metacognitive-guard

Metacognitive Guard

Awareness of struggle detection and guidance on when to engage deep-thinking resources.

When to Self-Escalate

Consider spawning deep-think-partner when:

1. Architectural decisions — multiple valid approaches, trade-offs span different dimensions
2. Ambiguous requirements — multiple reasonable interpretations, wrong choice has high rework cost
3. Multi-domain synthesis — problem spans multiple technology areas, prior art doesn't directly apply
4. Edge case analysis — happy path is clear but failure modes need systematic exploration

Self-Assessment

Before responding to complex questions:

• Can I give a concrete recommendation (not "it depends")?
• Do I have high confidence in my answer?
• Is this answerable without multiple follow-up exchanges?
• Would structured analysis add significant value?

If "no" to any — escalate.

Escalation Protocol

Task tool:
  subagent_type: deep-think-partner
  prompt: |
    Context: [system/codebase]
    Constraints: [solution limits]
    Success criteria: [how to judge]
    Question: [specific decision needed]
  description: [3-5 word summary]

Struggle Signals

Signal	Better Approach
Hedging	Escalate for deeper analysis
Deflecting	Answer then ask clarifying questions
Verbose without concrete output	Structure response with code/tables
Contradiction mid-response	Stop, think, give one coherent answer
Non-committal	Make a recommendation with confidence level

Using Deep-Think Results

1. Synthesize for the user (don't paste raw output)
2. Highlight the key non-obvious insight
3. Present the recommendation clearly
4. Offer the implementation plan