deep-scout_skill

This skill converts a natural language query into a structured research report with source citations through a four-stage intelligence pipeline.
  • Python

2.5k

GitHub Stars

5

Bundled Files

2 months ago

Catalog Refreshed

3 months ago

First Indexed

Readme & install

Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.

Installation

Preview and clipboard use veilstrat where the catalogue uses aiagentskills.

npx veilstrat add skill openclaw/skills --skill deep-scout

  • _meta.json632 B
  • clawhub.json346 B
  • config.yaml243 B
  • README.md2.1 KB
  • SKILL.md5.7 KB

Overview

This skill implements a multi-stage intelligence pipeline that turns a natural-language query into a structured research report with full source citations. It performs web search, filters results with an LLM, fetches content with tiered escalation (fast fetch, Firecrawl, browser snapshot), and synthesizes a final report in several styles. Outputs include citations, optional file writing, and configurable depth, freshness, and locale parameters.

How this skill works

On invocation the skill runs four stages: SEARCH collects search results from Brave with configurable freshness and count. FILTER uses a prompt-driven LLM step to score and keep the most relevant results, deduplicating by domain and selecting the top URLs up to the requested depth. FETCH attempts content extraction per URL in tiers: web_fetch, Firecrawl (JS-aware), and browser snapshot with an LLM extractor; if all fail it falls back to the original snippet. SYNTHESIZE assembles fetched content into the chosen style (report, comparison, bullets, timeline) and produces a final report with inline source blocks and citations.

When to use it

  • Produce a concise, sourced research report from web material.
  • Compare vendors, products, or claims with side-by-side dimensions.
  • Create a timeline or bullet summary of recent events or trends.
  • Archive investigative research with direct source excerpts and links.
  • Automate repeated monitoring queries with freshness and country settings.

Best practices

  • Start with a focused natural-language question to reduce noise.
  • Adjust --depth to balance breadth vs. synthesis cost (1–10).
  • Use --freshness and --country to localize results for relevance.
  • Set --min-score to filter low-relevance hits before fetching.
  • Prefer --style comparison with explicit --dimensions for structured comparisons.

Example use cases

  • Competitive intelligence: compile features and claims from top vendor pages into a comparison report.
  • Incident triage: gather recent coverage and timeline of a security event, with source excerpts.
  • Market research snapshot: summarize analyst posts, blogs, and news for a target market in a single report.
  • Due diligence: fetch product pages, press releases, and coverage and synthesize a sourced dossier.
  • Content brief: generate a referenced outline and sources for writing an article or whitepaper.

FAQ

The pipeline escalates from fast fetch to Firecrawl to browser snapshot. If all tiers fail it uses the original search snippet and marks the source as [snippet only].

How do I control how many sources are fully fetched?

Use --depth N (1–10) to set how many filtered URLs are fully fetched; search_count and min_score influence which results are considered for that depth.

Built by
VeilStrat
AI signals for GTM teams
© 2026 VeilStrat. All rights reserved.All systems operational