The Agent Skills Directory

[SAFE]: The skill is designed to manage Architecture Decision Records (ADRs) by reading from and writing to local directories such as docs/adr/ and docs/architecture/. These operations are well-defined and limited to the skill's stated purpose.
[SAFE]: The 'one question per message' constraint acts as a behavioral guardrail, preventing the agent from generating large blocks of text that could be used for complex prompt injection or exfiltration attempts.
[SAFE]: No remote network operations (e.g., curl, wget), hardcoded credentials, or obfuscated code were found in the skill definition.

draft-adr