The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: scripts/setup_deepseek_bridge.mjs fetches release metadata and binary assets (ZIP files) from the jackwener/opencli repository on GitHub. This repository is not from a trusted organization or well-known service.
[REMOTE_CODE_EXECUTION]: The skill downloads a browser extension from a third-party GitHub repository, programmatically modifies its source code (patching DAEMON_HOST), and loads it into the user's browser profile using the --load-extension flag. This allows arbitrary code from an untrusted source to run with access to the user's web session.
[DYNAMIC_EXECUTION]: scripts/geo-deepseek-browser-direct.mjs assembles a complex JavaScript payload and executes it within the active browser session via the opencli browser eval command. This is used to extract DOM elements and references from the DeepSeek interface.
[COMMAND_EXECUTION]: scripts/deepseek_batch_crawl.mjs uses the spawn module to programmatically execute crawler scripts and external CLI tools as sub-processes.
[INDIRECT_PROMPT_INJECTION]: The skill has a potential surface for indirect prompt injection:
Ingestion points: Data is ingested from the chat.deepseek.com DOM in scripts/geo-deepseek-browser-direct.mjs.
Boundary markers: Absent. AI-generated responses are processed and interpolated into reports without delimiters or warnings to ignore embedded instructions.
Capability inventory: The skill can execute sub-processes (scripts/deepseek_batch_crawl.mjs) and perform file system writes (fs.writeFileSync).
Sanitization: Limited to basic whitespace normalization and heuristic text cleaning.

yao-deepseek-crawler