tokens. If the problem is vague, ask the user to sharpen it before spawning.
- Agents must produce structured output — freeform prose can't be aggregated. Every agent returns a defined format.
- Cost scales with agent count — 3 debate agents × 3 rounds ≈ $0.30-0.50. 10 poll agents ≈ $0.30-0.50. Default to sonnet unless user requests opus.
Inputs Required
- A specific problem, decision, or question from the user
- Optional: agent count, round count, preferred mode (debate/poll), custom roles
Output
.agents/meta/multi-lens-report.md— synthesis with consensus, disagreements, recommendation
More from hungv47/meta-skills
task-breakdown
Decomposes a spec or architecture into buildable tasks with acceptance criteria, dependencies, and implementation order for AI agents or engineers. Produces `.agents/skill-artifacts/meta/tasks.md`. Not for clarifying unclear requirements (use discover) or designing architecture (use system-architecture). For code quality checks after building, see fresh-eyes.
79discover
Conversational discovery — adapts from quick scoping (3-5 questions) to deep interviews (multi-round). Talk until we're clear, then build. Produces inline decisions; optionally saves spec.md or scope contract. Not for multi-perspective debate (use agents-panel). Not for decomposing work (use task-breakdown). Not for diagnosing a known metric decline or root-causing a problem (use diagnose).
76agent-room
Multi-agent discussion rooms — debate or poll a problem from multiple perspectives. Standalone or invoked by other skills as a sub-routine. Mode=debate: N agents argue in rounds, converge. Mode=poll: N agents independently analyze, aggregate by consensus. Not for implementation (use system-architecture). Not for verification (use review-chain). For clarifying requirements first, see discover. For decomposing work after a decision, see task-breakdown.
67review-chain
Post-implementation quality check via fresh-eyes review. Chain: Implement → Review (independent agent) → Resolve (if issues). Max 2 rounds. Auto-triggers for security-sensitive and data-mutation code. Not for code refactoring (use code-cleanup). Not for decision analysis (use agent-room).
61navigate
Artifact status + multi-phase orchestration. Scan what exists, check freshness, compose and track complex workflows across sessions. Not for skill routing (the agent does that proactively).
57fresh-eyes
Post-implementation quality check via fresh-eyes review. Chain: Implement → Review (independent agent) → Resolve (if issues). Max 2 rounds. Auto-triggers for security-sensitive and data-mutation code. Not for code refactoring (use code-cleanup). Not for decision analysis (use agents-panel).
16