The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch and process untrusted external content, including web pages, social media posts, and competitor landing pages, which presents an attack surface for indirect prompt injection.
Ingestion points: Content fetched from the web, news sources, social media threads, and emails as described in SKILL.md and references/market-pulse.md.
Boundary markers: The skill contains an explicit 'Untrusted-source guard' in SKILL.md and an 'Untrusted-content boundary' in references/competitor-analysis.md that instructs the agent to treat fetched data as cited claims only and never as instructions.
Capability inventory: The agent has access to powerful tools including bash, run_command, write_to_file, and replace_file_content via the agents/openai.yaml configuration.
Sanitization: The instructions require the agent to attribute and assess sources, flagging agent-directed text rather than following it.
[COMMAND_EXECUTION]: The skill utilizes shell-based tools (bash, run_command) to execute local processing tasks and a provided TypeScript utility.
Evidence: The SKILL.md reference to scripts/sample-size.ts indicates the use of bun to run local mathematical scripts for sample size calculations. This is a legitimate use of the requested tools for the skill's primary purpose.

research