The Agent Skills Directory

[SAFE]: The skill follows security best practices for AI agents, including maintaining a clear boundary between the generic skill logic and project-specific environment data. It explicitly restricts its operations to local development environments and enforces read-only access for database verification.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted data from the application under test, including browser logs, network payloads, and database records. While this creates a surface for indirect prompt injection, the skill includes a rigorous 'Proof Discipline' framework. This requires the agent to verify all findings across multiple independent signals (UI snapshots, SQL SELECT queries, and network logs) before reporting a failure, effectively mitigating the impact of poisoned data.
[COMMAND_EXECUTION]: The skill includes Python scripts (resolve_customization.py and bootstrap_adapter.py) to manage local configuration. These scripts were flagged by static analysis for their use of subprocess.run within test files (scripts/test_bootstrap_adapter.py and scripts/test_resolve_customization.py). Upon review, these calls are used exclusively to execute the skill's own scripts during automated unit testing and do not pose a risk of arbitrary command execution.

manual-testing