The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from untrusted external sources, including competitor websites, pricing pages, and third-party review platforms.
Ingestion points: Phase 1 and Phase 3 of the Research Process (SKILL.md) ingest content from competitor URLs and review sites like G2, Capterra, and Product Hunt.
Boundary markers: The skill mandates the use of evidence markers ([fact: url, date], [inference]) defined in references/templates.md and includes a "Boundaries" section that restricts the agent from bypassing site controls or contacting competitors.
Capability inventory: The skill uses web-fetching tools (e.g., Firecrawl), directory listing (ls -R), and filesystem write access to the local competitor-profiles/ directory.
Sanitization: The instructions require saving all raw browser or tool captures as markdown files for auditing and verification against the final synthesized profile.
[COMMAND_EXECUTION]: The skill uses a standard directory listing command to verify the presence of evidence before finalizing its output.
Evidence: Phase 4 of the research process (SKILL.md) executes ls -R against the local competitor-profiles/raw/ directory to ensure all required data points were successfully captured.

suede-competitor-profiling