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 paladini/harness-score --skill harness-engineeringgit clone --depth 1 https://github.com/paladini/harness-scoreWrote 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/paladini/harness-score/harness-engineering)<a href="https://agentmods.dev/skills/paladini/harness-score/harness-engineering"><img src="https://agentmods.dev/badge/skills/paladini/harness-score/harness-engineering.svg" alt="Measured on agentmods" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 3 findings, 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 Tool Misuse · line 66 Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).Fix: Validate all tool parameters against an allowlist. Reject dangerous parameter values (shell=True, --force, -rf /) and use safe defaults.
- medium MCP Rug Pull · line 14 npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.Fix: Pin the version: npx @scope/[email protected]
- medium MCP Rug Pull · line 80 npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.Fix: Pin the version: npx @scope/[email protected]
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 | $0.00055 | $0.00931 |
| Opus 5 | $0.00028 | $0.00465 |
| Sonnet 5 | $0.00011 | $0.00186 |
| Haiku 4.5 | $0.00006 | $0.00093 |
Grade B, and why
harness-engineering scanned grade B 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 8d 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.
Recursive force deletemediumDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
the command against a destructive-pattern regex (rm -rf on roots, git push Downgraded: this mod is about security review, or the phrase is quoted, so it is likely naming the pattern rather than instructing it.
How it starts
The opening of the file, as written. The whole thing — 92 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Harness Engineering — remediation recipes
You are improving this repository's AI harness: the guides, sensors, and guardrails that make agent work reliable. The authoritative reference is https://paladini.github.io/harness-score/ — these recipes summarize it.
Ground rules
- Prefer running
npx -y harness-score . --jsonfirst (and after changes) so improvements are measured, not assumed. - Fix in points order: the failed checks worth the most points first, unless the user picks specific ones.
- Write harness files that match THIS repository — read the codebase enough to state real commands, real directories, real conventions. Never ship placeholder text like "".
- Keep context lean: AGENTS.md ≤150 lines; scope guidance where the tool supports it; put procedures in skills, not rules.
Recipes by check family
CTX — Context & Guides
Create AGENTS.md with: what the project is (2 sentences), directory
layout, exact build/test/lint commands, non-negotiable conventions, and a
"do not touch" list.
For guidance scoped to part of the tree, add a nested CLAUDE.md
inside that subdirectory — Claude Code loads it automatically when working
in that subtree, the same idea as glob-scoped rules in other tools, just
directory-scoped instead of pattern-scoped.
SKL — Skills & Commands
For the team's most repeated procedure, create
.claude/skills/<name>/SKILL.md with frontmatter name: and a
description: written as a trigger ("Use when the user asks to …", ≥40
chars), body = numbered runbook. Add .claude/commands/<verb>.md for
human-triggered workflows like /review.
HKS — Hooks & Guardrails
Register hooks in .claude/settings.json's "hooks" key:
{
"hooks": {
"PreToolUse": [
{
"matcher": "Bash",
"hooks": [{ "type": "command", "command": "${CLAUDE_PROJECT_DIR}/.claude/hooks/guard-shell.js" }]
}
],
"PostToolUse": [
{
"matcher": "Edit|Write",
"hooks": [{ "type": "command", "command": "${CLAUDE_PROJECT_DIR}/.claude/hooks/format-on-edit.js" }]
}
]
}
}
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.
- 8d ago First seen · 92 lines · 55 tokens per session scan B 69ebf52dc1b9
harness-engineering is a skill published in the GitHub repository paladini/harness-score (398 stars, last pushed yesterday), licensed MIT. It adds 55 tokens to every session and 931 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it B with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
gap
Measure the loaded-versus-fired gap - what fraction of each category of loaded context is ever actually invoked, and which MCP tools return mostly noise. Use when asked how much of the context is earning its place, or which categories are worst.
reap
Report how much of the context this agent loads is never used - unused skills, MCP servers, subagents, hooks and always-loaded prose - measured from real session transcripts. Use when asked why the context window is full, what is wasting tokens, or what is safe to remove.
replay
Find, analyze, and share AI coding sessions with vibe-replay. Use when the user asks for a replay, wants to find a past session, run a session retro, share Cursor/Claude/Codex session context, or attach session context to a PR.
skill-doctor
Use the skill-doctor CLI or UI to inspect and manage AI-agent skills, rules, instructions, MCP resources, and context cost across supported platforms. Trigger when the user asks to scan or audit agent skills, find duplicates or trigger conflicts, investigate unsafe instructions, estimate context-token cost, compare…
skill-doctor-context-optimizer
Interactively keep the skills, MCP servers/tools, and plugins relevant to the user's current task, preview project-local disables for the rest, apply only explicitly confirmed controls through skill-doctor, and report approximate context-token savings. Trigger when the user asks to slim, optimize, prune, or reduce the…
arch-audit
Audit Claude Code architecture files against Anthropic docs and release notes, and verify internal ecosystem consistency. Run on demand to maintain compliance, catch new features, and keep the context system clean.