Review

Skill pnascimento9596/codex-buddy-reviewer/skills/review

Run a fresh, independent, read-only review of the current Git working tree or a branch diff through an explicitly selected supported adapter. Use after Codex finishes code work, when the user asks for a second engineer or another perspective, or when they explicitly request a buddy review. Do not use this skill to edit code or install automatic hooks.From its SKILL.md

Install

npx -y skills add pnascimento9596/codex-buddy-reviewer --skill review

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.2 KB, as published. Nobody here has run it

Independent Buddy Review

Run the plugin's manual reviewer in the foreground and present its validated result as an independent second-engineer opinion.

Workflow

Resolve the plugin root as the directory two levels above this file.
Inspect the target repository's Git status and diff summary so the requested scope is understood.
Translate the explicit user request into only the documented flags below. Reject unknown or free-text tokens. Invoke the command with every resolved value passed as a separately quoted argument:
```
node "<plugin-root>/scripts/buddy-review.mjs" review --cwd "<repo-root>" [documented flags]
```
Never splice raw skill arguments or user text into a shell command.
Require an explicit provider selection before any egress. If the user does not name a supported provider, ask which connection to use. Never default to a provider and never silently change the provider used for a previous review.
For GLM through Ollama Cloud, use --provider ollama --model glm-5.2:cloud; for Claude, use --provider claude --model claude-opus-4-8; for OpenCode, use --provider opencode --model <provider/model>.
For a committed branch review, require an explicit base and add --scope branch --base <ref>.
Wait for the foreground process to finish. Present its status, summary, and any findings. Keep clear that these are reviewer findings, not independently proven facts.
Do not modify the reviewed code unless the user separately asks for fixes.

Safety Rules

Never bypass the evidence collector's path policy or send excluded files manually.
Run only after the user explicitly requests a Buddy or independent review. Patch egress must not be triggered implicitly.
Never give the external reviewer extra tools, memory, subagents, or write permission.
A clean no_findings result is valid; never demand findings.
Preserve an abstain result when evidence is incomplete or confidence is too low.
Retain patch evidence in receipts only when the user explicitly requests --retain-evidence.
Do not install hooks, alter Codex configuration, or toggle automatic mode from this manual one-shot skill. Use the separate buddy-review control skill when the user requests automatic review.
Report provider failures exactly. Do not silently switch models or providers.

Useful Arguments

--dry-run --no-store: inspect sanitized evidence metadata without calling a model.
--json: return the validated result as machine-readable JSON.
--confidence <0..1>: change the publication threshold; default is 0.75.
--max-patch-bytes <n>: bound the outbound sanitized patch.
--timeout-seconds <n>: set a hard reviewer deadline.
--no-store: avoid writing a local receipt.

Read references/review-contract.md when explaining the result contract, privacy behavior, or provider boundary.