CtrlK
BlogDocsLog inGet started
Tessl Logo

uinaf/review-gang

Review existing code, diffs, branches, or pull requests with a risk-scaled set of concern-specific reviewer subagents, then synthesize a ship-it / needs-review / blocked verdict.

93

1.18x
Quality

95%

Does it follow best practices?

Impact

91%

1.18x

Average score across 4 eval scenarios

SecuritybySnyk

Passed

No known issues

Overview
Quality
Evals
Security
Files

Evaluation results

87%

3%

Auth Token Module Review

Criteria
Baseline
With context

any type flagged

100%

100%

Unsafe cast flagged

30%

100%

Non-null assertion flagged

100%

100%

Dead code identified

100%

100%

Duplicate logic identified

100%

100%

Catch-all error flagged

100%

100%

Error classification recommended

60%

100%

Narrating comments flagged

100%

100%

Findings tied to impact

100%

100%

Valid verdict

80%

0%

Compact verdict block

50%

50%

91%

20%

Code Review: Authentication Middleware Refactor

Criteria
Baseline
With context

Default gang spawned

0%

100%

Subagent authorization honored

50%

87%

Types persona included

80%

70%

Cleanup persona included

75%

87%

Comments persona omitted

100%

100%

Repo guidance loaded

100%

100%

Personas listed in output

0%

100%

Verdict present

75%

100%

Scope stated

83%

83%

Findings ordered by severity

100%

100%

Evidence for findings

100%

100%

Silent failures finding

87%

100%

Unverified areas acknowledged

50%

100%

Recommended follow-up

100%

100%

Compact verdict block

33%

33%

86%

5%

Quick Review: Documentation Typo Fix

Criteria
Baseline
With context

Small tier selected

100%

100%

General persona used

70%

70%

Comments persona used

70%

80%

Irrelevant personas omitted

100%

100%

Verdict present

100%

100%

Separate reviewers evidenced

12%

0%

No nit inflation

100%

100%

Docstring accuracy noted

100%

100%

Unverified areas or residual risk

83%

100%

Recommended follow-up

50%

100%

Compact verdict block

62%

87%

100%

28%

Review: Payment Retry Logic Changes

Criteria
Baseline
With context

CLAUDE.md loaded

100%

100%

File references in findings

100%

100%

Line-level evidence

100%

100%

Verdict present

62%

100%

Findings by severity

100%

100%

Error classification finding

100%

100%

Silent failure finding

100%

100%

Mock-heavy test concern

33%

100%

Dead code identified

0%

100%

Unverified surfaces marked

37%

100%

Recommended follow-up

100%

100%

No nit inflation

100%

100%

Reviewer gang listed

0%

100%

Compact verdict block

50%

100%

Evaluated
Agent
Claude Code
Model
Claude Sonnet 4.6

Table of Contents