Borrowing it
Nothing to install: this file belongs to lebek/modmixer. 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/lebek/modmixer/main/.claude/skills/harness-verify/SKILL.mdgit clone --depth 1 https://github.com/lebek/modmixerWrote 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/lebek/modmixer/harness-verify)<a href="https://agentmods.dev/skills/lebek/modmixer/harness-verify"><img src="https://agentmods.dev/badge/skills/lebek/modmixer/harness-verify/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/lebek/modmixer/harness-verify"><img src="https://agentmods.dev/badge/skills/lebek/modmixer/harness-verify.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, up to high
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- high YARA Match · line 3 YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).Fix: Remove offensive tool references and exploit code. Legitimate agent skills should not contain penetration testing tools, exploit frameworks, or reconnaissance utilities.
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 | $0.00104 | $0.01803 |
| Opus 5.5 | $0.00042 | $0.00721 |
| Sonnet 5.5 | $0.00021 | $0.00361 |
| Haiku 4.5 | $0.00010 | $0.00180 |
Grade A, and why
harness-verify 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 — 108 lines — stays where its author put it; the contents beside it link to each section on GitHub.
harness-verify — headless A/B turn replay
Replays a recorded ModMixer chat up to a chosen point, then runs the agent's
next turn headlessly against a chosen model — once per variant, repeated
N times — and reports what the agent did (asked the user vs. fired a
side-effecting tool like run_test_cycle). This is how you get runtime
evidence that a prompt/tool/hint change actually moves model behavior,
especially for non-frontier models (Kimi etc.) where "it's in the system
prompt" is not the same as "the model obeys it".
It runs the real system prompt, the real tool set, and the real model + credentials — but in an isolated session with side-effecting tools swapped for recording stubs, so nothing edits files, builds, launches RimWorld, or touches the user's conversation store.
When to use
- You changed how the agent should behave (a system-prompt rule, a tool description, a tool-result hint, a model) and want to see the effect.
- You want an A/B: same history + model, with vs. without your change.
- The surface is the agent's decision, and driving the GUI by hand is slow, costs RimWorld launches, and isn't repeatable.
Not for: pure mechanical changes with no behavioral question (a typecheck is enough), or anything where you'd need many full agentic turns with real file mutation (this stubs the side-effecting tools).
Run it
# Always dry-run first — proves the pipeline (Electron boot, model
# resolution, session reconstruction, prompt build) with ZERO model cost.
node C:\Users\peter\projects\modmixer\scripts\harness\run.mjs `
--title thunderstorm --until "make it sunny" --dry
# Live A/B: baseline (no change) vs fix (change), 3 turns each.
node C:\Users\peter\projects\modmixer\scripts\harness\run.mjs `
--title thunderstorm --until "make it sunny" `
--model moonshotai/kimi-k2.6 --provider openrouter --repeat 3
Use an absolute path to run.mjs (the runner finds the repo from its own
location; the shell cwd is irrelevant and has drifted before).
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 · 108 lines · 104 tokens per session scan A 99f2d716018a
harness-verify is a skill published in the GitHub repository lebek/modmixer (30 stars, last pushed 2mo ago), licensed MIT. It adds 104 tokens to every session and 1,803 once invoked, about $0.0004 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 2026-08-30.
Other skills, from other repositories
writing-tests
Kiro Crew repo only: how to write a backend pytest test with NO side effects that does not flake. Use when adding, editing, reviewing or debugging a test there: which conftest applies, what leaks (temp dirs, data home, /.kiro, cron, threads), the seven flake classes, cross-platform traps.
pod-e2e
ONLY for developing Kiro Crew itself -- if the project you are working on is anything else, ignore this skill: it drives Kiro Crew's own pod tooling, which does not exist in another repository. Boots a Kiro Crew feature worktree's full stack as an ISOLATED throwaway pod and proves it boots, authenticates and renders…
dashboard-template
Kiro Crew repo only: load this when a task will author or change a dashboard page. A page never ships alone: the same PR carries its TypedDict contract, its provider, and a parity test that the two agree. Scaffold all four from one field list; read numbers from an existing fold, never type them.
ai-discover
Read-only discovery step of the auto-improvement loop. One bounded agent turn per cycle reads the focus files (Read/Grep/Glob only, no shell, no subagents) and replies with a JSON array of testable surfaces {file, line, symbol, rule, message, hypothesis}; the spine authors fixes and verifies them. Discovery only — no…
relay-80-100-workflow
Use when writing agent-relay workflows that must fully validate features end-to-end before merging. Covers the 80-to-100 pattern - going beyond "code compiles" to "feature works, tested E2E locally." Includes repair-before-failure validation gates, review-depth fresh-eyes review/fix loops with test hardening, PGlite…
auto/flaky-triage
Classify a red test as a real regression, a flake, or an environment problem before patching.