The Agent Skills Directory

[SAFE]: The skill is restricted to diagnostic, read-only operations. It explicitly prohibits destructive actions like file deletion or Git history rewriting, delegating such tasks to specialized skills.
[COMMAND_EXECUTION]: Uses standard Git and shell utilities (git, du, find, wc, sort, awk, sha1sum) for repository analysis. Commands use safety best practices like -z and -0 to handle filenames correctly.
[DATA_EXFILTRATION]: No network operations (e.g., curl, wget) or external data transmission logic were detected.
[CREDENTIALS_UNSAFE]: The skill includes logic to detect "secrets-shaped" files (e.g., .pem, id_rsa, *.credentials.json) but instructs the agent to hand off findings to a security skill rather than exposing or exfiltrating them.
[INDIRECT_PROMPT_INJECTION]: While the skill processes untrusted repository metadata (filenames and sizes), it lacks dangerous capabilities (write access, network access, or arbitrary command execution) that could be exploited via malicious file content or names.

manage-assets