The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation instructions in SKILL.md suggest using curl -sL https://skills.sh/paperclip | bash. This pattern downloads and executes code from an external server directly in the user's shell, which bypasses manual inspection and poses a significant security risk if the remote source is compromised.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from the Paperclip API, such as issue identifiers, titles, and descriptions, which are then presented to the agent.
Ingestion points: scripts/issues.sh, scripts/status.sh, and scripts/comment.sh fetch data from API endpoints.
Boundary markers: None; API data is formatted as plain text and displayed to the agent without delimiters or instructions to ignore embedded content.
Capability inventory: The skill can execute CLI tools, perform network requests via curl, and potentially write to the filesystem through standard shell redirection.
Sanitization: While scripts use python3 -c to safely parse JSON structures, they do not filter or sanitize the content of the data fields for potentially malicious instructions.
[PERSISTENCE]: The documentation in SKILL.md recommends that users add their PAPERCLIP_API_KEY and PAPERCLIP_COMPANY_ID to shell configuration files (~/.zshrc or ~/.bashrc). This practice ensures the credentials persist across sessions but stores sensitive authentication tokens in plain text.
[COMMAND_EXECUTION]: The provided bash scripts (scripts/comment.sh, scripts/issues.sh, and scripts/status.sh) execute Python code at runtime using python3 -c. While these instances specifically use environment variables to safely handle data, the pattern involves dynamic execution of code blocks to process API responses.

paperclip