implementation-review
Skill0 starsChanged 46 days agoMITView on GitHub
Use when the user asks to review code changes, a branch, or a pull request, check an implementation, find what's missing, or validate a feature is complete. Use when another skill needs a post-implementation quality gate.
8 other projects use a version of this.
Adopted by
Most recently taken up by
- matt-riley/agent-skillstheir own version, first seen Nov 15, 2026
- mqmalagris/agent-skillstheir own version, first seen Nov 5, 2026
- drvoss/everything-copilot-clitheir own version, first seen Oct 26, 2026
- futex-ai/skillstheir own version, first seen Oct 25, 2026
- cellrishi-code/claude-skills-atlastheir own version, first seen Oct 4, 2026
- HugoLopes45/design-skillstheir own version, first seen Aug 26, 2026
Other versions8 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 8
- drvoss/everything-copilot-cli★ 49same name
- cellrishi-code/claude-skills-atlas★ 5same name
- ProviderProtocol/ai★ 4same name
- matt-riley/agent-skills★ 3same name
- aku11i/skills★ 1same name
- futex-ai/skills★ 1same name
- HugoLopes45/design-skills★ 1same name
- mqmalagris/agent-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
We split this skill into the rules it gives an agent
9 rules. 0 are common, 8 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 8 edited versions we read keep these.
8
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 17
Before inspecting code, ask: “Review in the current checkout or in a separate worktree?” Wait for the answer.
- Very specific: line 20
Separate worktree: ask “Which branch should I review, or should I use the current branch?” unless the target is already specified.
- Distinctive: line 24
Use the bundled helper through the shell tool (resolve its path relative to this skill directory):
- Distinctive: line 38
Read the system before judging the change.
- Distinctive: line 57
When reviewing a feature rather than a diff, ask the user which files or commits constitute the change.
- Distinctive: line 59
Completion criterion: every changed file listed, intent articulated in domain vocabulary, and each change mapped to the subsystem it …
- Distinctive: line 73
Completion criterion: every changed file examined against every applicable lens.
- Distinctive: line 85
Completion criterion: every finding from step 3 classified, reported, and traceable to a specific ground truth.
- Distinctive: line 95
Does not replace linters, type checkers, or CI.
“Common” is measured: kept by at least 6 in 10 of the 8 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Implementation Review
Probe implemented changes against the codebase's ground truth — the established patterns, domain model, conventions, and architectural decisions. The goal is not style nitpicks but structural gaps: what's missing, what breaks contracts, what violates the system's own rules.
Steps
0. Choose the review workspace
Before inspecting code, ask: “Review in the current checkout or in a separate worktree?” Wait for the answer. If the user already specified their choice, honour it without asking again.
- Current checkout: stay on the current branch and use the existing working directory. Do not create a worktree or switch branches.
- Separate worktree: ask “Which branch should I review, or should I use the current branch?” unless the target is already specified. “Current” means
HEAD. No PR access is assumed; ask for the branch name if given only a PR number or URL.
A worktree contains committed code only, not staged, unstaged, or untracked changes. Explain this before creating it; if the requested review includes uncommitted work, ask whether to review in place or review committed code only. Never stash, reset, commit, or copy the user's changes to create a review workspace.
Use the bundled helper through the shell tool (resolve its path relative to this skill directory):
bash /absolute/path/to/implementation-review/scripts/review-worktree.sh /absolute/path/to/repository [branch-or-ref]Omitting the target uses the current HEAD. The helper validates the ref, creates a unique temporary worktree outside the repository at a pinned commit in detached-HEAD mode, and prints its absolute path. This also works when the target branch is already checked out elsewhere. It does not switch branches, create a branch, or fetch. If the branch is unavailable or creation fails, report the error and ask how to proceed; do not silently fall back to the current checkout.
Use the returned path as the review root for all reads, searches, Git commands, checks, and delegated agents. Use explicit working directories or git -C; a shell cd does not persist across tool calls. Report the path and reviewed commit. The user can keep working in the original checkout; the worktree isolates files, not shared databases or external services.
