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 naveedharri/benai-skills --skill rule-rewritergit clone --depth 1 https://github.com/naveedharri/benai-skillsWrote 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/naveedharri/benai-skills/rule-rewriter)<a href="https://agentmods.dev/skills/naveedharri/benai-skills/rule-rewriter"><img src="https://agentmods.dev/badge/skills/naveedharri/benai-skills/rule-rewriter/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/naveedharri/benai-skills/rule-rewriter"><img src="https://agentmods.dev/badge/skills/naveedharri/benai-skills/rule-rewriter.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 Rogue Agent · line 13 Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.Fix: Prevent the skill from modifying its own code, SKILL.md, or configuration files. Treat skill files as read-only at runtime.
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 | $0.00109 | $0.01169 |
| Opus 5 | $0.00055 | $0.00584 |
| Sonnet 5 | $0.00022 | $0.00234 |
| Haiku 4.5 | $0.00011 | $0.00117 |
Grade A, and why
rule-rewriter 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 6d 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 — 60 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Rule Rewriter
Claude 5 models follow judgment better than they follow rules. A bare "never do X" gets ignored or overfitted; "X breaks because [reason], so avoid it" generalizes. This skill audits an instruction file and rewrites it accordingly.
Grounded in Anthropic's published guidance:
- Rules to judgment: https://claude.com/blog/the-new-rules-of-context-engineering-for-claude-5-generation-models
- Give the why: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices
- Remove verification instructions: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-opus-5
Process
-
Locate the target. Default to the CLAUDE.md in the current working folder. If the user names a skill or another file, use that. If nothing is found, ask for the path and stop.
-
Inventory every instruction line. Classify each into exactly one bucket:
- RETIRED: instructions the model now does by itself. Delete candidates. Examples: "double-check your work", "verify before responding", "think step by step", "be thorough", "always re-read the file", "you are an expert...".
- ONE-SIDED: rules that state a cost or a command without the counterweight, so the model overfits. Rewrite with both sides.
- BARE PROHIBITION: "never / always / do not" lines with no reason. Rewrite as judgment plus the why.
- AGGRESSIVE: "CRITICAL", "YOU MUST", all-caps emphasis written to fix undertriggering on old models. Soften to normal language; Claude 5 overtriggers on these.
- OBVIOUS: anything the model can see by listing the file system or reading the repo. Delete candidates.
- KEEP AS HARD RULE: places where being wrong is expensive. Safety, destructive or irreversible actions, client-facing sends, money, permissions. Do not soften these; a hard rule is correct here.
- KEEP AS GOTCHA: non-obvious project facts the model cannot discover on its own. These are the most valuable lines in the file; leave them alone.
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.
- 6d ago First seen · 60 lines · 109 tokens per session scan A 810d243227f8
rule-rewriter is a skill published in the GitHub repository naveedharri/benai-skills (61 stars, last pushed 7d ago), licensed MIT. It adds 109 tokens to every session and 1,169 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 2026-09-05.
Other skills, from other repositories
project
A single starting point for setting up an AI-assisted project in Claude Cowork. It asks questions about the work, reviews available add-ons, and creates project instructions, custom agents, and connected workflows.
doc-html-slide
A renderer that turns presentation content into a single HTML slide deck that opens directly in a browser. It creates a 16:9 slide sequence with navigation, fullscreen viewing, printing to PDF, and speaker-note controls.
cs-channel-message
A channel-message writing tool for search ads, advertising, customer relationship messages, and app notifications. It uses the NCM sequence—Need, Channel, Moment, Message, and CTA—to adapt wording to where and when customers see it.
design-sync-upload
An uploader for design-system files such as DESIGN.md, tokens, logos, fonts, and images into Claude Design. It can either use an authenticated connection or prepare a folder and guide for manual upload.
design-tokens-transformer
A converter for design tokens, which are named values for colors, fonts, spacing, borders, shadows, and motion. It translates one shared token source into CSS variables and Tailwind or shadcn-style theme files, and can convert them back for checking.
media-higgsfield-explainer
A Higgsfield workflow for making non-photorealistic narrated explainer videos. It pairs each narration line with a 10-second animated clip and joins the clips into one finished video.