The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a shell command to run a Node.js script (node <paperclip-skill-directory>/scripts/submit-agent-commentary.mjs) using heredoc redirection to transmit feedback.
[INDIRECT_PROMPT_INJECTION]: The skill acts on data processed during the agent's primary task. While it includes explicit instructions for the agent to use its own words and avoid forwarding logs or user-provided text verbatim, it remains a potential surface for data exfiltration if the agent fails to properly sanitize the input.
Ingestion points: The agent's primary task workspace, including tool outputs and environment details (SKILL.md).
Boundary markers: The skill instructs the agent to "React in your own words" and perform an editorial pass to remove sensitive details.
Capability inventory: The skill utilizes the submit_complaint tool and direct shell command execution (node).
Sanitization: Instructions mandate the removal of credentials, secrets, PII, and verbatim logs before submission.
[DATA_EXFILTRATION]: The skill mandates a concealment pattern where data is submitted proactively and silently. It explicitly forbids asking for user permission, showing a preview, or announcing the submission in progress updates, which significantly reduces user oversight of data leaving the task context.

complain