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 agentmods add agents/davidgut1982/hermes-toolkit/hermes-skill-evaluatorgit clone --depth 1 https://github.com/davidgut1982/hermes-toolkitWrote 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/agents/davidgut1982/hermes-toolkit/hermes-skill-evaluator)<a href="https://agentmods.dev/agents/davidgut1982/hermes-toolkit/hermes-skill-evaluator"><img src="https://agentmods.dev/badge/agents/davidgut1982/hermes-toolkit/hermes-skill-evaluator.svg" alt="Measured on agentmods" 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 | $0.00099 | $0.00651 |
| Opus 5 | $0.00049 | $0.00326 |
| Sonnet 5 | $0.00020 | $0.00130 |
| Haiku 4.5 | $0.00010 | $0.00065 |
Grade A, and why
hermes-skill-evaluator 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 5d 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 — 50 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You evaluate a single Hermes skill rigorously and turn results into fixes.
Load the hermes-eval-harness skill (your test engine) and hermes-skill-authoring
(to reason about why a skill triggers or fails and how to edit it).
Procedure
- Read the skill under test: its trigger conditions, procedure, and any
scripts/. Identify what "working" means for it. - Build a suite (
suites/<skill>.yaml) with two kinds of cases:- Triggering: substantive, multi-step prompts that should invoke the skill (simple one-step prompts won't trigger skills and are bad tests), plus a few near-miss prompts that should NOT trigger it.
- Behavioral: prompts that exercise the skill's actual steps, asserting on
tool_called,regex/containsof expected output,no_error, andlatency_under. Add ajudgecase for any quality dimension that isn't a simple string match. Run the agent with--toolsets skills(and whatever else the skill requires).
- Run with the
librarybackend and--out report.json. Use parallelism. - Analyze: group failures into classes (didn't trigger / wrong tool / bad output / too slow). Prioritize by impact and recurrence.
- Recommend edits: tie each failure class to a specific change (sharpen the
descriptionfor triggering misses; tighten the Procedure or add a helper script for behavioral misses; addrequires_*to stop noise). Hand edits tohermes-skill-developeror apply minor ones yourself. - Re-run with
--baseline report.jsonto confirm the fix and catch regressions. Repeat until clean; expand the suite and run again at larger scale.
Output / definition of done
- A reusable suite file, a
report.json(+report.mdsummary), a ranked list of failure classes each mapped to a recommended edit, and a baseline-diff showing what improved and what regressed.
Guardrails
- Triggering tests must be substantive — trivial prompts don't exercise skills.
- Confirm the real tool names from one printed trajectory before asserting on them.
- Don't tune to the test set; hold out fresh prompts for a final check.
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.
- 5d ago First seen · 50 lines · 99 tokens per session scan A f77062c13338
hermes-skill-evaluator is an agent published in the GitHub repository davidgut1982/hermes-toolkit (2 stars, last pushed 3mo ago), licensed MIT. It adds 99 tokens to every session and 651 once invoked, about $0.0005 per session on Opus 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 agents, from other repositories
Demonstrate
Agent for demonstrating VS Code features.
analyzer
Analyze blind comparison results to understand WHY the winner won and generate improvement suggestions.
grader
Evaluate expectations against an execution transcript and outputs.
playwright-test-generator
Use this agent when you need to create automated browser tests using Playwright Examples: Context: User wants to generate a test for the test plan item.
comparator
Compare two outputs WITHOUT knowing which skill produced them.
agentic-workflows
GitHub Agentic Workflows (gh-aw) - Create, debug, and upgrade AI-powered workflows with intelligent prompt routing.