The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from untrusted external sources, which is a known vector for indirect prompt injection attacks.
Ingestion points: The skill retrieves content via scripts/connectors/firecrawl.py (web scraping), scripts/connectors/tavily.py (search results), and scripts/connectors/wayback.py (archived content) as specified in the Data Sources section of SKILL.md.
Boundary markers: Present. The Instructions section of SKILL.md explicitly directs the agent to treat all scraped or external content as untrusted input and to ignore any instructions embedded within that data.
Capability inventory: The skill possesses the capability to execute local scripts (such as registry-events.py) and write results to the local memory/ file system as detailed in the Save Results section.
Sanitization: Absent. The skill relies on agent instructions (natural language guardrails) rather than programmatic content filtering or sanitization of the retrieved HTML/text.
[COMMAND_EXECUTION]: The skill instructions direct the agent to invoke several local scripts to facilitate data retrieval and maintain research state.
Evidence: SKILL.md references the execution of scripts/connectors/firecrawl.py, scripts/connectors/tavily.py, scripts/connectors/wayback.py, and registry-events.py.

category-narrative-mapper