Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add SII-Holos/synergy --skill develop-benchmarkgit clone --depth 1 https://github.com/SII-Holos/synergyWrote 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/skills/sii-holos/synergy/develop-benchmark)<a href="https://agentmods.dev/skills/sii-holos/synergy/develop-benchmark"><img src="https://agentmods.dev/badge/skills/sii-holos/synergy/develop-benchmark/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/skills/sii-holos/synergy/develop-benchmark"><img src="https://agentmods.dev/badge/skills/sii-holos/synergy/develop-benchmark.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 | $0.00046 | $0.07341 |
| Opus 5.5 | $0.00018 | $0.02936 |
| Sonnet 5.5 | $0.00009 | $0.01468 |
| Haiku 4.5 | $0.00005 | $0.00734 |
Grade A, and why
develop-benchmark 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 3d 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 — 179 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Develop the Benchmark Evaluator
Freeze the experiment
- Read benchmark ownership, commands and formats, and the matrix decision.
- Separate evaluator code, measured harness source or package version, runtime composition, model profile, native task inputs and repeat identity. Declare configuration differences as named variants before observing scores. A model profile that enables thinking must also declare its reasoning tier. Providers expose depth through a separate control from the enable switch, so an omitted tier silently inherits the provider default and the retained evidence cannot state the condition that produced a score. Name the tier in the model key and set it explicitly; treat a different tier as a new named condition rather than a reinterpretation of a completed run. See the tier decision.
- Freeze a new experiment after changing the evaluator. Resume only the same run with its frozen current evaluator and current-format records. Reject old matrix/result formats and retired CLI commands; do not add translators or automatic old-evaluator dispatch. Preserve old evaluators and reports unchanged, and freeze sealed historical cost summary references in
prior_costsfor new family totals. Product-version adapters are separate and remain supported. - Put disposable live-task workspaces outside protected assistant configuration directories. Validate that the actual sandbox can run a fixture-local Node/Bun test before interpreting model recovery loops as harness regressions. A fixture inside a denied ancestor can permit some reads while breaking module resolution and current-directory discovery.
- Keep provider keys in environment references. Use deterministic provider tests when changing the container, native harness, inference proxy, observer or streaming recorder. Algorithm comparisons start formal tasks directly; never prepend paid probes, full-task prewarming, oracle audits or a model-based admission phase.
Freeze the provider's actual thinking switch, including strict boolean
enable_thinkingwhere required. Explicitly disabled thinking is not a reasoning capability. Inspect retained upstream requests from primary and auxiliary calls; native reasoning defaults must not survive a disabled profile. Honor endpoint-specific direct transport settings in the isolated evaluator process without changing live configuration.
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.
- 3d ago Changed · +2 lines 6729233b1edb
- 11d ago Changed 2a3bd69b8d08
- 14d ago Changed · +2 lines 2b786511920c
- 15d ago Changed · +4 lines a716e26fa632
- 16d ago Changed · +7 lines 8bf6f320d936
- 18d ago Changed · +10 lines a8b197c016c0
- 20d ago Changed · +2 lines d3168f3fea11
- 21d ago Changed · +69 lines e06d50c42734
- 24d ago Changed · +6 lines e49101b61736
- 26d ago First seen · 77 lines · 46 tokens per session scan A 87ab8f389f90
develop-benchmark is a skill published in the GitHub repository SII-Holos/synergy (386 stars, last pushed today), licensed MIT. It adds 46 tokens to every session and 7,341 once invoked, about $0.0002 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 2026-09-15.
Other skills, from other repositories
research-engineer
An uncompromising Academic Research Engineer. Operates with absolute scientific rigor, objective criticism, and zero flair. Focuses on theoretical correctness, formal verification, and optimal implementation across any required technology.
eval-dataset-design
A guide to creating reliable test sets for AI agents. Each evaluation task includes the request, the simulated user's behaviour, a reset starting state, and a success check that can be independently verified.
tika-eval-compare
Compare extracts from two Tika builds over a corpus to detect regressions in content, encoding, exceptions, and embedded-document handling. Use for "compare before/after extracts", "eval this change against the corpus".
cua-driver
Use Cua Driver for desktop or browser tasks that are awkward or unavailable through Bash/APIs, or when the user explicitly wants GUI interaction: app testing, visual bug reproduction, form filling, calendar entry, screenshots, and demo recording. Also covers Cua setup; not OpenAI Codex Computer Use or web research.
jetson-validate-image
Use after jetson-flash-image to run static BSP checks, on-target smoke/regression tests on a flashed DUT, or both. Not for build or flash steps. Triggers: validate bsp, on-target validation.
atmos-validation
Validate Atmos projects, components, arbitrary JSON Schema inputs, EditorConfig, and GitHub Actions; use affected-file selection and native CI annotations.