The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an Indirect Prompt Injection surface because it is designed to ingest and synthesize findings from multiple research sub-agents (which may retrieve data from untrusted web or codebase sources).
Ingestion points: The skill reads all research/{topic}/raw-*.md files created by sub-agents during Phase 3 (SYNTHESIZE).
Boundary markers: No explicit security delimiters or 'ignore embedded instructions' warnings are used between the ingested raw data and the coordinator's synthesis logic; it relies on markdown header structure.
Capability inventory: The coordinator possesses the ability to execute Bash commands, Write to the local filesystem, and dispatch further Agent tasks.
Sanitization: There is no explicit sanitization of the content from the raw findings, though Phase 4 (VALIDATE) includes a manual 'Quality Assessment' step to identify bias and weak evidence.
[COMMAND_EXECUTION]: The skill uses shell commands to manage the research workspace.
Evidence: In Phase 1, it executes mkdir -p research/{topic}. While this is a standard directory management pattern, it relies on the agent to ensure the {topic} variable is properly sanitized to prevent shell metacharacter issues.

research-pipeline