Borrowing it
Nothing to install: this file belongs to dbcls/togomcp. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/dbcls/togomcp/main/.claude/skills/benchmark-runner/SKILL.mdgit clone --depth 1 https://github.com/dbcls/togomcpWrote 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/dbcls/togomcp/benchmark-runner)<a href="https://agentmods.dev/skills/dbcls/togomcp/benchmark-runner"><img src="https://agentmods.dev/badge/skills/dbcls/togomcp/benchmark-runner/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/dbcls/togomcp/benchmark-runner"><img src="https://agentmods.dev/badge/skills/dbcls/togomcp/benchmark-runner.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 | $0.00000 | $0.02833 |
| Opus 5.5 | $0.00000 | $0.01133 |
| Sonnet 5.5 | $0.00000 | $0.00567 |
| Haiku 4.5 | $0.00000 | $0.00283 |
Grade A, and why
benchmark-runner scanned grade A with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 21d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
Pass `--usage-log` a production tool-call JSONL (download: `curl -u "$TOGOMCP_STATS_USER:$TOGOMCP_STATS_PASSWORD" How it starts
The opening of the file, as written. The whole thing — 155 lines — stays where its author put it; the contents beside it link to each section on GitHub.
TogoMCP Benchmark Runner
Contract item 2.4 (periodic evaluation). Two scripts do the work; this file is the procedure and the rules around them.
| Script | Does |
|---|---|
scripts/run_benchmark.py |
answer → guard → judge → manifest → summarize. Spends API money unless --dry-run. |
scripts/summarize_run.py |
rebuilds summary.csv + report.md for any run dir, free. Re-run it after adding a judge or to attach a usage log. |
Procedure
- Check the question set is frozen.
python benchmark/scripts/make_set_manifest.py --check. If it fails, questions changed since the freeze: stop and ask the user whether to re-freeze (make_set_manifest.py, then commit). Every run records the set hash; runs on different hashes are not comparable question for question. - Dry run and estimate.
uv run python .claude/skills/benchmark-runner/scripts/run_benchmark.py --mode canary --dry-run(or--mode full). It checks the freeze, the judges (Ollama model pulled), the config, boots the local server, and prints the plan and an estimate. - Get approval before any paid run. Show the user: mode, question count, replicates, cells, answering and judge models, target (local corpus or production), estimated cost and wall time. Wait for an explicit yes. Planning figures (2026-10-01, Sonnet 5.5 + Opus 4.8): about $0.31 per cell and 1.1 min per cell; canary ≈ $3 and 15 min, full (110 × 3) ≈ $100 and 6 h. Gemma4 judging is free but slow: about 46 s per judgement on the developer Mac (two per cell, TogoMCP and no-tool), so ≈ 15 min for a canary and ≈ 8.5 h for a full run, after answering.
- Run detached. A full run outlives a tool timeout, so launch with
nohupandANTHROPIC_API_KEYset (e.g.ANTHROPIC_API_KEY=$MY_ANTHROPIC_API_KEY nohup uv run python .../run_benchmark.py --mode full > <log> 2>&1 &). Answering and the Claude judge go through the API; theclaude loginsubscription degrades into "Not logged in" stubs on long runs. - Read the report (
report.md) and relay to the user: scores per judge with the paired Δ and CI, the held-out vs development split, excluded-cell counts, operational metrics, judge agreement, the diff against the previous run, and the usage-log section if a log was given.
What ships with it
2 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 21d ago Changed · +7 lines a0ca4ddb1c82
- 22d ago Changed · +12 lines 40c4e7ec2e55
- 28d ago First seen · 136 lines · 0 tokens per session scan A b7a2a26aea66
benchmark-runner is a skill published in the GitHub repository dbcls/togomcp (20 stars, last pushed 2d ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 2,833 tokens. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-10-02.
Other skills, from other repositories
open-ontologies
AI-native ontology engineering using 110 MCP tools backed by an in-memory Oxigraph triple store. Build, validate, query, and govern RDF/OWL ontologies with a generate-validate-iterate loop. Use when building ontologies, knowledge graphs, RDF data, SPARQL queries, BORO/4D modeling, SHACL validation, clinical…
post-cortex-memory
Long-term project memory through the post-cortex MCP server (tools named session, semanticsearch, updateconversationcontext, assemblecontext, queryconversationcontext, getstructuredsummary, manageworkspace, manageentity, admin). Use it whenever those tools are available and the work could benefit from what was learned…
api-simulation
Runs your code for real against stand-ins for the HTTP services it calls, served together as one environment on one local port or one deployed URL, whether that code is a service you're changing, a frontend, a test suite or a demo. Use when you add or change code that calls other services and need to run it, test it…
application-testing
Sets up, runs, debugs, and interprets results from postman application test — the Postman layer that runs your Playwright tests, captures their network traffic, matches it against a collection contract, runs pm.test assertions, and publishes results to Application Inventory. Covers the Postman side of the integration…
variables
Use when parameterizing requests across environments, choosing variable scopes, chaining values in scripts, or referencing credentials without committing or exposing them. Keeps one reusable collection while separating target configuration from secret values. Use for requests such as "where should I put this API key".
api-mocking
Finds, builds, runs and publishes a stand-in for one API — the owning team's published mock when there is one, otherwise one built from its spec or collection that behaves like the real API, stateful when callers write then read. Use when something needs an API that isn't built or reachable yet (a frontend, client or…