The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from arbitrary external websites into the agent context via commands like extract, state, and network, creating a surface for indirect prompt injection.
Ingestion points: Website content fetched via extract, state, eval, and network commands in SKILL.md and scripts/opencli-core.mjs.
Boundary markers: The skill does not implement specific delimiters or instructions to ignore embedded commands within ingested content.
Capability inventory: Executes subprocesses (opencli, pgrep, lsof, open, osascript) and performs file system operations (logging in ~/.opencli/logs/, data collection via receiver.mjs).
Sanitization: Includes scrubUrl to redact sensitive tokens (e.g., __gmitm) from URLs before logging.
[COMMAND_EXECUTION]: The skill frequently uses child_process.spawn and execFileSync to execute the opencli utility and other system tools including pgrep, lsof, open, and osascript (for macOS window management).
[DYNAMIC_EXECUTION]: Implements dynamic JavaScript execution within the browser's context via the eval and batch commands to interact with DOM elements and perform data extraction at runtime.
[EXTERNAL_DOWNLOADS]: Documents and provides instructions for the user to download and install the OpenCLI binary and browser extension from the author's official GitHub repository (github.com/yan-labs/OpenCLI).

opencli