The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local scripts using uv run python -m runtime.harness to perform research tasks. These scripts include scripts/mcp/nia_docs.py, scripts/mcp/perplexity_search.py, and scripts/mcp/firecrawl_scrape.py. The execution is parameterized based on input research questions provided to the agent.
[EXTERNAL_DOWNLOADS]: The skill retrieves data from the public internet via the Perplexity search engine and Firecrawl web scraper. This content is inherently untrusted and provided by external sources.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8) because it processes untrusted external data into research findings without sanitization or protective delimiters.
Ingestion points: Untrusted data enters the agent context through search results and web scraping via the Perplexity and Firecrawl MCP tools mentioned in SKILL.md.
Boundary markers: The instructions do not define boundary markers or explicit commands for the agent to ignore instructions embedded within the retrieved research data.
Capability inventory: The agent has the ability to execute local Python scripts and write structured handoff files to the filesystem as described in SKILL.md.
Sanitization: There is no evidence of sanitization or filtering of external content before it is synthesized into the final handoff document.

research-agent