humanitarian-intelligence-workflows

FIRST: Use cloudera-agent-studio-workflows for base orchestration patterns.

This skill provides domain-specific operating guidance for humanitarian intelligence pipelines that combine API data, web intelligence, social signals, and LLM synthesis.

When To Use

• Building workflows for country/theme/time-window intelligence briefs.
• Creating recurring monitoring workflows for conflict, displacement, or food security.
• Designing report-ready outputs for operations teams and decision-makers.

Domain Workflow Pattern

1. Intake mission scope (country, theme, time range, decision audience).
2. Collect multi-source evidence (APIs, web, social).
3. Score relevance and filter low-signal data.
4. Synthesize findings by sub-theme and confidence.
5. Generate recommendations with source-backed rationale.
6. Expose endpoint or app for interactive follow-up Q&A.

Minimum Agent Set

• Manager Agent: orchestration, quality gates, fallback handling.
• Data Collection Agent: structured API sources.
• Web Intelligence Agent: contextual reporting and qualitative evidence.
• Social Media Agent: early warning and real-time signals.
• Relevance Filter Agent: semantic filtering and deduplication.
• Report Agent: decision-ready narrative and recommendations.

Source Coverage Priorities

• Conflict: ACLED, INSO, ReliefWeb, contextual news analysis.
• Displacement: UNHCR, IOM DTM, ReliefWeb, social verification.
• Food security: IPC, FEWS NET, ACAPS, market/context intelligence.

Output Contract

Every final report should include:

• Executive summary (what changed + why it matters).
• Key findings by sub-theme with evidence bullets.
• Confidence labeling (high/medium/low) per major finding.
• Action recommendations tied to evidence.
• Source appendix (traceable citations/IDs/URLs).

Required Reading by Phase

| Phase | Read | |------|------| | Source strategy | references/source-prioritization.md | | Report generation | references/report-template.md |

Guardrails

• Do not treat social content as verified fact without corroboration.
• Separate observed evidence from model inference.
• Prefer explicit uncertainty over overconfident claims.
• Preserve provenance for every high-impact recommendation.