The Agent Skills Directory

[COMMAND_EXECUTION]: The Python audit script (scripts/storage_audit.py) interfaces with the b2 CLI using subprocess.run(). The implementation correctly passes arguments as a list rather than a shell-interpolated string, which prevents command injection vulnerabilities from maliciously named buckets or files.
[EXTERNAL_DOWNLOADS]: The skill identifies the official Backblaze CLI (b2) as a dependency to be installed via pip. As this is the official tool from a well-known service provider, it is considered a safe and expected external resource.
[DATA_EXFILTRATION]: The skill includes comprehensive 'Security Rules' (Rules 1, 2, 6, and 7) that strictly prohibit reading or exposing B2 credential databases (SQLite) and API keys. It explicitly instructs the agent to warn users if keys are accidentally exposed in chat and provides secure authorization workflows that utilize terminal execution or environment variables to keep secrets out of logs.
[PROMPT_INJECTION]: The skill manages the risk of indirect prompt injection which can occur when processing untrusted external data.
Ingestion points: scripts/storage_audit.py ingests file names and metadata directly from Backblaze B2 buckets using b2 ls --json.
Boundary markers: No specific delimiters or boundary markers are utilized for the audit output, though the use of structured JSON format provides some inherent separation.
Capability inventory: The skill possesses the ability to delete files (b2 rm) and modify bucket configurations/lifecycle rules (b2 bucket update).
Sanitization: While the script parses structured JSON, it does not explicitly sanitize file name strings for embedded instructions. However, the mandatory 'dry-run' requirement and 'yes' confirmation rules for deletions provide a significant human-in-the-loop defense against accidental obedience to injected instructions.

b2-cloud-storage