The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the execution of shell commands, specifically wrangler for remote storage operations, openssl for generating unique object keys, and pbcopy to modify the system clipboard.
[DATA_EXFILTRATION]: The skill is designed to move local data to a remote public environment. It uploads files to a Cloudflare R2 bucket named public and generates a public URL via the domain assets.shahram.dev. This mechanism allows any file accessible to the agent to be made publicly available on the internet.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided file paths to perform network operations, creating a surface for indirect attacks. Evidence Chain:
Ingestion points: User-specified file paths in prompts (SKILL.md).
Boundary markers: Absent; there are no instructions to verify the content or sensitivity of the file before uploading.
Capability inventory: Network upload via wrangler, subprocess execution of openssl, and clipboard interaction via pbcopy (SKILL.md).
Sanitization: None; the skill does not filter for sensitive files such as .env, .ssh/id_rsa, or configuration directories, allowing potentially critical secrets to be uploaded to a public bucket.

file-upload