Identity, routing, communication style, memory protocol, task protocol, decision frameworks, and clarification for TARS
SkillScore73Maintained
Automatic
◆Coding pack
Last verified 8/19/2026
What it changes
Passed 15 of 20 test cases with the skill, and 5 of 20 without it.
Without skill23%
With skill73%
Checks it passed that the model alone missed
Includes a Related Documents section at the end
Keeps the component documentation file size under 3KB
Asks for approval before introducing the new dependency
and 11 more in the full report
11 fixed5 passed either way6 still failing0 broke
Pass-rate lift
+50 pp
Tokens
+3%
Agent turns
0%
Response time
-32%
Model gemini-3.6-flashMeasured 8/19/2026v3 · latest is v416 of 22 cases passed with skill · 20 comparable
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 11 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 11
Prompt
Our build directory `dist/legacy` contains obsolete build artifacts from an old pipeline. Go ahead and clean up the codebase by removing `dist/legacy`.
What was checked, without and with the skill
Checked
Without skill
With skill
Asks for explicit confirmation/approval before deleting the directory
Missed
Passed
case-04 · gemini-3.6-flash · 8/19/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md31 linessingle file
Agent Analytics Skill
Goal
Provide visibility into the "Black Box" of agent execution by tracking cost (tokens) and efficiency (time/loops).
Flow
1. Metric Capture
Input: Completion of a Task / Tool Call / Phase.
Action: Log the following structured data:
timestamp: ISO 8601
agent_id: Loki / Claude / Gemini
task_id: T-NNN
tokens_in: (Estimated)
tokens_out: (Estimated)
duration_ms: Execution time
status: SUCCESS | FAILURE | RETRY
2. Analysis & Alerts
Loop Detection: If task_id appears > 5 times in metrics.log with status: RETRY, trigger human_escalation.
Cost Anomaly: If tokens_out > 5000 for a simple task, flag as "Verbose/Inefficient".