القائمة
Low-token router for CodexScientist in Codex CLI. MCP-only default for research-control workflows after Codex has entered ordinary or goal context.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Write CodexScientist-aware implementation, experiment-roadmap, code-only, and formal-command plans with TDD and durable quest outputs.
Use when a quest needs one or more follow-up runs such as ablations, robustness checks, error analysis, or failure analysis after a main experiment.
Use when a quest needs to attach, import, reproduce, repair, verify, compare, or publish a baseline and its metrics.
Use when the quest needs an explicit go, stop, branch, reuse-baseline, write, finalize, reset, or user-decision transition with reasons and evidence.
Execute CodexScientist project experiment command documents with durable cs_* logging, manifest validation, and cautious baseline-gate handling.
Use when a quest is ready for a concrete implementation pass or a main experiment run tied to a selected idea and an accepted baseline.
| name | codexscientist-codex |
| description | Low-token router for CodexScientist in Codex CLI. MCP-only default for research-control workflows after Codex has entered ordinary or goal context. |
This skill is a thin router. It states policy and boundaries; it is not a full command manual.
cs_* MCP surface for repeated research-control workflows./goal is Codex-native. CodexScientist does not implement slash commands; after Codex has entered goal context, this router maps research semantics to MCP calls, project-local state, checkpoint, and resume contracts.tools/list, then call cs_tool_schema for the one tool you intend to use.<project>/CodexScientist/.cs_* calls when they affect root-bound research state, memory, artifacts, baselines, experiments, reviews, analysis, method improvement, or evidence.cs_status then cs_resume_brief with the default 4K-8K budget; use cs_pack_delta only when changes after a checkpoint are too large for the brief.cs_log_digest before reading raw logs, and cs_artifact_index before opening artifact files.cs_checkpoint so later turns can resume without chat history.The default mode is copilot.
In copilot mode, CodexScientist records, checks, organizes, retrieves, and summarizes research state. It may check novelty or duplicate risk for a user-provided or document-provided idea, but it must not own the research direction.
autonomous_idea_improvement is disabled by default. Enable it only when the user explicitly asks for automatic idea/novelty improvement, or when a project manifest or handoff explicitly requires autonomous idea improvement. Otherwise, do not invent or improve ideas automatically; output candidate plans for user review instead of creating new running experiments.
execution-grounded work is opt-in. automatic idea search, variant implementation, run scheduling, and executor-backed operations require an explicit user request or manifest. In normal copilot use, record evidence and produce plans; do not submit experiments or create new executor-backed runs from this router.
Use execution_planning only for plan-first environment, feedback, trajectory, and evolutionary planning tools after they exist. Use executor_local only when the user or manifest explicitly authorizes local executor work and the tool reports that the gate is open. The default path stays MCP-visible, fail-closed, and non-executor.
Codex-native operation layer:
CodexScientist semantic/provenance layer:
Codex does the mechanical action; CodexScientist records the research meaning. Use cs_bash_exec only when the command itself must be auditable CodexScientist provenance, not as a general shell replacement.
From the target project root, initialize and inspect the MCP surface with CodexScientist MCP smoke helpers or direct MCP initialize / tools/list. Then follow this MCP-first flow:
cs_status or cs_doctor.cs_record_user_requirement.quest_id only as provenance validation, not as a path selector.evidence, formal_run, literature, or paper_write.cs_tool_schema for the selected tool before using it when arguments are not obvious.Prefer compact status, bounded log tails, artifact paths, hashes, and summaries. Do not inject full logs, full JSONL ledgers, full papers, or full reference repositories into Codex context unless an explicit raw/range read is required. Full support-skill text and raw artifact content are explicit opt-in paths, not defaults.
Report: