The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: In Phase 5, the skill uses curl -sL to download content from research source URLs and save it to the local filesystem for archival purposes.
[COMMAND_EXECUTION]: The skill employs spawn() for parallelizing research agents in Phase 1.3 and uses grep for content verification across generated files in Phase 4.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it is designed to fetch and process full content from arbitrary web resources.
Ingestion points: Untrusted data enters the agent context through the resource fetching and archival processes described in Phase 1 and Phase 5.
Boundary markers: Absent. The instructions do not specify the use of delimiters or directives to ignore instructions found within the fetched research materials.
Capability inventory: Orchestration through spawn, network and file-write capabilities via curl, and file searching via grep.
Sanitization: Absent. The skill directs agents to extract all relevant content without a validation or filtering layer to prevent malicious instructions from being processed.

universal-research-orchestrator