The Agent Skills Directory

[DATA_EXFILTRATION]: The skill's diagnostic guidelines in SKILL.md recommend logging environment variables using env | grep IDENTITY and system keychain information via security list-keychains. These actions expose potentially sensitive credentials, tokens, and system paths to the agent's context and logs.\n- [COMMAND_EXECUTION]: The skill contains a utility script find-polluter.sh that executes npm test on local files. It also instructs the agent to run various shell commands for evidence gathering, such as codesign, security, and git.\n- [DYNAMIC_EXECUTION]: The find-polluter.sh script dynamically identifies and executes test files within the local project directory using npm test.\n- [INDIRECT_PROMPT_INJECTION]: The debugging workflow involves ingesting and analyzing untrusted error messages, stack traces, and logs. The skill lacks boundary markers or sanitization steps to prevent malicious instructions embedded in this data from affecting the agent's operation.\n
Ingestion points: SKILL.md (Phase 1) and root-cause-tracing.md (Observing symptoms).\n
Boundary markers: Absent.\n
Capability inventory: Shell execution (npm, security, codesign, git).\n
Sanitization: Absent.

systematic-debugging