The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from pull request diffs and application runtime states to formulate and execute test cases. This creates a surface for indirect prompt injection if the processed data contains malicious instructions. * Ingestion points: The skill ingests data from external sources including the 'blast-radius' skill output and task-supplied work units in SKILL.md. * Boundary markers: No explicit delimiters or instructions are provided to the agent to disregard instructions found within the data being probed. * Capability inventory: The skill possesses the ability to execute shell commands, perform file system operations outside the repository tree, and invoke secondary code-generation skills like 'poteto-tdd'. * Sanitization: The skill lacks explicit sanitization or validation logic for the external content it processes.
[COMMAND_EXECUTION]: The skill is designed to run literal commands on the host system to interact with a target application's verification driver. * Evidence: The instructions explicitly state: 'A literal command is the entire invocation, not a prefix: run it as written, once.'
[DYNAMIC_EXECUTION]: The skill generates and executes new test code at runtime to verify bugs found during the probing phase. * Evidence: The skill references the 'poteto-tdd' skill for staging failing tests before fixes are implemented.

runtime-probes