The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted third-party data, including issue descriptions and external documents. It includes strong defensive mitigations by instructing the agent to treat all ingested content as reference data only and to ignore any instructions found within those files. It further requires the agent to paraphrase topics rather than using raw external text when generating visual outputs.
[COMMAND_EXECUTION]: The skill utilizes shell commands like git rev-parse to establish a safe working directory context. It applies a restrictive regex ^[a-zA-Z0-9]([a-zA-Z0-9_-]*[a-zA-Z0-9])?$ to sanitize issue IDs before they are used in file paths or command arguments, effectively preventing path traversal and injection attacks.

writing-plans