Works with the AI Agents you already use
Claude Code
Cursor
Codex CLI
DeepSeek
Muse+22 moreClaude CodeCursorCodex CLIDeepSeekMuse
Dots
Manus
Grok
Grok Bot+18 agentsTested Guardrails for Coding Agents
These rules were tested on two small models with automated grading.
$19
About this skill
Most rule packs are guesses. These rules were tested on two small models with automated grading. The package includes the evidence summary (EVIDENCE.md), with results at every level:
- Each rule alone: pass rates on rule-sensitive tasks went from 56% to 81% on gpt-6-luna and from 63% to 93% on claude-haiku-5-5.
- Each skill's rules together: as good as the right rule alone (86% vs 84%, 98% vs 99%) and far above no rule (67% and 70%).
- Choosing the right skill: both models loaded the right skill for 100% of 120 runs on the requests that needed a skill, and loaded no skill for 88% of 24 runs on requests that needed no skill.
- No side effects: ordinary tasks passed 18 of 18 with every skill.
- Weakest result: on gpt-6-luna, scope tasks passed 28% with the requested-changes skill and 28% with its rule alone (EVIDENCE.md §2).
- Small: each skill is about 210–340 tokens and loads only when it fits the task.
EVIDENCE.md also shows what failed and was left out: a one-skill version with a menu, an all-in-one version, and a rule that showed no gain.
The task bank, grader and run logs are not included, so the numbers can be read but not re-run. Small test banks, two models, short requests: these are evidence, not guarantees.
Usage example: Fix the failing test in tests/test_api.py Works with: claude-code, cursor, codex-cli
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.
- Small task banks: 9 runs per rule per model, so one run moves a cell by 11 points.
- Two models only (gpt-6-luna and claude-haiku-5-5). Results on other models will differ.
- Decision tasks, not live work: rule tasks ask for a decision about prior turns; they don't watch a model edit real files over a long session.
- Short requests: the skill-choice requests were short and clear. Messier real requests weren't tested.
- Summary only: the task bank, grader and run logs don't ship, so the results can't be re-run from the package.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 23 days ago
Permissions
No special permissions declared or detected