The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves processing external data, including source code, bug reports, and test execution results. This ingestion of untrusted content presents a surface where malicious instructions could be embedded (e.g., in code comments, log files, or issue descriptions) to influence the agent's behavior during the 'Reproduce' or 'Fix' phases.
Ingestion points: External source code files, test logs, and reproduction scripts as defined in the 'Reproduce first' section of SKILL.md.
Boundary markers: The skill does not define specific delimiters or wrapping instructions for external content to prevent instruction leakage, though it does mandate logic gates.
Capability inventory: File system modification (code fixes) and shell command execution (test and reproduction checks).
Sanitization: Explicitly mandates the redaction of credentials, tokens, and connection strings using a '' placeholder in commands, files, and summaries.
[COMMAND_EXECUTION]: The instructions require the agent to build and run reproduction checks, which involve executing shell commands or test suites. This execution environment is a potential vector if the reproduction steps or the test environment are influenced by malicious data within the project being debugged.

do-debug