메뉴
Use when a task needs the right repository, file and document context assembled without unnecessary noise.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Use when reviewing code before merge to assess correctness, tests, maintainability, security and user impact.
Use when reviewing plans, recommendations, prioritisation, risk registers, retrospectives or agent outputs for judgement distortion from cognitive bias — not for checking whether an argument's premises validly support its conclusion.
Use when evaluating whether premises in plans, ADRs, reviews, security justifications or agent recommendations validly support their conclusions — for logical fallacies, not judgement distortion from cognitive bias.
Use when a coding agent needs robust problem solving, explicit decomposition, alternative-path exploration, program-aided reasoning, ReAct loops, or self-correction.
Use when generated code, plans, tests, prompts or architecture need critique, repair and verification before completion; especially for coding tasks requiring self-review, test-driven repair or quality gates.
Identifies risks, scores likelihood and impact, selects treatment, assigns owners and review cadence. Use when building or updating a risk register, treating agent autonomy risks, or deciding mitigate, transfer, accept or avoid.
| name | context-engineering |
| description | Use when a task needs the right repository, file and document context assembled without unnecessary noise. |
| aliases | ["context-selection","context-assembly"] |
Use this skill when starting a session, switching tasks, or preparing context for a workflow that depends on the right local files and guidance.
Load enough context to work accurately while avoiding irrelevant material that dilutes attention or exceeds the useful scope.
For a validator change, load the validator, its tests, the library contract and only the skill files affected by the new rule. Do not load unrelated UX or reliability skills unless the failing evidence points there. If a later test reveals more affected files, refresh context and record why the scope expanded.
