The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to autonomously reproduce bugs by running failing commands, tests, or endpoints. This involves executing arbitrary shell commands based on its own analysis of the project.
[DYNAMIC_EXECUTION]: In Phase 3 and 5, the skill generates instrumentation code (logging blocks) and injects it directly into the project's source files. This modified code is then executed to collect runtime data, which constitutes dynamic code generation and execution.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted source code and logs, creating a surface for indirect prompt injection. Malicious instructions embedded in the code being debugged could influence the agent's instrumentation logic or reproduction commands.
Ingestion points: Project source code files, user-provided reproduction steps, and runtime logs generated during execution.
Boundary markers: None identified. The skill does not use specific delimiters or instructions to ignore embedded commands within the code it analyzes.
Capability inventory: Includes file system read/write (modifying code and writing logs), arbitrary command execution (running reproduction steps), and network operations (fetch POST for browser logs).
Sanitization: None identified. The skill does not specify any sanitization or validation of the code or logs before they are processed or used to generate new code.
[DATA_EXFILTRATION]: The skill implements a logging mechanism that captures variable states and execution paths. In browser environments, it uses a fetch POST to a 'debug API route', which could be leveraged to exfiltrate sensitive data if the agent is directed to a malicious endpoint or if the logs contain credentials/PII.

debug-mode