The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill's integration instructions in reference/project-integration.md specify the use of the -ExecutionPolicy Bypass flag when running PowerShell scripts, which disables the system's protections against untrusted code. Additionally, the Chrome launch configuration in reference/chrome-debug-start.ps1 utilizes the --remote-allow-origins=* flag, which allows any website to potentially interact with the browser's debugging port via the Chrome DevTools Protocol.
[COMMAND_EXECUTION]: The skill requires the execution of shell commands to manage the environment, including the use of Stop-Process -Name chrome -Force to kill existing browser instances.
[DATA_EXFILTRATION]: The skill documentation identifies sensitive file paths for credential storage, such as ../secrets/instagram/, which are accessible to the agent and could be exposed if handled improperly.
[DYNAMIC_EXECUTION]: The agent is instructed to generate and execute inline Python scripts (python -c) to perform runtime image conversion tasks, such as converting PNG files to JPEG.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the web and local project files, creating a surface for injection attacks.
Ingestion points: The agent reads data from the Instagram user interface and local marketing files.
Capability inventory: The skill can execute shell commands, write files, and perform network requests.
Boundary markers: No explicit delimiters or instructions are used to prevent the agent from executing instructions found in the processed data.
Sanitization: There is no process for validating or cleaning the captions or images before they are uploaded to Instagram.

instagram-chrome-post