The Agent Skills Directory

[PROMPT_INJECTION]: The instructions focus on structured documentation and do not contain patterns for bypassing safety filters or overriding agent behavior.
[DATA_EXFILTRATION]: No hardcoded credentials, sensitive file access (e.g., SSH keys, AWS configs), or unauthorized network operations were identified.
[EXTERNAL_DOWNLOADS]: The skill does not perform automated downloads or execution of remote code. It incorporates a manual validation step in the 'Resilience phase' to check the security of proposed project dependencies.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user stories. While this creates an ingestion surface for indirect instructions, the logic is constrained to producing documentation (TRI.md), which poses minimal risk to the agent environment.
[COMMAND_EXECUTION]: No dangerous system commands or privilege escalation attempts were found. The skill primarily interacts with the file system to read local documentation and write markdown files.

requirements-interlock