Borrowing it
Nothing to install: this file belongs to FlorianBruniaux/ctxharness. 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/FlorianBruniaux/ctxharness/main/.claude/skills/eval-rules/SKILL.mdgit clone --depth 1 https://github.com/FlorianBruniaux/ctxharnessWrote 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/florianbruniaux/ctxharness/eval-rules)<a href="https://agentmods.dev/skills/florianbruniaux/ctxharness/eval-rules"><img src="https://agentmods.dev/badge/skills/florianbruniaux/ctxharness/eval-rules/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/florianbruniaux/ctxharness/eval-rules"><img src="https://agentmods.dev/badge/skills/florianbruniaux/ctxharness/eval-rules.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
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 | $0.00077 | $0.01401 |
| Opus 5.5 | $0.00031 | $0.00560 |
| Sonnet 5.5 | $0.00015 | $0.00280 |
| Haiku 4.5 | $0.00008 | $0.00140 |
Grade A, and why
eval-rules 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 — 149 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Rules Evaluator
Discover all rule files, validate their structure and glob patterns against the real project, then run an interactive session to confirm (or improve) each rule.
The goal is not just to score — it is to leave the rules directory in better shape than it was.
When to Use
- First time writing
.claude/rules/files (validate before committing) - A rule seems to never trigger, or fires on every file
- Migrating
@imports from CLAUDE.md to path-scoped rules - Periodic hygiene: "are these rules still relevant to how we work?"
- Before shipping ctxharness as a new project to users
Key Concepts
| Mechanism | When it loads | Notes |
|---|---|---|
@file in CLAUDE.md |
Session start, always | Even inside a conditional sentence |
No paths: in rule |
Session start, always | Same cost as @import |
paths: frontmatter |
When Claude reads a matching file | Trigger = Read tool, not Write |
The paths: field is the main lever for keeping rules contextual. An always-on rule with 80 lines loads on every session even if you're fixing a typo in README.md.
Scoring Criteria (12 pts per rule)
| # | Criterion | Max | What is checked |
|---|---|---|---|
| 1 | frontmatter block | 1 | File has YAML frontmatter (--- delimited) |
| 2 | paths: field | 2 | Present (1pt) + at least one pattern listed (1pt) |
| 3 | pattern validity | 3 | Each pattern matches ≥1 file in project (up to 3 patterns checked) |
| 4 | scope | 2 | Not dead (≥1 match) + not too broad (<30% of project source files) |
| 5 | content quality | 3 | Has clear header/title (1pt) + rules are specific/actionable (1pt) + under 150 lines (1pt) |
| Bonus | focus | +1 | Under 15 rules in file |
Thresholds:
- ✅ Good: ≥10/12 (≥83%)
- ⚠️ Needs work: 7–9/12 (58–82%)
- ❌ Fix: <7/12 (<58%)
Always-on rules (no paths: field): skip criteria 2, 3, 4. Score on 5 pts max. Flag with 🔵 and go through the interactive step to decide if scoping is needed.
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 · 149 lines · 77 tokens per session scan A 2fa02701cd89
eval-rules is a skill published in the GitHub repository FlorianBruniaux/ctxharness (26 stars, last pushed 12d ago), licensed MIT. It adds 77 tokens to every session and 1,401 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 2026-08-30.
Other skills, from other repositories
review
Adversarial review of an increment before it ships, by a fresh context or another model family; every finding cites path:line and is re-verified. Use for "review", "qa", "critique this".
sw-review
Adversarial review of an increment before it ships, by a fresh context or another model family; every finding cites path:line and is re-verified. Use for "review", "qa", "critique this".
review
Independently assess designs, plans, implementations, PRs, or repository quality; return evidence and SHIP, FIX-FIRST, or BLOCKED without applying fixes.
fix
Resolve blocking review gaps through bounded repairs and independent re-review; diagnose stalled attempts without expanding scope.
deslop
Clean up the prose, code or interface already in scope — protect its meaning, voice and behaviour, remove what only adds reading work, and return findings instead of edits when a review was asked for.
ce-code-review
Structured code review for bugs, regressions, tests, and standards. Use before PRs or when asked for review; interactive mode can fix locally, while mode:agent reports only for pipeline callers.