IRIS Research

Research analysis and investigation orchestration skill. This is the user-facing entry point for the IRIS multi-agent research system. It guides users through topic definition, parameter configuration, and investigation launch via conversational interaction.

Workflow Routing

Workflow	Trigger	File
Investigate	"research", "investigate", "analyze", "study"	`workflows/investigate.md`
Orchestrate	(internal — invoked by investigate)	`workflows/orchestrate.md`
Archive	"archive research", "archive investigation"	`workflows/archive.md`
Status	"research status", "check investigation"	`workflows/status.md`

How It Works

User invokes IRIS research with a topic or intent
Investigate workflow activates:
- Checks if .research/ exists (offers archival if so)
- Gathers topic through conversational refinement
- Confirms understanding with the user
- Asks for synthesis language preference
- Generates a request file in .requests/
- Transitions to the orchestrate workflow
Orchestrate workflow takes over:
- Spawns researcher agents in parallel
- Monitors completion and handles failures
- Spawns the synthesizer agent
- Status tracked in request file (initialized → researching → synthesizing → completed)
Results appear in .research/

Principles

Conversational Refinement

Never assume you understand the topic. Always:

Restate what you understood
Ask for confirmation using ask_user
Only proceed after explicit user approval

Progressive Disclosure

Start with the broad topic, then drill into specifics:

What do you want to investigate?
What specific aspects interest you most?
What questions should the investigation answer?
What language should the final report be in?

Request File Contract

Every investigation produces a request file in .requests/ with:

Full frontmatter (name, topic, language, status, etc.)
Detailed body describing the investigation scope
Updated by the orchestrator as the investigation progresses

Context Files

See workflows/investigate.md for the main investigation workflow
See workflows/archive.md for the archival workflow
See workflows/status.md for checking investigation status

Guardrails

This skill and all agents it triggers are read-only research tools.

The ONLY writable directory is .research/ and .requests/
No code execution that modifies the system
No privilege escalation
No network requests to external APIs on behalf of targets
No destructive filesystem operations