code-reviewer is a Codex skill for reviewing a pull request or branch as one
full diff. It re-checks earlier findings, analyzes resource and server-load
costs, prioritizes crash and service-wide availability risks, and produces an
evidence-backed Markdown report.
- Correctness, authorization, data loss, concurrency, durability, and migration safety.
- CPU, memory, disk, database/WAL, network, thread, scheduler, and external-service costs.
- Crash/startup-loop and service-wide outage paths as the highest-priority blockers.
- Conditional occurrence ratios and production trigger estimates.
- Previously reported findings as
FIXED,PARTIALLY_FIXED,NOT_FIXED, orREGRESSED.
This is the recommended installation method because it also makes future updates easy.
mkdir -p "$HOME/.codex/skills"
git clone https://github.com/nsqning/code-reviewer.git \
"$HOME/.codex/skills/code-reviewer"The installed structure should be:
~/.codex/skills/code-reviewer/
├── SKILL.md
├── agents/
│ └── openai.yaml
└── references/
└── report-format.md
Restart Codex or start a new Codex conversation so the skill is rediscovered.
- On GitHub, select Code → Download ZIP.
- Extract the archive.
- Copy the extracted folder to
~/.codex/skills/code-reviewer. - Confirm that
SKILL.mdis directly inside that folder, rather than inside an additional nested directory. - Restart Codex or start a new conversation.
On Windows, the equivalent destination is:
%USERPROFILE%\.codex\skills\code-reviewer
On macOS or Linux:
test -f "$HOME/.codex/skills/code-reviewer/SKILL.md" \
&& echo "code-reviewer is installed"You can also verify it from Codex by starting a new conversation and invoking:
$code-reviewer
Invoke the skill and provide a pull request URL, branch, or local checkout:
Use $code-reviewer to review this pull request as one full diff:
https://github.com/OWNER/REPOSITORY/pull/123
For a repeated review after fixes land:
Use $code-reviewer to pull the latest PR commits, re-check every finding from
the previous Markdown report, analyze resource costs and crashability, and
write a new review-latest-<sha>.md report.
The skill does not modify product code, push branches, or post pull-request comments unless the user separately requests those actions.
If installed using Git:
git -C "$HOME/.codex/skills/code-reviewer" pull --ff-onlyRestart Codex or start a new conversation after updating.
- The skill does not appear: Verify that
~/.codex/skills/code-reviewer/SKILL.mdexists, then restart Codex or open a new conversation. - The clone destination already exists: Update the existing Git checkout with the command above instead of cloning again.
- The ZIP created a nested directory: Move the contents so that
SKILL.mdis directly under thecode-reviewerdirectory. - The skill does not activate automatically: Invoke it explicitly with
$code-reviewer.
SKILL.md— workflow, review lenses, crashability priorities, and evidence rules.references/report-format.md— required Markdown report structure and fields.agents/openai.yaml— Codex display metadata and default invocation prompt.
Before installing a skill from any repository, review its instructions and
supporting files. OpenAI documents skills as directories containing a required
SKILL.md plus optional references, scripts, and assets:
Build skills.