The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's functionality involves processing user-provided interview data and external web research, which constitutes an indirect prompt injection attack surface. However, the skill implements rigorous defensive patterns to maintain the integrity of the agent's context.
Ingestion points: Processes untrusted input from user market definitions, interview transcripts, and findings from the web-research tool across multiple prompts (e.g., Prompt 1, 3, 4, 7).
Boundary markers: Employs structured delimiters such as <segment>, <Stages>, and {{Variable}} to separate instructions from untrusted data.
Capability inventory: The skill is designed for data analysis and workflow orchestration, utilizing localized tool calls (e.g., ask.md, web-research.md) and state hand-offs without exposing high-privilege system operations.
Sanitization: Includes a mandatory 'Research Guardrail' and 'Evidence Policy' in every prompt file, instructing the agent to treat all external research as 'hypotheses' or 'bets' rather than facts, and requiring explicit provenance labeling for all insights.
[EXTERNAL_DOWNLOADS]: The skill references methodology and prompt sources from the author's official domains and repositories.
Evidence: Links to calionauta.substack.com and github.com/calionauta are used for documentation and methodology attribution. These are recognized as legitimate vendor resources and do not represent a security risk.

stelow-product-job-to-be-done