SKILL: intel-synthesis

Description

Advanced intelligence processing pipeline optimized for high-context models (Gemini 1.5 Pro/Ultra). Ingests raw multi-source data, performs cross-verification, deduplication, and conflict analysis, and produces authoritative geopolitical briefings.

Usage

Run with the target directory path. Example: openclaw run intel-synthesis --dir /Volumes/Intel/news/2026-02-13_0500

Implementation

This skill leverages the large context window to process entire intel dumps in single passes, reducing hallucination and improving narrative coherence across multiple sources.

1. Ingest & Analyze

• Read MANIFEST.json to validate completeness.
• Load all .json content files.
• Smart Temporal Filter:
  • Ali provides the files, but the content may contain old "Related Articles" or sidebars.
  • Rule: Extract the main narrative event time.
  • Keep: Events occurring within the assigned cycle (Last 12-18 hours).
  • Discard: Obvious historical noise (e.g., "2015", "10 years ago") found in page clutter.
  • Note: Do not auto-reject "yesterday" if it falls within the 12h operational window (e.g., late-night news for morning brief).

2. Analyze (Single-Shot Chain of Thought)

• Persistence Check: Read the most recent [YYYY-MM-DD]_briefing_EN.md from /Volumes/Intel/NewsBriefs/.
• Deduplication: Identify stories that were already reported.
• Update Protocol:
  • If a story is new: Include as a full entry.
  • If a story is an update: Frame it as an update (e.g., "UPDATE: Following the PLC order reported in the previous cycle, UAE forces have begun withdrawing from...") rather than repeating the baseline facts.
  • If no new information exists for a previously reported story: Omit it to maintain high signal-to-noise.
• Entity Extraction: Identify key actors, locations, and equipment.
• Timeline Reconstruction: Order events chronologically, resolving "2 hours ago" vs absolute timestamps.
• Conflict Detection: Highlight discrepancies between sources (e.g., casualty numbers, attribution).
• Source Weighing: Prioritize primary/local sources over aggregators for facts; use aggregators for broad sentiment.