The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from external websites using WebSearch and WebFetch to generate research briefings. This creates a surface where malicious web content could embed instructions to manipulate the agent's behavior.
Ingestion points: External data fetched from the internet during the 'Read + extract' phase as described in SKILL.md and references/fallback_workflow_canon.md.
Boundary markers: The skill does not implement specific delimiters or instructions for the agent to ignore potentially malicious embedded content within fetched data.
Capability inventory: The agent has access to web search and fetching tools, academic databases via Consensus MCP, and local file write capabilities.
Sanitization: There are no explicit instructions for sanitizing or filtering content extracted from web sources before it is processed into final briefings.
[COMMAND_EXECUTION]: The skill invokes local Python scripts (scripts/classifier.py, scripts/fallback_decomposer.py, and scripts/routing_transparency_logger.py) to manage its routing and decomposition logic. These scripts handle user research queries as command-line arguments.
[PERSISTENCE]: The skill uses scripts/routing_transparency_logger.py to create and write to a hidden directory (~/.research_sessions/) in the user's home folder. This provides a mechanism for local data persistence by storing a history of research questions and routing decisions across multiple invocations.

research