code-review
Review a committed range or complete work-in-progress candidate against repository standards and its governing contract.
Skill by luizomf
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/luizomf-code-review
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: documents
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md and code. Checked Oct 26, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/code-review
What's inside
- Skill name
code-review- Description
- Review a committed range or complete work-in-progress candidate against repository standards and its governing contract.
- License
- MIT
- Instructions
- 216 lines in SKILL.md · view on GitHub
Files (1)
- SKILL.md
SKILL.md (first lines)
# Code Review Read applicable repository instructions and domain documents. Read issue-tracker configuration only when the review contract is tracked; an untracked direct request uses its confirmed conversation as the contract and does not require tracker setup. If required configuration is unavailable, follow [setup-omskills](../setup-omskills/SKILL.md) and its scoped authorization gate before review. A read-only reviewer returns the missing prerequisite to its responsible caller instead of writing setup. Review exactly one candidate in one of two modes: - **Committed:** a fixed base commit through `HEAD`. - **WIP:** the current staged, unstaged, and untracked worktree state. Honor an explicit mode. Otherwise infer committed mode for a branch, PR, or supplied fixed point and WIP mode for a worktree request. Ask only when both remain plausible; never combine them into a partial hybrid. Review against two separately reported criteria sets: - **Standards:** applicable repository instructions and conventions, maintainability, and relevant code smells. - **Spec:** the exact Ticket, accepted behavior, acceptance criteria, omissions, incorrect behavior, and changes outside scope. ## Adversarial review contract Defend the project's interests, not the author's solution or the status quo. Challenge decisions that compromise correctness, simplicity, or maintainability against accepted requirements and evidence. The normal path is one review followed by author corrections and verification: complete the investigation now instead of relying on a later pass. - Inspect every candidate path and contract requirement, reading surrounding context and tracing affected callers, consumers, tests, and governing sources wherever needed to establish correctness. Challenge assumptions, failure and boundary cases, regressions, omissions, and contradictory instructions that apply to this candidate; passing checks and author claims are evidence to examine, not substitutes for inspection. - Challenge unnecessary complexity as part of Standards: premature abstractions, pass-through layers, speculative generality, extra configuration or dependencies, and redundant workflow steps. For each complexity finding, show the concrete maintenance, reasoning, testing, or operational cost and a substantially simpler alternative that preserves required behavior and constraints. Evaluate total complexity, including what callers must absorb; fewer lines or a personal preference alone is not evidence. Apply this to code and governing documents without turning candidate review into an unrelated redesign. - Continue across the complete candidate after finding a blocker. Return every concrete, supported problem found, including non-blocking ones, without a finding quota or a top-findings cutoff. Consolidate duplicates only when all affected locations and distinct consequences remain explicit. - State defects, impact, and severity directly, without sugar-coating, praise padding, or downgrading a finding to soften the verdict. Be adversarial toward the work, not the author. Evidence determines severity; inventing requirements, speculative defects, or stylistic preferences is not adversarial review. - For every finding, identify Standards or Spec, severity/blocking status, file/line (or exact artifact section), the violated rule or expected behavior, supporting evidence, and consequence. Distinguish evidence gaps from confirmed defects. Report relevant discovered problems outside the authorized correction scope as findings without expanding the investigation into unrelated work. The pass is complete only when every candidate path and applicable requirement has been examined, unnecessary complexity has been assessed, and every supported finding is reported. State coverage, including the simplification assessment, and verification limitations explicitly; an unexamined required area makes the review incomplete, not clean. A complete pass with no supported findings reports that result without manufacturing faults or claiming proof of defect-free work. ## Prepare the complete candidate ### Committed mode 1. Use the fixed point supplied by the caller. If none was supplied, infer one and ask only when none can be established. Resolve review HEAD to a full SHA. For standalone committed review, resolve the base as the merge base of that fixed point and review HEAD, preserving the `fixed-point...HEAD` candidate. For coordinated review, use the supplied exact full base SHA and verify it is an ancestor of review HEAD. 2. In a coordinated Ticket review, verify the supplied candidate path, branch and exact HEAD before capture. Use that exclusive candidate, never an inherited caller checkout; do not create another workspace. Unexpected branch/HEAD drift stops review as incomplete. 3. Capture the complete `git diff <base-sha> <review-sha>` and `git log <base-sha>..<review-sha> --oneline` in that candidate. Check that its identity/HEAD remains unchanged after capture; report exact path, branch, base and review SHAs with any capture limitations. 4. If the diff is empty, stop and report that there are no committed changes to review. ### WIP mode 1. Capture staged changes with `git diff --cached`. 2. Capture unstaged changes with `git diff`. 3. Inventory every untracked path and capture its complete content or binary status without treating ignored files as candidate work. 4. Report every unreadable or unrepresentable path as a capture limitation. If all three parts are empty, stop and report that there is no WIP candidate. For either mode, locate the current contract and every applicable repository instruction, governing source, and standard. The contract may be a tracked Ticket or Spec, or the concise confirmed request for untracked direct work. If no governing contract exists, review Standards and observable correctness while stating that contract compliance could not be verified. ## Select the caller-safe review path This skill requires one isolated review pass; assigning a reviewer name does not create isolation, read-only behavior, tools, or delivery semantics. Direct code or behavior changes and changes to Specs, ADRs, workflow, security, or other governing authority require this fresh independent pass. Purely editorial documentation may instead be self-reviewed. Before using disposable notes or transport files, read [Working files](../../productivity/WORKING-FILES.md); read-only review protects the candidate, not the reviewer's own working files. - An interactive responsible agent, including a Ticket owner, may use the active harness's documented asynchronous delivery or [tmux-worker](../../productivity/tmux-worker/SKILL.md) for a visible isolated
From luizomf/omskills, MIT.
