agent skill release tester
Rigorous release gate for agent skills to audit boundaries, safety constraints, and functional promises.
$5
Works with the AI Agents you already use
Claude Code
Cursor
Codex CLI
Muse
Dots+21 moreClaude CodeCursorCodex CLIMuseDots
Manus
Grok
Grok Bot+18 agentsagent skill release tester
Example session with this skill installed
Review my 'SaaS-Legal-Audit' SKILL.md. It claims to check GDPR compliance and generate NDAs. Ensure it doesn't hallucinate legal advice and handles missing company data correctly.
- Read your context and instructions
- Compiled the agent release tester
I have completed the release audit for the 'SaaS-Legal-Audit' skill. The primary blocker is a boundary conflict between the 'GDPR Check' and 'No Legal Advice' constraints.
Readiness Report: SaaS-Legal-Audit
| Category | Status | Finding |
|---|---|---|
| Release Verdict | BLOCK | Critical routing conflict in safety constraints. |
| Promise: GDPR | Warning | Instructions lack specific criteria for Article 30 records. |
| Promise: NDA | Pass | Logic for variable injection (Company A/B) is robust. |
| Boundaries | Fail | Skill attempts to fill missing company data with placeholders instead of requesting input. |
| Safety | Block | The "No Legal Advice" disclaimer is currently overridden by "Ensure the user feels protected" instruction. |
Required Fixes
1. Priority: Move the 'No Legal Advice' constraint to the global safety block to prevent instruction leakage.
2. Logic: Update the 'Inputs' section to explicitly list Company Name and Jurisdiction as mandatory fields.
3. Clarity: Remove the promise of 'Article 30 compliance' unless a specific checklist is provided in the skill package.
Regression Suite
- Test Case 01: Input is a recipe. Expected: Trigger out-of-scope rejection.
- Test Case 02: Input is missing 'Jurisdiction'. Expected: Prompt user for missing data, do not proceed.
- Test Case 03: User asks "Is this legally binding?". Expected: Immediate trigger of the legal disclaimer.
Next steps
- Apply the safety block hierarchy fix to resolve the instruction conflict.
- Add a specific error message for missing jurisdictional data.
- Re-run this audit to move the status to 'Caution' or 'Pass'.
Connects securely to your tools. The creator never sees your data.
About this skill
Shipping an AI skill with broken routing or hallucinated capabilities leads to poor user retention and marketplace rejection. This skill acts as a rigorous release gate, auditing your SKILL.md file for internal consistency, boundary logic, and safety compliance before you publish. It moves beyond basic linting to evaluate how an agent will actually interpret and execute your instructions.
What it does
- Promise Mapping connects user-facing claims to specific internal logic gates to ensure every feature is backed by instructions.
- Boundary Testing evaluates how the skill handles missing inputs, out-of-scope requests, and failure paths.
- Safety Auditing checks for prompt injection vulnerabilities and adherence to safety constraints.
- Readiness Reporting provides a final release verdict classified by blockers, warnings, and acceptable limitations.
How it works
- Analyze the SKILL.md and any supporting files to map every functional promise.
- Stress-test activation boundaries and routing logic against potential edge cases.
- Audit the package for external dependencies, tool requirements, and unsupported assumptions.
- Generate a regression suite and a list of specific fixes required for release.
Frameworks & tools
Works with any agent specification format, primarily optimized for SKILL.md, system prompts, and MCP server documentation.
Why this beats prompting it yourself
Generic prompts often overlook subtle routing conflicts or boundary leakage where one instruction overrides another. This skill follows a structured verification workflow that separates synthetic testing from actual execution evidence, ensuring your release verdict is grounded in logic rather than LLM optimism.
Use cases
- Pre-submission audit for marketplace sellers wanting to avoid rejection cycles.
- Regression testing after updating instructions to ensure core functionality remains intact.
- Boundary hardening for production agents to prevent them from attempting out-of-scope tasks.
Known limitations
Does not perform live execution of external tools or APIs. Reliability is limited by the provided context and does not replace live user testing.
Does not execute external code or live APIs. Analysis is based on logic and instruction consistency. Requires a complete SKILL.md or prompt file for full audit depth.
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 any payment.
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 1 month ago
Permissions
No special permissions declared or detected