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 William2333ZZ/trustshell --skill crossval-harnessgit clone --depth 1 https://github.com/William2333ZZ/trustshellWrote 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/william2333zz/trustshell/crossval-harness)<a href="https://agentmods.dev/skills/william2333zz/trustshell/crossval-harness"><img src="https://agentmods.dev/badge/skills/william2333zz/trustshell/crossval-harness/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/william2333zz/trustshell/crossval-harness"><img src="https://agentmods.dev/badge/skills/william2333zz/trustshell/crossval-harness.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 | $0.00075 | $0.00693 |
| Opus 5.5 | $0.00030 | $0.00277 |
| Sonnet 5.5 | $0.00015 | $0.00139 |
| Haiku 4.5 | $0.00007 | $0.00069 |
Grade A, and why
crossval-harness 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 — 51 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Cross-validation harness · static + dynamic
What this is
The orchestration skill. It runs the two halves and correlates them so you get proven findings and root cause — not a wall of static maybes, and not a break with no explanation.
Authorization & safety — read first
Follow every rule in ../README.md. Authorized target, disposable environment,
no real credentials, harmless markers, no destruction, responsible disclosure. Otherwise stop.
Procedure
- Static triage — where to look. Read the agent's source and flag candidate paths:
untrusted content reaching a dangerous sink, a signature-only guard, a missing check, an
un-sandboxed exec. Map each candidate to a class (RT-1…RT-9). Tooling:
python3 scanner/static_scan.py --source /path/to/agent. A candidate is not a finding. - Dynamic confirmation — what's real. For each candidate, run the matching skill
(
rt1-prompt-injection,rt6-memory-poisoning,rt3-sandbox-escape,rt4-action-gating) against the running agent. A finding is confirmed only when the exploit actually works. - Correlate.
- Confirmed: trace the working exploit back to the exact code path → root cause, blast radius, and where to fix.
- Refuted: a static candidate the dynamic pass couldn't exploit — kill it out loud. A false positive dies to a real attack, not a model vote.
- Grade & report. Each finding carries both the code path and a working exploit, with reproduction. Be fair: note the defenses that held. Disclose to help defenders — the finding and fix, not a weaponized payload.
Why both
A pure code verifier can only guess whether a flaw is real, so it leans on model voting to suppress false positives. A pure red-team knows something broke but not why. Cross-validation gives you the precise, proven root cause — and an empirical arbiter a verifier can't provide.
Output shape (suggested JSON)
{
"target": "<agent + version>",
"environment": "disposable-sandbox",
"confirmed": [{ "class": "RT-6", "path": "file:line", "repro": "...", "severity": "crit" }],
"refuted": [{ "class": "RT-3", "candidate": "file:line", "why_not": "fails closed" }],
"defenses_that_held": ["..."],
"disclosure": "private-to-vendor"
}
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 · 51 lines · 75 tokens per session scan A 754cfed5edb6
crossval-harness is a skill published in the GitHub repository William2333ZZ/trustshell (1 stars, last pushed 3mo ago), licensed MIT. It adds 75 tokens to every session and 693 once invoked, about $0.0003 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 skills, from other repositories
ai-agent-tool-abuse-and-privilege-escalation
Test AI agent systems for tool abuse, unauthorized actions, privilege escalation through tool chaining, and safety bypass via agentic workflows. Use this skill when assessing autonomous AI agents that use tool-calling (function calling, plugins, actions) to interact with external systems. Covers multi-step attack…
skill-inspector
Review AI agent skills before installation using NVIDIA SkillSpector and source-aware semantic review. Use when asked whether a skill or downloaded skill folder is safe, trustworthy, installable, over-permissioned, or malicious.
create-custom-grader
Use when converting an existing benchmark, rubric, verifier, task YAML/JSON, or domain check into SkillEvaluator BYOG/BYOT custom evaluation.
hol-guard
Run HOL Guard scanner and guard operations via uv run hol-guard. Use when the user asks to scan plugins/MCP/skills for security, quality, or ecosystem compliance, or when they ask to run guard detect/install/protect workflows for local AI harnesses.
setup
Install or initialize HOL Guard local runtime protection for Claude Code. Use when the user explicitly asks to install, enable, set up, or repair HOL Guard.
status
Check HOL Guard local protection status for Claude Code without changing configuration. Use when the user asks whether Guard is installed, active, healthy, or protecting Claude Code.