What does this skill do, and when should you use it?
This skill defines rules for maintaining the model context used by Codex. Message history must grow incrementally rather than being rewritten. Every injected item needs a bounded size and hard cap, with no individual item larger than 10K tokens. New items that may exceed 1K tokens must be marked P0 for additional manual review.
It specifies context-management constraints: preserve history, avoid frequent context changes that cause cache misses, bound every injected item, and require injected fragments to be structs in core/context that implement the ContextualUserFragment trait.
- Codex engineers maintaining a context pipeline that must preserve incremental message history.
- Developers optimizing prompt-cache stability by limiting unnecessary context changes.
- Engineers designing context-fragment data structures that need a shared type and trait contract.
- Teams reviewing large context additions that may cross the 1K-token P0 threshold.
How do you install this skill?
- This file reads more like an internal design constraint than a directly callable skill; implementation location, invocation, input/output contracts, and failure handling are missing.
- It does not explain how credentials, personal data, or secrets in context are handled, nor does it define confirmation or rollback mechanisms.
- The required trait is named without API, version, or test evidence, so integration feasibility is unverified.
The skill is located at openai/codex/.codex/skills/code-review-context/SKILL.md; no standalone installation procedure is documented. To install the Codex CLI collection, run curl -fsSL https://chatgpt.com/codex/install.sh | sh on macOS or Linux, or powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex" on Windows. It is also available through npm install -g @openai/codex.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/openai/codex.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/.codex/skills/code-review-context" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Read and apply it as a repository-level specification for context design. The source does not document a standalone trigger command, parameters, example prompt, or separate integration workflow.
What are this skill's strengths and limitations?
- Sets a clear 10K-token maximum for individual injected items.
- Addresses correctness, cache stability, and human review together.
- Requires a consistent `core/context` struct and trait implementation.
- Provides constraints but no implementation code or migration procedure.
- Does not explain token counting, hard-cap enforcement, or the P0 review process.
- The source provides no independent test suite or runtime validation evidence for this skill.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Context Governance this page ✓ OpenAI · Official | 35 · Not recommended | ★ 129k | 11d ago | Apache-2.0 |
| Agents Best Practices | 61 · Recommended | ★ 2.4k | 12d ago | MIT |
| Handoff — Session Handoff Skill for Claude Code | 52 · Use with care | ★ 64 | 8d ago | MIT |
| Codebase Design: Deep Module Design Guide | 57 · Use with care | ★ 291k | 16d ago | MIT |
| NeMo Relay Adaptive Tuning ✓ NVIDIA · Official | 57 · Use with care | ★ 3.6k | 11d ago | Apache-2.0 |
How did FollowSkills review this skill?
The skill requires incremental context construction, forbids history rewriting, imposes a 10K-token item cap, and requires ContextualUserFragment structs, showing some boundary and capacity controls. It does not specify user confirmation, sensitive-data handling, isolation, data-flow disclosure, rollback, or external-effect controls, so points are deducted.
The rules are broadly consistent and include hard limits, but there is no implementation, dependency guidance, abnormal-input handling, failure feedback, or key-path test evidence. The score remains within the static-review ceiling.
The name and body identify model-visible context as the target area, but inputs, outputs, trigger conditions, non-fit boundaries, Chinese-language support, and environment fit are unspecified. This limits accurate invocation.
The file is concise and readable with explicit rules, but lacks installation or dependency notes, examples, FAQs, known limitations, versioning, changelog, maintenance ownership, and update instructions. The repository license and stated OpenAI provenance provide governance context but do not fill these skill-level gaps.
The rules address context growth, cache misses, item-size limits, and structural integration, so they may provide engineering value. However, no implementation or verifiable output is supplied, so task completion cannot be established and substantial engineering work remains.
The text is auditable as a static policy, but contains no test suite, CI coverage, third-party execution evidence, or corroborating sources. The assessment therefore relies on a single file and receives a very low score.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →