The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection (Category 8) because it retrieves and processes untrusted data from the open web to provide search results to the agent.
Ingestion points: Data enters the context via WebResults (Title, SiteName, Snippet, Summary) and ImageResults (Title, SiteName) fetched from the Volcengine API in scripts/web_search.py.
Boundary markers: Absent. The format_output function in scripts/web_search.py formats results as plain text without using delimiters or instructions to ignore embedded commands.
Capability inventory: The script has network access (via requests.post) and the ability to read local configuration files.
Sanitization: None detected. The script performs basic string formatting and truncation but does not escape or sanitize content fetched from external websites.
[EXTERNAL_DOWNLOADS]: The skill depends on the requests library and optionally the veadk-python package for IAM credential management.
Evidence: _require_requests function in scripts/web_search.py checks for the library; references/docs-index.md and references/setup-guide.md point to official vendor repositories on GitHub and PyPI.
[COMMAND_EXECUTION]: The script uses shlex.split when parsing a local environment file (/root/.openclaw/.env) to safely handle shell-style syntax in configuration values, which is a defensive measure.

byted-web-search