Breaking Change Reviewer
Systematically detect compatibility risks across external integration surfaces before they reach users.
The skill limits its stated purpose to reviewing breaking changes in external integration surfaces, with no explicit malware, exfiltration, or destructive behavior; no red-line risk is evidenced. However, it specifies no least-privilege model, user confirmation, data-flow disclosure, sensitive-data handling, rollback, or result attribution, so 11 points are deducted.
It identifies four concrete surfaces—app-server APIs, CLI parameters, configuration loading, and session resumption—and is internally consistent. It provides no procedure, input/output contract, dependency guidance, abnormal-input handling, or failure feedback, so 15 points are deducted under static review.
The intended review scenario and main surfaces are reasonably clear. Trigger semantics, non-fit boundaries, expected inputs and outputs, Chinese-language support, and variation across repository layouts are unspecified, so 8 points are deducted.
The name and purpose broadly match breaking-change review, while repository-level Apache-2.0 licensing and official provenance provide limited governance evidence. The skill lacks installation notes, dependencies, examples, FAQs, limitations, versioning, changelog, maintenance ownership, and update path, so 11 points are deducted.
The instruction to continue beyond the first finding can improve review completeness. However, there is no report format, decision rubric, or directly usable output contract, and no statically verifiable result, so only partial core-task value is supported and 10 points are deducted.
The skill contains only a short behavioral instruction, with no tests, sample outputs, execution evidence, or corroborating sources. Traceability and reproducibility are therefore weak, so 8 points are deducted.
- The skill defines no output structure or severity rubric, leaving the user to design the review method and report.
- The phrase “external integration surfaces” may be too broad; it does not clarify whether to inspect code, documentation, migrations, compatibility behavior, or only interface definitions.
- This static review did not execute the skill or verify detection of real breaking changes; repository tests must not be treated as tests of this skill itself.
- Handling boundaries for sensitive configuration, session data, and external-service information are unspecified.
What it does & when to use it
This Agent Skill is designed for software engineering reviews focused on breaking changes in external integration surfaces. It covers app-server APIs, CLI parameters, configuration loading, and resuming sessions from existing rollouts. The instructions explicitly require continuing beyond the first finding and analyzing every plausible breakage path. The SKILL.md contains review guidance only; it does not define installation, invocation, automation scripts, or a report schema.
Reviews a change, integration, or upgrade for possible breakage in app-server APIs, CLI parameters, configuration loading, and session resumption from existing rollouts. It is instructed to keep analyzing additional breakage paths after finding an initial issue. The source does not specify a required input format, output format, or execution tool.
- An engineer maintaining an app-server API checks a proposed interface change for client compatibility before release.
- A CLI team evaluates whether changing parameters or command behavior could break existing callers.
- A configuration owner reviews changes to loading logic for incompatibilities with existing configuration.
- A team supporting session recovery checks whether an upgrade breaks resumption from existing rollouts.
- A reviewer wants systematic coverage across several external integration surfaces instead of stopping at one known issue.
Pros & cons
- Names four concrete integration surfaces that are easy to overlook.
- Explicitly requires continued analysis, reducing the chance of stopping after the first issue.
- Uses a minimal SKILL.md with only name and description frontmatter and no visible platform-specific extensions.
- Provides no test suite, scripts, or automated checking command.
- Does not define review inputs, output structure, severity levels, or remediation format.
- Does not document a dedicated installation, invocation, or permissions workflow.
- The source does not establish direct support for other Agent platforms or API-only deployments.
How to install
The Skill is located at .codex/skills/code-review-breaking-changes/SKILL.md. The README documents installation for the full Codex CLI skill collection: on macOS or Linux, run curl -fsSL https://chatgpt.com/codex/install.sh | sh; on Windows, run powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex"; alternatively use npm install -g @openai/codex or brew install --cask codex. The source does not document installing or placing this Skill independently.
How to use
In an Agent that supports the Skill, provide the change, interface, or upgrade under review and ask it to inspect app-server APIs, CLI parameters, configuration loading, and session resumption from existing rollouts. Example: Review this change for breaking changes across app-server APIs, CLI parameters, configuration loading, and resuming sessions from existing rollouts. Analyze every possible breakage path. The SKILL.md does not define a specific trigger command, input format, or report format.