changocr
User跨 Codex 与 Claude 的可验证 AI agent 工程治理工作台
Categories
Indexed Skills (9)
analyze
Produce evidence-backed quantitative or qualitative analysis with explicit provenance, calculations, denominators, assumptions, uncertainty, and limits. Use when a decision or claim depends on interpreting available data or artifacts.
build
Implement a bounded software or configuration change in this repository with surgical edits, focused tests, privacy safeguards, and current verification evidence. Use after a task contract is approved.
data-pipeline
Build or revise privacy-safe data ingestion, cleaning, joining, and feature pipelines with explicit provenance, grain, keys, null semantics, validation, and output contracts. Use for data engineering tasks involving confidential or large structured data.
orchestrate
Coordinate a multi-step repository task across creators, read-only reviewers, and verifiers. Use when work has dependencies, distinct write scopes, review gates, expensive operations, or cross-session artifact handoffs.
plan-task
Convert an ambiguous software, data, or agent-workflow request into a bounded task contract with scope, interfaces, constraints, risks, and measurable acceptance criteria. Use before implementation when requirements or success conditions are not already decision complete.
plugin-dev
Create, update, synchronize, and validate the repository's Codex plugin and its shared skills. Use when changing `plugins/agents-lab-core`, canonical skills, generated runtime mirrors, or plugin metadata.
review
Run an independent read-only review of code, configuration, data workflows, analysis, documentation, or plugins. Use after creator changes and before completion to apply severity gates and identify unsupported claims or missing tests.
handoff
Create a concise durable handoff for work that crosses agents, clients, sessions, or operators. Use when responsibility changes, execution is deferred, or a task ends with blockers or verified next actions.
verify
Verify execution evidence, tests, generated-file freshness, artifact integrity, privacy checks, and repository readiness without modifying files. Use after review findings are cleared and before claiming delivery.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.