More from luizomf/omskills
- 0.609score
diagnosing-bugs
luizomf
Reproduction-first diagnosis loop for hard bugs and performance regressions.
SkillAFreshness: A★ 3Claude Code +9 - 0.604score
research
luizomf
Investigate a bounded question or produce a durable cited research artifact.
SkillAFreshness: A★ 3Claude Code +9 - 0.604score
dispatch-tickets
luizomf
Coordinate a multi-issue Mission by tracking dependencies, responsible agents, blockers, and verified deliveries.
SkillAFreshness: A★ 3Claude Code +9 - 0.604score
codebase-design
luizomf
Design or improve deep modules.
SkillAFreshness: A★ 3Claude Code +9 - 0.604score
setup-omskills
luizomf
Configure repository instructions, issue-tracker operations, triage-label mappings, and domain-doc locations.
SkillAFreshness: A★ 3Claude Code +9 - 0.604score
orchestrate
luizomf
Carry a bounded change through implementation, independent review, and verified delivery in the current conversation.
SkillAFreshness: A★ 3Claude Code +9 - 0.601score
wayfinder
luizomf
Plan work that exceeds one agent session as a shared map of investigation tickets, then resolve one ticket per session until the route to the destination is…
SkillAFreshness: A★ 3Claude Code +9 - 0.601score
to-tickets
luizomf
Break a plan, spec, or conversation into tracer-bullet tickets with explicit blocking and conflict edges, then publish them to the configured tracker.
SkillAFreshness: A★ 3Claude Code +9 - 0.601score
triage
luizomf
Categorize and verify issues or external pull requests, then move them through configured triage states and record the applicable brief or notes.
SkillAFreshness: A★ 3Claude Code +9 - 0.601score
grill-with-docs
luizomf
Stress-test a repository plan or design through bounded Question rounds while updating domain terms and durable architecture decisions.
SkillAFreshness: A★ 3Claude Code +9 - 0.601score
to-spec
luizomf
Synthesize the current conversation and repository context into a spec, then publish it to the configured issue tracker without a requirements interview.
SkillAFreshness: A★ 3Claude Code +9 - 0.599score
resolving-merge-conflicts
luizomf
Resolves an in-progress git merge or rebase conflict by understanding each change's intent and preserving both where possible.
SkillAFreshness: A★ 3Claude Code +9
Signals
- Stars
- 3
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Oct 16, 2026 (13 days ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/luizomf/omskills
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from luizomf
- 0.609score
diagnosing-bugs
luizomf
Reproduction-first diagnosis loop for hard bugs and performance regressions.
SkillAFreshness: A★ 3Claude Code +9 - 0.604score
dispatch-tickets
luizomf
Coordinate a multi-issue Mission by tracking dependencies, responsible agents, blockers, and verified deliveries.
SkillAFreshness: A★ 3Claude Code +9 - 0.604score
research
luizomf
Investigate a bounded question or produce a durable cited research artifact.
SkillAFreshness: A★ 3Claude Code +9 - 0.604score
codebase-design
luizomf
Design or improve deep modules.
SkillAFreshness: A★ 3Claude Code +9 - 0.604score
orchestrate
luizomf
Carry a bounded change through implementation, independent review, and verified delivery in the current conversation.
SkillAFreshness: A★ 3Claude Code +9 - 0.604score
setup-omskills
luizomf
Configure repository instructions, issue-tracker operations, triage-label mappings, and domain-doc locations.
SkillAFreshness: A★ 3Claude Code +9
Cite this page
code-review skill. Agent Mart. https://agentmart.dev/skills/luizomf-code-review (last updated 2026-10-24)