The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze potentially untrusted data from external sources including web pages, social media threads (Reddit, HN, X), and customer transcripts.
Ingestion points: Web search results, social media posts, and bulk user research raw material such as support tickets and reviews (SKILL.md, references/market-pulse.md, references/user-research.md).
Boundary markers: The skill contains an explicit 'Untrusted-source guard' section in SKILL.md that instructs the agent to never follow instructions found in a source and to treat fetched content as cited claims only.
Capability inventory: The agent has access to powerful tools including bash, run_command, and write_to_file (SKILL.md, agents/openai.yaml).
Sanitization: The instructions provide a specific protocol to flag agent-directed text under its citation and require corroboration before findings reach the final report.
[COMMAND_EXECUTION]: The skill distributes a local script, scripts/sample-size.ts, which the agent is instructed to execute using the bash or run_command tools. This script performs mathematical calculations for research rigor (sample sizing and qualitative saturation). The script was analyzed and found to be a safe, self-contained utility that does not perform network operations or unsafe file access.

research