Use when the user wants to externalize a fuzzy intuition, unnamed judgment, unease about an AI response, or meta-cognitive question into candidate language, a cognitive map, validation probe, grounded principle, prompt, or reusable framework. Trigger when something feels missing or too smooth, the user cannot explain why a framing feels wrong, wants to analyze their thinking or AI collaboration, is overloaded by abstraction, needs to distinguish levels from dimensions, roles, states, or operations, or wants an abstract principle to change concrete action. Do not use for ordinary brainstorming, summarization, domain explanation, or decisions with explicit criteria and output.
Use when a response, document, implementation, or test may dilute a clear user objective or best-supported direction through defensive caveats, excessive disclaimers, imagined edge cases, protective alternatives, one-sided risk weighting, conservative scope reduction, or success criteria that prove only the absence of failure. Preserve the intended outcome, detect when defensibility is replacing evidence or user value as the optimization target, and let only constraints or uncertainties that materially change the decision alter the result.
Use when direct use of a skill from Biaoo/skills reveals something that may help the project improve that skill, or when a fresh independent session receives a sanitized handoff to prepare or submit such feedback. Preserve the active task, report the observation faithfully without deciding how it should be absorbed, isolate feedback preparation from the originating work, and complete a narrowly scoped GitHub submission whenever verified authority and access are available.
Use when choosing or designing an implementation path and the agent may trade away requested quality, fidelity, capability, or completeness to reduce implementation effort, code volume, technical unfamiliarity, or default to KISS or MVP. Encodes the user's preference to give the agent's own execution effort little weight unless it creates a concrete user-relevant consequence or violates an explicit budget. Do not use when the user explicitly prioritizes speed, minimum change, or lowest cost.
Use when designing or changing AI/Agent-native product entities, schemas, database models, JSON schemas, DTOs, events, workspace artifacts, memories, profiles, plans, evaluations, or agent outputs; guides Codex to choose the right degree of formalization and freedom instead of defaulting to either rigid schema or unbounded text.
Use when designing or changing public CLI command behavior, help text, output shape, pagination, JSON modes, file-output behavior, error messages, or Agent-facing next-action guidance; guides Codex to make CLI commands useful as both execution tools and context delivery tools.
Use when reviewing recorded agent runs for avoidable context consumption, comparing quality-qualified matched traces, designing a context-efficiency experiment, or validating the evaluator skill itself. Applies to broad or repeated retrieval, oversized tool results, premature reference loading, weak continuation, hidden cost transfer, and unsupported efficiency claims; do not use for generic context-window explanations, prompt shortening without run evidence, or directly implementing an optimization.
Use when designing or materially reviewing an Agentic content-generation or content-transformation project, especially when its purpose, routes, artifacts, evidence, authority, evaluation, feedback, human intervention, recovery, or adoption boundaries remain unsettled; applies an inquiry-led method and stops before implementation or production authorization.