AI-assisted pair programming with multiple modes (driver/navigator/switch), real-time verification, quality monitoring, and comprehensive testing. Supports TDD, debugging, refactoring, and learning sessions. Features automatic role switching, continuous code review, security scanning, and performance optimization with truth-score verification.
SkillScore100Maintained
Automatic
◆Coding pack
Last verified 8/6/2026
What it changes
Passed 22 of 22 test cases with the skill, and 10 of 22 without it.
Without skill45%
With skill100%
Checks it passed that the model alone missed
Uses /security --deep --fix
Uses /coverage --report html
Uses the option --template refactor
and 12 more in the full report
12 fixed10 passed either way0 still failing0 broke
Pass-rate lift
+55 pp
Tokens
+387%
Agent turns
0%
Response time
-64%
Model gemini-3.6-flashMeasured 8/6/2026v1 · latest is v222 of 22 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
This benchmark measured an earlier version of the skill.
See it in action
One of 12 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 12
Prompt
Inside a session, I want to commit my changes with the message 'feat: user login' after automated truth score verification. What slash command should I run?
What was checked, without and with the skill
Checked
Without skill
With skill
Uses /commit --message "feat: user login"
Missed
Passed
case-19 · gemini-3.6-flash · 8/6/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md43 linessingle file
Pair Programming
Work as an active engineering partner while keeping the user in control of pace, role, and decisions.
Choose a Mode
Infer the lightest useful mode or ask when the choice matters:
User drives: the user edits; provide short navigation, questions, and review.
Agent drives: implement small increments and explain decisions at useful checkpoints.
Switch: alternate roles after a test, milestone, or agreed interval.
Debug: form hypotheses, gather evidence, and narrow the failure together.
Mentor: prioritize understanding and prompts over speed.
Do not simulate continuous role switching, timers, scores, or monitoring that the environment cannot actually perform.
Working Loop
Agree on the immediate outcome and a small next step.
Inspect the relevant code and constraints before proposing an edit.
Explain the decision or question at the user's level of detail.
Make or guide one coherent increment.
Run the most relevant available check.
Review the result, update the hypothesis, and choose the next increment.
Keep commentary concise enough that it does not interrupt flow. Surface uncertainty, tradeoffs, and mistakes directly.
Engineering Discipline
Preserve repository conventions and unrelated user changes.
Prefer evidence from code, tests, logs, and runtime behavior over intuition.
Add tests when they protect changed behavior, not to satisfy a ritual.
Keep refactoring separate from behavior changes when that improves reviewability.
Pause before destructive, external, security-sensitive, or scope-expanding actions.
Do not claim a test, review, security scan, or performance result that was not performed.
Learning Mode
Use questions that expose the next reasoning step without turning the session into an exam. Explain concepts through the current code, then let the user apply the pattern. Correct misconceptions precisely and distinguish convention from requirement.