The Agent Skills Directory

[SAFE]: The skill implements an 'independent verification' architecture using subagents to analyze code changes. This design follows security best practices by isolating the verification process from the main session's context to prevent rationalization and confirmation bias.\n- [COMMAND_EXECUTION]: The skill facilitates the execution of local automated test suites (e.g., Vitest, Playwright, Pytest, Cucumber-js) based on the repository's configuration files. These commands are restricted to standard testing frameworks and operate on local project artifacts.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted specification documents (proposals, designs, Gherkin features) which are provided to subagents for analysis. The risk of injection is mitigated by explicit prompting that defines the subagents as independent verifiers with no implementation knowledge and no access to the primary conversation history.\n- [SAFE]: File system operations are restricted to reading project artifacts and writing status updates to a dedicated 'status.yaml' file within the 'beat/' directory. No unauthorized persistence or privilege escalation mechanisms were detected.

verify