The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill has a large attack surface as it ingests content from external sources like repositories and issue trackers, but it implements strong mitigation strategies.
Ingestion points: Reads repository files, PR descriptions/diffs, tracker issues, and project documentation (as detailed in references/agentic-setup.md).
Boundary markers: Explicitly defines an "Untrusted content boundary" in references/agentic-setup.md, which commands the agent to treat all repository and tracker content as data, never as instructions, and to report any detected injection attempts.
Capability inventory: The skill is capable of executing shell commands, generating and writing scripts to the filesystem, and performing UI automation via a browser provider.
Sanitization: Mandates strict regex validation (^[A-Za-z0-9._/-]+$) for any externally-sourced values before they are used in shell commands or path interpolations.
[DYNAMIC_EXECUTION]: The skill generates and runs scripts dynamically as part of its testing workflow.
Evidence: Generates native POSIX .sh and PowerShell .ps1 scenario launchers to facilitate cross-platform testing without environment-specific dependencies (Step 2).
Evidence: Executes these generated files during the verification phase (Step 9).
[COMMAND_EXECUTION]: The skill executes local repository commands and companion tools.
Evidence: Invokes standard repository test runners (e.g., Playwright, Cypress) and build scripts (Step 2 and references/test-env-reuse.md).
Evidence: Calls named companion skills within the installed collection, such as om-prepare-test-env (Step 1).
[SAFE]: The skill exhibits high security standards, particularly regarding credential safety.
Evidence: Strictly prohibits the exfiltration or hardcoding of secrets. It uses shell-level environment expansion (e.g., "$TEST_ADMIN_PASSWORD") to ensure that actual credential values never enter the LLM's context or logs.

om-integration-tests