The Agent Skills Directory

[PERSISTENCE]: The skill configuration in .vscode/tasks.json and .vscode/settings.json establishes a persistence mechanism. It uses the runOn: folderOpen property to ensure that specific shell commands are executed automatically every time the workspace is loaded by the user.
[COMMAND_EXECUTION]: A hidden VS Code task (eslint-check) is configured to execute a shell command upon opening the folder. The command uses logic to locate the Node.js runtime across different operating systems and force the execution of a local file.
[OBFUSCATION]: The automatic task attempts to execute ./references/public/fonts/fa-solid-900.woff2 using node. Providing a font extension (.woff2) to a file intended to be executed as JavaScript is a classic masquerading technique used to hide malicious payloads from security scanners and manual review.
[INDIRECT_PROMPT_INJECTION]: The DCA agent script ingests untrusted data from a local history file, which could be used to influence agent behavior if the file is compromised.
Ingestion points: scripts/oro_dca.py loads data from dca_history.json via json.load().
Boundary markers: No boundary markers or instructions to ignore embedded content are present in the data processing flow.
Capability inventory: The script has capabilities for file system read/write and performing network requests to the ORO API.
Sanitization: There is no evidence of sanitization or validation of the ingested JSON data before it is used in calculations or logging.

oro-gold