The Agent Skills Directory

[DATA_EXFILTRATION]: The documentation in SKILL.md provides diagnostic command examples that access sensitive system state, including environment variables (e.g., IDENTITY) and macOS keychain data via the security utility. These examples demonstrate how to expose credentials and identity secrets during the investigation of build and signing issues.
[COMMAND_EXECUTION]: The utility script find-polluter.sh performs local command execution by running npm test on files identified through user-provided patterns. This is a standard debugging technique but represents a controlled execution of local code.
[DATA_EXFILTRATION]: The debugging methodology encourages the agent to implement diagnostic logging at component boundaries to trace data flow. This practice can lead to the unintended exposure of sensitive operational data, such as authentication tokens, secrets, or personally identifiable information (PII), if the data being logged is not properly sanitized.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it instructs the agent to ingest and analyze untrusted external content, such as error messages and stack traces, to guide its diagnostic and fix-implementation logic.
Ingestion points: Error messages, stack traces, and log files parsed during Phase 1.
Boundary markers: No explicit markers or instructions are provided to isolate or neutralize potential commands embedded in the ingested logs or error data.
Capability inventory: File system access, local code execution via npm test, and diagnostic shell command execution (e.g., codesign, security).
Sanitization: The process does not mandate any validation or sanitization of the input data before analysis.

systematic-debugging