CtrlK
BlogDocsLog inGet started
Tessl Logo

metacognition

Routes a reasoning task to the right pattern (Structured Reasoning, Skeleton-of-Thought, Chain-of-Code, Tree of Thought, Committee/Panel) and applies confidence scoring, bias scan, and verify-the-verifier. [EXPLICIT] Meta-skill that operationalizes the metacognition canon. Trigger: 'metacognition', 'reasoning pattern', 'how should I think about', 'tree of thought', 'committee', 'second opinion', 'high-stakes decision'.

SKILL.md
Quality
Evals
Security

Metacognition

"Pick how to think before you think. The wrong reasoning shape costs more than a slow answer." [INFERENCE]

TL;DR

Selects the reasoning pattern a task needs — Structured Reasoning · Skeleton-of-Thought · Chain-of-Code · Tree of Thought · Committee/Panel — then applies the cross-cutting supports ([CONFIANZA: x] scoring, bias scan, verify-the-verifier). It is a router, not a re-implementation: the canon lives in orchestration/metacognition-protocol.md; the executable map in docs/METACOGNITION.md; this skill turns a task into the right pattern + the skill that runs it. Loads only the pattern playbook the task needs (ICM layered context). [EXPLICIT]

When to use

  • You are unsure how to think about a task (which reasoning shape fits). [EXPLICIT]
  • A decision is high-stakes, ambiguous, or has no obvious winner (→ ToT or Committee). [EXPLICIT]
  • You need a second opinion or to check your own reasoning for bias before committing. [EXPLICIT]
  • NOT for a direct one-step answer (just answer) and NOT for parallelism mechanics — that is agentic-parallelism-decision. [INFERENCE]

Selection (route by task shape + confidence)

Task shapePatternRuns via
análisis / diagnósticoStructured Reasoning (DECOMPOSE→SOLVE→VERIFY→SYNTHESIZE→REFLECT)the skill's own Procedure
documento / entregableSkeleton-of-Thoughtscaffold_skill.py --rich, outline-first
workflow / decisión condicionalChain-of-Codedynamic-workflow-forge, katas-deterministic-agent-loop
sin ganador obvio / alto costo de errorTree of Thought → references/tree-of-thought.mdsocratic-debate, trade-off-analysis
high-stakes / ambiguo / multi-dominio / disputaCommittee/Panel → references/committee-panel.mdtriada agents/{lead,support,guardian}, agent-team-orchestration

Confidence gate: score the decision [CONFIANZA: x] first. ≥0.8 → a single pattern is enough · 0.5–0.7 → escalate to ToT or Committee · <0.5 → escalate to the user, do not proceed on assumption. [DOC]

Procedure

  1. Classify the task shape and score [CONFIANZA: x]. [EXPLICIT]
  2. Route to the pattern via the table; for ToT/Committee open the matching references/ playbook (L3, on-demand — do not load both). [EXPLICIT]
  3. Run the pattern through the skill it maps to; ToT generates branches, the Committee judges them, Structured Reasoning synthesizes. [EXPLICIT]
  4. Bias scan before committing: anchoring / confirmation / availability / sunk-cost; if detected, seek counter-evidence. [DOC]
  5. Verify with an independent checker (Guardian ≠ Lead, Art. 9); record minority dissent even when overruled. [EXPLICIT]

Quality Criteria

  • Pattern chosen by task shape + confidence, not by habit. [INFERENCE]
  • For wide solution spaces, ≥2 independent branches were generated (not variants of one). [DOC]
  • High-stakes decisions passed a multi-perspective panel; the verifier ≠ the producer. [DOC]
  • Conclusion carries [CONFIANZA: x]; <0.5 escalated, not assumed. [DOC]
  • Bias scan run; claims evidence-tagged. [EXPLICIT]

Anti-Patterns

  • One reasoning shape for everything (Structured Reasoning on a wide-open design → premature convergence). [INFERENCE]
  • ToT "branches" that are three rewordings of the same idea (no real diversity → anchoring survives). [DOC]
  • Self-verification (the producer grades its own output → confirmation bias). [DOC]

Contract

  • Aceptación: tarea clasificada + [CONFIANZA: x]; patrón ruteado; ToT con ramas independientes; high-stakes con panel + verify-the-verifier; bias scan corrido; disidencia registrada. [EXPLICIT]
  • Límites: rutea y aplica patrones de razonamiento; NO reimplementa el canon (orchestration/metacognition-protocol.md) ni mecánica de paralelismo (agentic-parallelism-decision). [EXPLICIT]
  • Casos borde: dos ramas empatan → Comité o [CONFIANZA: x]+usuario; confianza <0.5 → escalar, no proceder. [DOC]
  • Supuestos: las skills destino (socratic-debate, agent-team-orchestration, katas) están disponibles. [SUPUESTO]
  • Trade-off: ToT/Comité cuestan más tokens/tiempo a cambio de menos error en decisiones caras; por eso el confidence gate los reserva para <0.8. [EXPLICIT]

Related Skills

  • socratic-debate — la encarnación de Tree of Thought (tesis→antítesis→síntesis). [EXPLICIT]
  • agentic-parallelism-decision — mecánica de ejecución paralela (complementaria, no metacognición). [EXPLICIT]
  • agent-team-orchestration — ejecuta el modo Committee/Panel. [EXPLICIT]
  • trade-off-analysis, feasibility-validation — consumidores típicos del routing ToT. [EXPLICIT]

Packet

Capas del packet, cargables bajo demanda (disciplina ICM: una capa por vez, nunca todas juntas): references/ guías de profundidad (cargar UNA por etapa) · knowledge/ cuerpo de conocimiento · prompts/ prompts listos · examples/ salida de ejemplo · agents/ subagentes del packet · assets/ recursos estáticos.

Repository
JaviMontano/claude-plugins
Last updated
First committed

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.