Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Evaluate and rank agent results by metric or LLM judge for an AgentHub session. Use when the user runs /hub:eval or asks to score, compare, or pick a winner among completed AgentHub agents.
.claude/skills/alirezarezvani-eval/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -28% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 204% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -37% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 15% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 6% | 0% |
Rank all agent results for a session. Supports metric-based evaluation (run a command), LLM judge (compare diffs), or hybrid.
/hub:eval # Eval latest session using configured criteria
/hub:eval 20260317-143022 # Eval specific session
/hub:eval --judge # Force LLM judge mode (ignore metric config)Run the evaluation command in each agent's worktree:
bashpython {skill_path}/scripts/result_ranker.py \ --session {session-id} \ --eval-cmd "{eval_cmd}" \ --metric {metric} --direction {direction}
Output:
RANK AGENT METRIC DELTA FILES
1 agent-2 142ms -38ms 2
2 agent-1 165ms -15ms 3
3 agent-3 190ms +10ms 1
Winner: agent-2 (142ms)For each agent:
git diff {base_branch}...{agent_branch}.agenthub/board/results/agent-{i}-result.mdPresent rankings with justification.
Example LLM judge output for a content task:
RANK AGENT VERDICT WORD COUNT
1 agent-1 Strong narrative, clear CTA 1480
2 agent-3 Good data points, weak intro 1520
3 agent-2 Generic tone, no differentiation 1350
Winner: agent-1 (strongest narrative arc and call-to-action)bashpython {skill_path}/scripts/session_manager.py --update {session-id} --state evaluating
/hub:merge to merge the winner/hub:merge {session-id} --agent {winner} to be explicit| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-06 | fail→pass | 7,495 | 1,765 | -76% | 1 | 1 | 0% | 1,284 | 922 | -28% | 0 | 0 | — |
case-01 | fail→fail | 7,010 | 6,005 | -14% | 1 | 1 | 0% | 1,102 | 961 | -13% | 0 | 0 | — |
case-02 | fail→pass | 5,463 | 16,598 | +204% | 1 | 1 | 0% | 1,050 | 3,187 | +204% | 0 | 0 | — |
case-03 | fail→fail | 11,556 | 5,258 | -54% | 1 | 1 | 0% | 2,109 | 891 | -58% | 0 | 0 | — |
case-04 | fail→pass | 9,190 | 2,745 | -70% | 1 | 1 | 0% | 1,753 | 1,104 | -37% | 0 | 0 | — |
case-05 | fail→fail | 4,016 | 3,447 | -14% | 1 | 1 | 0% | 650 | 1,285 | +98% | 0 | 0 | — |
case-07 | fail→pass | 5,736 | 3,079 | -46% | 1 | 1 | 0% | 1,066 | 1,229 | +15% | 0 | 0 | — |
case-08 | fail→pass | 6,598 | 3,342 | -49% | 1 | 1 | 0% | 1,170 | 1,242 | +6% | 0 | 0 | — |
case-09 | fail→pass | 18,624 | 3,681 | -80% | 1 | 1 | 0% | 1,199 | 1,052 | -12% | 0 | 0 | — |
case-10 | pass→pass | 10,204 | 2,829 | -72% | 1 | 1 | 0% | 1,815 | 1,009 | -44% | 0 | 0 | — |
case-20 | fail→pass | 7,942 | 1,537 | -81% | 1 | 1 | 0% | 1,383 | 852 | -38% | 0 | 0 | — |
case-11 | fail→fail | 8,405 | 1,468 | -83% | 1 | 1 | 0% | 1,402 | 809 | -42% | 0 | 0 | — |
case-12 | pass→pass | 5,629 | 1,337 | -76% | 1 | 1 | 0% | 955 | 821 | -14% | 0 | 0 | — |
case-13 | fail→pass | 9,385 | 1,574 | -83% | 1 | 1 | 0% | 1,464 | 818 | -44% | 0 | 0 | — |
case-14 | pass→pass | 5,637 | 1,465 | -74% | 1 | 1 | 0% | 924 | 855 | -7% | 0 | 0 | — |
case-15 | fail→pass | 7,002 | 1,832 | -74% | 1 | 1 | 0% | 1,252 | 859 | -31% | 0 | 0 | — |
case-16 | pass→pass | 9,334 | 2,660 | -72% | 1 | 1 | 0% | 1,579 | 1,056 | -33% | 0 | 0 | — |
case-17 | fail→pass | 11,175 | 1,250 | -89% | 1 | 1 | 0% | 1,879 | 820 | -56% | 0 | 0 | — |
case-18 | fail→pass | 7,350 | 1,650 | -78% | 1 | 1 | 0% | 1,274 | 871 | -32% | 0 | 0 | — |
case-19 | fail→pass | 6,962 | 2,041 | -71% | 1 | 1 | 0% | 1,275 | 1,012 | -21% | 0 | 0 | — |
case-21 | fail→pass | 5,699 | 3,149 | -45% | 1 | 1 | 0% | 1,019 | 990 | -3% | 0 | 0 | — |
case-22 | fail→pass | 7,305 | 4,736 | -35% | 1 | 1 | 0% | 1,314 | 1,451 | +10% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 19 counted toward the lift figure. The other 3 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +64 percentage points is the difference between those two pass rates over the 19 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.