metacognitive-guard

Dispatch two competing reviewers (arch-reviewer and impl-reviewer) before deep analysis. Competition produces more thorough results. Use before creating code, modifying architecture, or making technical decisions.

Structured 4-phase analysis with transparent reasoning and adversarial self-review. Use for complex decisions, architecture choices, debugging dead-ends, or any problem where jumping straight to implementation would be premature.

Force verification before answering questions involving versions, dates, status, or "current" state. Prevents hallucinations at the REASONING level by checking assertions.yaml and WebSearch before forming beliefs.

Monitors Claude's responses for struggle signals and suggests escalation to deep-thinking agents when complexity exceeds comfortable reasoning capacity.

Force verification before claiming success or completion. Prevents false "it works" claims. Triggers when about to say "done", "complete", "works", "fixed", or "the implementation is ready".

Architecture-focused competitive reviewer. Finds structural problems like dependency violations, SSOT violations, layer boundary issues, coupling problems, and SOLID principle violations. Competes with impl-reviewer - whoever finds more valid issues gets promoted.

Asynchronous reasoning amplifier for complex problems requiring deep analysis. Use when facing multi-step logical challenges, architectural trade-offs, strategic decisions, or when you need a second brain to validate reasoning chains. Runs in background to multiply thinking capacity. Examples: <example> Context: Complex architectural decision with competing constraints user: "Should we use event sourcing or CRUD for order management? High reads, complex rules." assistant: "This has significant trade-offs across multiple dimensions. Let me spawn a deep-think-partner to analyze implications while I gather more context." <commentary>Strategic technical decisions with long-term implications warrant deep collaborative analysis.</commentary> </example> <example> Context: Multi-step data transformation with edge cases user: "Transform nested JSON with circular refs to flat relational schema, preserving integrity." assistant: "Multiple competing constraints here. Spawning deep-think-partner in background to explore the solution space." <commentary>Complex data problems with interdependencies benefit from parallel deep reasoning.</commentary> </example> <example> Context: Non-deterministic concurrency bug user: "Race condition only under load. Optimistic locking but seeing phantom writes." assistant: "Race conditions require methodical state transition analysis. Let me engage deep-think-partner to reason through timing scenarios." <commentary>Concurrency debugging needs systematic exploration of state spaces.</commentary> </example>

Implementation-focused competitive reviewer. Finds code-level issues like banned API usage, version mismatches, wrong assumptions, and fact-checks claims using WebSearch. Competes with arch-reviewer - whoever finds more valid issues gets promoted.

5 hooks across 3 events