The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data through tools such as get_recent_events and get_recent_logs which return raw event attributes and log messages from project event rings. * Ingestion points: The tools get_recent_events and get_recent_logs described in references/tools.md read external project data. * Boundary markers: The documentation does not specify the use of delimiters or 'ignore' instructions when presenting these raw attributes to the agent's context. * Capability inventory: The skill includes mutation capabilities such as set_config_value, set_signal, and ship_experiment, along with the ability to execute local shell commands for verification. * Sanitization: The instructions explicitly state that raw attributes are exposed, which provides a surface for adversarial content in telemetry to influence agent behavior.
[COMMAND_EXECUTION]: The skill provides instructions for the agent to run various local project verification, testing, and stress-testing scripts. * Evidence: The Verification section in references/package.md lists commands including npm run lint, npm run test:js, npm run test:csharp, and node --test packages/server/test/handler.test.js.

wardx-server