The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The agent is instructed to fetch raw HTML from external URLs to verify testimonials, which allows untrusted data into the context.
Ingestion points: Usage of curl on a user-specified <references page URL> in reference/best-practices.md.
Boundary markers: Absent; no instructions provided to delimit or ignore instructions within the fetched HTML.
Capability inventory: Access to MCP tools for data modification and shell command execution (curl, python).
Sanitization: No instructions for the agent to sanitize the fetched content before processing.
[COMMAND_EXECUTION]: The skill describes using curl and python snippets to verify content integrity, which involves processing external inputs through shell commands.

creating-offers