The Agent Skills Directory

[DATA_EXFILTRATION]: The skill reads the full content of local markdown files and transmits them to an external API at https://www.proofeditor.ai. This represents a deliberate transfer of potentially sensitive information to a third-party service.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from a remote collaborative editor, creating a vulnerability to instructions embedded in that content. 1. Ingestion points: Remote document state is fetched via the v3/document endpoint in references/workflows.md. 2. Boundary markers: The skill suggests confirming with the user before overwriting files, providing a process-level boundary but no content-level sanitization. 3. Capability inventory: The agent has Bash (curl, jq, mv) and Write capabilities for filesystem modification. 4. Sanitization: No evidence of sanitization or safety filtering of markdown content is present.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute curl commands for all interaction with the web API.
[CREDENTIALS_UNSAFE]: The skill handles authentication tokens (accessToken and ownerSecret) and suggests storing the owner secret in session-based shell environment variables, which could be exposed in system logs or command histories.

ce-proof