rookie-ricardo/erduo-skills
Overview
This skill fetches content from a preset list of URLs, filters for high-quality technical items, and generates a daily Markdown report. It uses an orchestrator agent to schedule subagents, supports headless browser fetching for JS-heavy pages, and maintains a cache to avoid duplicates. The output is a standard Markdown file with top curated entries and metadata.
How this skill works
On each run the orchestrator reads source and cache configuration, then dispatches parallel subagent batches to fetch and extract structured items. Subagents return JSON items with title, summary, key points, keywords and a quality score; the orchestrator deduplicates, scores, and ranks results, applying early-stop rules once enough high-quality items are collected. For SPA or JS-rendered pages the system uses a headless browser flow and updates cache.json with run stats and fingerprints.
When to use it
- Daily technical briefing for engineering or product teams
- Automated curation of frontier research and productivity tips
- When you need a reproducible, timestamped Markdown digest
- Environments where some sources require JS rendering
- Operations that must avoid duplicate or low-quality items
Best practices
- Maintain sources.json with prioritized, trusted sources and update regularly
- Tune filtering constraints to your audience (e.g., advanced research vs. practitioner tips)
- Keep cache.json healthy to prevent re-including historical articles
- Prefer parallel SubAgent execution; fall back to serial mode if worker agents are unavailable
- Use the quality_score weighting and source credibility to adjust final ranking
Example use cases
- Generate a 20-item daily report for an engineering team summarizing relevant papers, product launches, and workflow tips
- Produce a condensed report for weekly newsletters by running daily and aggregating top items
- Automate monitoring of a list of blogs and communities, surfacing only high-value technical posts
- Run in environments lacking worker agents by using serial mode while still preserving cache and quality checks
FAQ
The orchestrator records the failure, may retry or mark the source disabled, and proceeds—single-source failures do not stop the report generation.
How does the system avoid duplicates and repeated coverage?
It deduplicates by exact URL, by title similarity threshold, and checks content fingerprints against cache.json to avoid historical repeats.
Can I add custom extraction rules or new sources?
Yes. Add entries to sources.json with priority and extraction method; the scheduler will include them in the next run and adjust weights accordingly.
