- Home
- Skills
- Code Quality & Review
- Code Review Specialist
Works with the AI Agents you already use
Code Review Specialist
Get a review that earns its verdict. Trace the change against its requirements, prioritize concrete defects and define what must change before acceptance.
$5
Code Review Specialist
Example session with this skill installed
Review the complete authorization predicate below. The application policy is: an authenticated reader can view their own private notes; an authenticated administrator can view any note. All actors in this example are already authenticated. Criterion AUTH-1: reader id=7 must be allowed to view private note owner_id=7. Criterion AUTH-2: reader id=8 must be denied that private note. Criterion AUTH-3: an administrator must be allowed.
Candidate excerpt C1, notes.py:
def can_view(actor, note):
return actor['role'] == 'admin' or note['public']
Note N1 is {'owner_id': 7, 'public': False}. No repository snapshot, diff, callers or execution logs are supplied. Produce a bounded static ReviewReport, identify the exact defect and close condition, and distinguish source-supported findings from unperformed checks. Do not edit source or authorize a merge. Respond in English.
- Read your context and instructions
- Compiled the code review specialist
Illustrative preview: this representative editorial result is not a recorded runtime session.
task_id: NOTE-REVIEW-1
artifact_id: REVIEW-C1
version: 1
status: CHANGES_REQUESTED
producer / recipient: reviewer / predicate owner
objective / scope: Check AUTH-1 through AUTH-3 against the supplied can_view predicate only.
inputs / sources: Candidate excerpt C1 and note N1 in the request.
confirmed_decisions: Actors are authenticated before this predicate; private-note rules are explicit.
recommendations / pending: Correct the ownership branch; obtain current source binding and affected checks.
criteria: AUTH-1, AUTH-2, AUTH-3
next_action: Predicate owner corrects AUTH-1 and provides the new candidate plus allowed/denied evidence.
ReviewReport
review_outcome: CHANGES_REQUESTED
candidate_ref: Supplied excerpt C1; no immutable repository binding provided.
base_ref / candidate_manifest: Not supplied.
spec_ref: AUTH-1 through AUTH-3 in the request.
reviewer / independence: Static review in this context; no distinct-context independence is established. Authorship, specification and evidence-collector overlap were not supplied.
scope_reviewed / exclusions: Complete supplied predicate; callers, endpoint enforcement and runtime behavior excluded.
Finding F1 — P2; established by source
- Candidate / criterion: C1 / AUTH-1.
- Location: notes.py, can_view return expression.
- Trigger: reader id=7 requests note N1, owned by id=7 and marked public=False.
- Expected: allow the owner. Actual by source reasoning: both comparisons are false, so the expression returns False. This is not a recorded execution.
- Consequence: the owner's legitimate private-note request is denied by this predicate. No data leak or production incident is established.
- Required action / owner: Predicate owner implements the agreed ownership behavior while retaining the administrator rule.
- Close condition: owner-allowed, other-reader-denied and administrator-allowed observations reference the corrected candidate.
Criterion disposition
- AUTH-1: NOT_SATISFIED by the supplied expression.
- AUTH-2: UNPROVEN as executed behavior; the expression is statically consistent with denial for this fixture.
- AUTH-3: UNPROVEN as executed behavior; the administrator branch is statically consistent with allowance.
Evidence gaps: No source snapshot/diff, current runtime checks or caller coverage. Obtain a bound candidate and the three checks; inspect the real call path if endpoint acceptance is requested.
Reviewer probes: None; no code or tests executed.
Candidate binding after review: Supplied excerpt only, unchanged.
Outcome limit: A concrete predicate defect was identified; this report grants no merge or release authorization.
Connects securely to your tools. The creator never sees your data.
About this skill
A useful code review gives you a reason to change the code, not another list of preferences.
Code Review Specialist equips your AI to inspect a software change against the behavior it was meant to deliver. It follows the diff into relevant callers and contracts, checks whether the supplied evidence supports acceptance and turns concrete problems into actionable findings. Get a review your engineer can use to make the next correction.
Bring a pull request, candidate excerpt or implementation change, its requirements and any check results. Use it before accepting a feature, after a bug fix or when a green test suite still leaves an important behavior uncertain. The review stays within the source and environment actually available.
Your report includes
- Prioritized findings with location, trigger and consequence.
- The expected behavior and a practical correction direction.
- A requirement-by-requirement account of what is supported or missing.
- Clear conditions for closing findings and the next review step.
For an authorization predicate, it can trace a legitimate owner's denied request to the exact missing branch, distinguish that defect from uninspected callers and define the cases that should demonstrate the correction. It also considers relevant maintainability, recovery and boundary risks without turning personal style into a blocker.
This standalone skill includes the full method, report template and buyer guidance. Supply your own AI access and project materials. Repository inspection or defect reproduction needs the corresponding host tools; the specialist reviews without silently rewriting the candidate.
A review covers only the supplied or inspected change and its evidence. Static findings do not establish runtime acceptance. The method does not silently edit the candidate or grant merge and release approval.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 1 month ago
Permissions
Reading the installed skill and supplied project materials is required for file-based use. Terminal or write access, where declared, is conditional on the requested execution or implementation mode and the buyer's authorization. Planning can use supplied text without those tools. Optional research uses only browsing/network access authorized by the buyer. No credential, environment-variable or account access is mandatory.