The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to run npx debugsk@latest server start --json, which downloads a package from the npm registry. This package is not associated with a known trusted vendor or organization.\n- [REMOTE_CODE_EXECUTION]: By using npx, the agent downloads and executes arbitrary code from the npm registry at runtime, representing a supply chain risk.\n- [COMMAND_EXECUTION]: The agent is directed to 'insert instrumentation' into the codebase, meaning it will autonomously write and modify source files within the user's environment.\n- [INDIRECT_PROMPT_INJECTION]: The skill requires the agent to read and analyze log files (.logs/*.jsonl). This creates a surface for indirect prompt injection if the logs include data from untrusted external inputs.\n
Ingestion points: Reading and Grepping log files in the .logs/ directory as part of the core investigation loop.\n
Boundary markers: None identified in the instructions for separating log data from agent instructions.\n
Capability inventory: The agent has the ability to write instrumentation (file-write), start the debug server (subprocess/npx), and read log files (file-read).\n
Sanitization: There is no mention of sanitizing or escaping the log content before the agent processes it for analysis.

code-debug-skill