The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Python subprocess module across several scripts (debug_session.py, collector_ide.py, collector_browser.py) to manage the log collector process, open the system web browser, and launch various IDEs (such as Cursor, VS Code, or Zed) for source code navigation. These operations are integral to the skill's purpose and include safeguards to ensure target files are within the authorized workspace.
[EXTERNAL_DOWNLOADS]: The dashboard frontend (index.html, dashboard-deps.js) fetches common web libraries, including React, SWR, and Tailwind CSS, from established CDNs such as esm.sh and tailwindcss.com.
[INDIRECT_PROMPT_INJECTION]: The skill's architecture involves the agent processing log data received from external runtimes through the /ingest endpoint. While this represents a potential surface for indirect prompt injection if a debugged application emits malicious strings, the skill mitigates this through structured summarization workflows and explicit instructions for the agent to treat logs as evidence rather than directives.
[OBFUSCATION]: A test script (scripts/test_dashboard_utils.mjs) uses a Base64-encoded data URI to load local JavaScript modules for testing in a Node.js environment. This is a common practice for file-based module testing and is not used to mask malicious activity.

debug