The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a data trust protocol in SKILL.md to mitigate risks from untrusted data sources. Ingestion points: The agent is instructed to handle untrusted data from source code (trusted), config files (verify), and user-provided content such as issue bodies, comments, and API responses (untrusted). Boundary markers: The instructions explicitly mandate that the agent 'Validate at boundaries' for untrusted content and 'Surface the conflict explicitly' if conventions are violated. Capability inventory: The skill provides instructions for general agent capabilities like searching, reading, and writing code, but does not define new tools. Sanitization: The protocol requires the agent to verify config files and validate external inputs before relying on them.
[SAFE]: The skill consists of guidance for repository execution and does not contain malicious code, remote downloads, or credential exposure. The author identity and the URL https://shipshit.dev are consistent with the skill metadata.

context-engineering