CtrlK
BlogDocsLog inGet started
Tessl Logo

matthew-a-carr/review-implementation

Repo-aware review of an implementation PR (the `ai:done` PR) against the SPEC it implements, the constitution, the ADRs, and the doc-staleness rules. Use when a routine fires on a PR labelled `ai:done`, when a human says "review impl PR #NNN" / "review the implementation for SPEC-NNN", or as a self-review step inside `implement-spec` before the PR is opened. Read-only — produces a structured report and never edits code or merges.

85

1.06x
Quality

90%

Does it follow best practices?

Impact

69%

1.06x

Average score across 3 eval scenarios

SecuritybySnyk

Advisory

Suggest reviewing before use

Overview
Quality
Evals
Security
Files

Evaluation results

67%

6%

Implementation PR Review: Notification Service

Architecture boundaries and ADR obligations

Criteria
Without context
With context

Report file exists

100%

100%

Verdict present

100%

100%

Verdict is Needs changes

100%

100%

Repo interface in wrong layer

66%

66%

Application imports infrastructure

50%

60%

Direct construction not DI

80%

70%

New library needs ADR

13%

60%

Deviation not logged

60%

60%

Report sections present

40%

20%

Finding path prefixes

60%

60%

No ADR 053 false positive

0%

0%

Read-only stance

100%

100%

61%

-6%

Implementation PR Review: Search Feature

Test quality and doc staleness review

Criteria
Without context
With context

Report file exists

100%

100%

Verdict present

100%

80%

Verdict is Needs changes

100%

40%

it.only Critical

100%

100%

Mocked DB for repository Critical

53%

53%

Missing integration test file

50%

50%

OpenAPI not updated Critical

53%

33%

Report sections present

20%

0%

Correct DI usage noted

0%

0%

CHANGELOG noted correct

0%

40%

Finding path prefixes

100%

100%

Read-only stance

100%

100%

79%

11%

Implementation PR Review: Payment Processing Feature

SPEC fidelity and conventions review

Criteria
Without context
With context

Report file exists

100%

100%

Verdict field present

40%

80%

Verdict is Needs changes

60%

80%

Report sections present

0%

0%

Money-as-float Critical

100%

90%

Throw-instead-of-Result Critical

50%

90%

Mocked-database Critical

30%

60%

Skipped-test Critical

50%

90%

Domain-imports-infrastructure Critical

100%

80%

Endpoint status code

100%

100%

TODO/FIXME warning

60%

60%

CHANGELOG warning

100%

80%

Finding path prefixes

40%

60%

Read-only — no code edits

100%

100%

Evaluated
Agent
Claude Code
Model
Claude Sonnet 4.6

Table of Contents