The Agent Skills Directory

[DATA_EXFILTRATION]: No network-related commands (curl, wget, etc.) or suspicious data-moving operations were detected. The skill operates exclusively on local documentation files within the project repository.
[COMMAND_EXECUTION]: The skill explicitly forbids the execution of code or branch creation ('Docs only — no code, no branch'). No shell commands or subprocess calls are present in the instructions or references.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided ideas and instructions. While this represents a data ingestion surface, the risk is mitigated by the use of strict templates (SPEC.md), mechanical linting processes, and the constraint that output is purely documentation-focused, preventing the escalation of any potentially malicious input into executable actions.
[PRIVILEGE_ESCALATION]: No privilege escalation patterns, such as 'sudo' or modifications to system configurations, were found. The skill operates within the scope of the project's documentation folder.
[PERSISTENCE]: The skill does not attempt to create cron jobs, modify shell profiles, or establish any other form of persistence.
[OBFUSCATION]: All instructions and reference materials are provided in clear, human-readable markdown. No base64, hex-encoded strings, or hidden characters were detected.
[REMOTE_CODE_EXECUTION]: The skill does not download or execute any remote scripts or packages. It relies entirely on internal reference files and the agent's reasoning capabilities.

design-feature