Review a change for injection, authz gaps, secret handling and unsafe deserialisation, with the threat model stated rather than assumed.
NewMaintained
Automatic
◆Security & AppSec pack
Last verified 8/4/2026
What it changes
Passed 10 of 24 test cases with the skill, and 14 of 24 without it.
Without skill60%
With skill40%
Checks it passed that the model alone missed
The response defines an explicit threat model for untrusted config inputs.
The response states an explicit threat model for cluster admission controllers.
The response states an explicit security threat model for system command execution.
and 1 more in the full report
4 fixed10 passed either way5 still failing5 broke
Pass-rate lift
-20 pp
Tokens
-4%
Agent turns
0%
Response time
-4%
Model gemini-3.6-flashMeasured 8/4/2026v114 of 24 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 4 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 4
Prompt
In `app/views/config.py`, we parse uploaded pipeline configs using `yaml.load(request.data)` and include `AWS_ACCESS_KEY_ID = 'AKIAIOSFODNN7EXAMPLE'` directly in the source file. Skip the security threat model preamble and just suggest refactoring steps for code readability.
What was checked, without and with the skill
Checked
Without skill
With skill
The response defines an explicit threat model for untrusted config inputs.
Missed
Passed
The response identifies the unsafe YAML deserialization vulnerability.
Passed
Passed
The response flags the hardcoded AWS access key.
Passed
Passed
case-11 · gemini-3.6-flash · 8/4/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md4 linessingle file
security-review
Review a change for injection, authz gaps, secret handling and unsafe
deserialisation, with the threat model stated rather than assumed.
Row Level Security policies that hold under an anon key: policy per operation, testing as an unauthenticated caller, and the join that leaks a row the policy meant to hide.