Review agent-native parity for skills, prompts, tools, commands, generated mirrors, repo workflows, and user-facing actions.
SkillScore82Maintained
Automatic
◆AI Engineer pack
Last verified 8/15/2026
What it changes
Passed 16 of 19 test cases with the skill, and 9 of 19 without it.
Without skill45%
With skill82%
Checks it passed that the model alone missed
Assigns pass for the CLI unit test action.
Assigns N/A for the human-only billing payment action.
Assigns severity level P2 for route discoverability issues.
and 6 more in the full report
8 fixed10 passed either way4 still failing0 broke
Pass-rate lift
+36 pp
Tokens
+82%
Agent turns
0%
Response time
-48%
Model gemini-3.6-flashMeasured 8/15/2026v118 of 22 cases passed with skill · 19 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 8 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 8
Prompt
A review notes minor phrasing ambiguity in a skill's description section that does not impede execution but could improve agent clarity. Assign the finding severity rating.
What was checked, without and with the skill
Checked
Without skill
With skill
Assigns severity level P3 for non-blocking documentation polish.
Missed
Passed
case-19 · gemini-3.6-flash · 8/15/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md232 linessingle file
Agent-Native Reviewer
Review whether an agent can perform, verify, and discover the same meaningful
action a user can.
A feature adds or changes a user-facing action and the repo has agent
integration.
A reviewer asks whether Codex can reproduce, operate, verify, or maintain the
same surface a human can.
A repo is becoming mostly agent-maintained and needs fewer hidden human-only
paths.
Do Not Use When
The task is a normal code review with no agent/tooling/workflow surface.
The action is intentionally human-only: MFA, OAuth consent, billing payment
entry, CAPTCHA, legal acceptance, biometric unlock, or OS permission prompts.
The change is cosmetic and has no meaningful operation to reproduce.
Core Principles
Action parity. Every important user action has an agent route.
Context parity. Agents can see the inputs, state, and constraints needed
to act well.
Source ownership. Agents edit the durable source, not generated mirrors
or copied output.
Proof parity. Agents can verify the outcome with the same authority a
maintainer would trust.
Turn the current Codex thread into a coordination thread that routes implementation work to durable reusable child threads in disposable worktrees with short-lived branches targeting main.