Completion criterion: the user chose the workspace and the review root and target are known.
1. Build ground truth
Read the system before judging the change. Load, in order:
CONTEXT.mdanddocs/adr/— domain language and architectural decisions (proceed silently if absent)- The module or domain area the change touches — its existing patterns, types, tests, and interfaces
- Adjacent modules that interact with the changed code — their contracts and expectations
Explore the unchanged code until you can name the conventions it follows. Look for: directory structure patterns, naming conventions, dependency injection style, test structure, error handling idiom, and type definitions. These are the rules the change must obey.
Completion criterion: you can describe how the affected subsystem works, what patterns it follows, and what contracts it maintains with its neighbours — without referencing the change itself.
2. Scope the change
Identify what changed and why:
- For a branch review, confirm the comparison base (ask if unspecified), resolve the base and target to commit IDs, and run
git diff --stat <base>...<target>andgit diff <base>...<target>from the review root. Keep those IDs fixed throughout the review. Plaingit diffin a fresh worktree is empty; it is not the branch's changes. - For an uncommitted review in the current checkout, inspect
git diff(unstaged),git diff --cached(staged), andgit ls-files --others --exclude-standard(untracked); read the relevant untracked files too. The checkout is live: if it changes during review, disclose that limitation or restart the affected inspection. - Articulate the intent in domain language: what behaviour is being added, changed, or fixed?
- Map each changed file to the ground truth area it touches
When reviewing a feature rather than a diff, ask the user which files or commits constitute the change. Do not guess the boundary.
Completion criterion: every changed file listed, intent articulated in domain vocabulary, and each change mapped to the subsystem it affects.
3. Probe
Examine each change through every applicable lens below. Probe for what's absent, not just what's wrong — the most damaging gaps are things that should exist but don't.
Lenses
- Pattern conformance — does the change follow the patterns established in the same module? Check: naming conventions, file structure, dependency direction, error handling style, test patterns. A deviation is a finding even when the new way is arguably better — the codebase chose, and unilateral divergence is a cost.
- Domain fidelity — does the change use the domain's vocabulary correctly? Are new concepts named consistently with
CONTEXT.md? Does the behaviour match the domain model's rules? - Completeness — what's missing? Systematically check: tests (unit and integration), types and interfaces updated, error paths handled, edge cases covered, migrations needed, documentation updated, logging and observability added. Compare the test coverage pattern in adjacent modules — if they test X, this change should test X too.
- Contract integrity — does the change maintain contracts with adjacent code? Check: interface changes propagated to all callers, event schemas honoured, API contracts preserved, database constraints respected, shared types updated.
- Regression surface — could this break existing behaviour? Look for: changed function signatures, altered default values, removed or renamed exports, changed error types, modified query patterns, implicit assumptions in callers.
Completion criterion: every changed file examined against every applicable lens. Every finding documented with its location, the problem, and the ground truth it violates.
4. Report
Deliver findings grouped by severity:
- Gap — something missing that the ground truth demands (a test the pattern requires, an error path the convention handles, a type the interface needs)
- Violation — something present that contradicts the ground truth (a pattern broken, a contract breached, a domain term misused)
- Risk — something that works now but creates future problems (a regression surface, an implicit coupling, a missing migration)
Each finding states: where (file and location), what (the problem), and against what (the ground truth that makes it a problem).
Completion criterion: every finding from step 3 classified, reported, and traceable to a specific ground truth. No finding left unclassified. No lens left unapplied.
5. Finish the workspace
Include the review root and compared commits (or uncommitted scope) in the report. If this review created a worktree, ask whether to keep or remove it. Remove only that worktree with git worktree remove <review-path>; never use --force or delete an existing user worktree. If removal refuses because files changed, leave it and report the path. Never clean the user's original checkout.
Boundaries
This skill probes — it does not fix. Output is findings, not patches. If the user wants fixes applied, that is a separate pass.
Does not replace linters, type checkers, or CI. Assumes automated checks pass. Probes for what automated tools cannot catch: structural gaps, missing pieces, and pattern violations that require system understanding.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

