Research Source Discovery

Overview

Use this skill to run Phase A candidate discovery for engineering research tasks. The objective is breadth with traceability: every query and shortlist decision must be logged.

Source Priority

1. GitHub (github / github-official) for repository and code discovery.
2. Reddit and Hacker News (mcp-reddit, mcp-hackernews) for implementation lessons.
3. Curated lists (awesome lists and tool indexes).
4. Blog/news discovery (brave, kagisearch, dappier).
5. X via API (mcp-api-gateway) when credentials are available.

Do not start from generic web search unless higher-priority sources are blocked.

Required Outputs

Write:

• reports/research-ros/<date>-phase-a-candidates.md
• reports/research-ros/<date>-discovery-records.jsonl

Records must match docs/research/schemas/discovery-record.schema.json.

Discovery Rules

• Log full query strings.
• Prefer recency (pushed, since) and quality constraints (stars, archived status, engagement thresholds).
• Deduplicate by canonical URL.
• Tag source class: github, social, curated, blog, news, x.

Exit Criteria

• 50-150 candidate sources collected.
• At least 3 distinct source classes.
• Each candidate has shortlist reason and confidence.