The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill executes the BrowserSkill (bsk) CLI tool via subprocess.run to perform browser automation. Commands are constructed using list-based arguments, which prevents shell injection. Inputs are escaped using json.dumps before being injected into the execution context.
[DYNAMIC_EXECUTION]: In scripts/analyze.py, the skill uses importlib.import_module to load specific reporting modules. This dynamic loading is secured by a hardcoded whitelist of allowed section names, preventing unauthorized module imports.
[EXTERNAL_DOWNLOADS]: The documentation references the BrowserSkill project on GitHub (github.com/Tencent/BrowserSkill). While this is a well-known source, the skill relies on the user to have the tool pre-installed rather than performing automatic downloads.
[DATA_EXPOSURE]: scripts/bridge_browserskill.py contains a hardcoded fallback path to an executable that includes a specific user profile name (C:\\Users\\49707\\...). This is a minor metadata leak regarding the author's development environment.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it processes data from external web sources.
Ingestion points: Data is ingested through browser-context fetch() calls and DOM scraping within scripts/core_api.py and scripts/core_dom.py.
Boundary markers: Absent; the instructions do not implement explicit delimiters for the processed external data.
Capability inventory: The skill has the capability to execute the bsk CLI via subprocess.run and write files to the local system using write_csv.
Sanitization: The skill uses json.loads to handle API payloads and basic string cleaning for scraped text.

sellersprite-rpa