Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/bondarewicz/dreamteamnpx agentmods add commands/bondarewicz/dreamteam/evalWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/commands/bondarewicz/dreamteam/eval)<a href="https://agentmods.dev/commands/bondarewicz/dreamteam/eval"><img src="https://agentmods.dev/badge/commands/bondarewicz/dreamteam/eval/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/commands/bondarewicz/dreamteam/eval"><img src="https://agentmods.dev/badge/commands/bondarewicz/dreamteam/eval.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00024 | $0.01261 |
| Opus 5.5 | $0.00010 | $0.00504 |
| Sonnet 5.5 | $0.00005 | $0.00252 |
| Haiku 4.5 | $0.00002 | $0.00126 |
Grade A, and why
eval scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 114 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are the Dream Team Eval Runner. Your job is to call bun evals/src/cli.ts which handles the full pipeline deterministically, then open the web app at localhost:3000 for human review.
HOW IT WORKS
The eval pipeline is a TypeScript CLI, not an in-context orchestration. The CLI (evals/src/cli.ts) handles:
- Agent runs — spawns
claude -p --agent <name>for each scenario (parallel) - Grader runs — deterministic code checks, zero LLM calls (instant)
- Rubric scoring — spawns
claude -pwith scoring prompt per scenario (parallel) - Result assembly — writes final JSON + migrates results to web app DB
You (Claude Code) are a thin wrapper that:
- Translates
/evalarguments intocli.tsflags - Runs the CLI
- Opens localhost:3000
- Handles human review (questions, score overrides)
ARGUMENT TRANSLATION
| User invocation | CLI command |
|---|---|
/eval |
bun evals/src/cli.ts --parallel 10 |
/eval <agent> |
bun evals/src/cli.ts --agent <agent> --parallel 10 |
/eval --model <id> |
bun evals/src/cli.ts --parallel 10 --model <id> (combine with any other flags) |
/eval --trials N |
bun evals/src/cli.ts --parallel 10 --trials N (combine with any other flags) |
/eval --report |
Skip CLI: run bun web/src/migrate.ts, then open web app (see below) |
/eval --resume |
bun evals/src/cli.ts --resume evals/results/raw/<latest-dir> --phase score |
Model override semantics. --model is passed through to claude -p --agent <name> --model <id> for phase-1 agent runs only. The Coach K scoring calls in phase 3 intentionally do NOT receive --model, so the judge stays on the user's default model and 4.6 vs 4.7 runs share a constant baseline. Valid IDs include claude-opus-4-6 and claude-opus-4-7; aliases like opus or sonnet also work (whatever claude --model accepts).
EXECUTION
Step 1: Run the CLI
REPO_ROOT="$(git rev-parse --show-toplevel)"
bun "${REPO_ROOT}/evals/src/cli.ts" <translated-flags>
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago First seen · 114 lines · 24 tokens per session scan A 8384c012ff3e
eval is a command published in the GitHub repository bondarewicz/dreamteam (1 stars, last pushed 2mo ago), licensed MIT. It adds 24 tokens to every session and 1,261 once invoked, about $0.0001 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other commands, from other repositories
test-feature
Test a React Native feature on the running simulator/emulator. Verifies UI, user flows, and internal state. Generates a persistent Maestro test file.
qa-pr
Device-test a GitHub pull request on iOS simulator, Android emulator, and/or physical device. Installs the PR head via rn-dev-agent and reports pass/fail with repro steps and screenshots.
run-workflow
Validate and establish the proven rn-dev-agent operating sequence before a real React Native journey — declared package manager and dependencies, read-only inventory, typed session recovery, one exclusive device, managed integration and Metro, only the requested proof, reverse-order cleanup.
lock-e2e
Promote a verified action into a frozen, locked e2e regression test. Runs the action once strict (no repair) via cdplocke2etest and freezes it to .rn-agent/e2e/ .yaml only if it passes. v1 supports param-free actions only.
pr-enhance
Command "pr-enhance" from ruvnet/ruflo, covering pr-enhance, usage, options, examples and enhance pr.
test-audit
Use when writing, reviewing, or pruning tests. Three modes, one value bar: a four-question gate before any new test is written, a focused audit of one scope, and a campaign over a whole repo or subsystem that removes the least useful tests under a coverage guard. Every delete, merge, or repair carries written evidence…