The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The upload_server.py script and the ShareX_Custom_Uploader.sxcu configuration file both contain a hardcoded authentication token (secret_vault_token_123). This token is used to authorize uploads to the local Flask server.
[DATA_EXFILTRATION]: The skill includes a Python script in gemini.md that accesses the ShareX history file located at %USERPROFILE%\Documents\ShareX\History.json. This file contains sensitive metadata about every capture event, including original file paths and remote upload URLs. Additionally, the local upload server in upload_server.py is configured to listen on 0.0.0.0, exposing the upload endpoint and stored files to the entire local network.
[COMMAND_EXECUTION]: The skill provides instructions and examples for executing local shell commands, including starting a Python Flask server and invoking the ShareX executable (ShareX.exe) with various command-line arguments across all files.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves the agent processing untrusted data, specifically screenshots and OCR text.
Ingestion points: Visual input from the ShareX Image Annotator and text extracted via the OCR engine as described in gemini.md and SKILL.md.
Boundary markers: None specified in the instructions for processing this external data.
Capability inventory: The agent is empowered to execute shell commands (ShareX.exe), write files (via Python scripts), and perform network operations (via the custom uploader).
Sanitization: No validation or sanitization of the OCR content or screenshot metadata is performed before the agent processes or acts on it.

sharex