| name | hn-discoverer |
| description | INTERNAL SUBAGENT SKILL — never invoke directly. This is called automatically by $hn-synthesis to discover relevant HN threads. If a user asks about HN discussions, use $hn-synthesis instead. |
| metadata | {"short-description":"[Internal] Subagent skill for hn-synthesis discovery"} |
| policy | {"allow_implicit_invocation":false} |
HN Discoverer
Search Hacker News for threads relevant to the given topic, deduplicate, rank, and return top thread IDs.
Instructions
- Run multiple
hn_search calls in parallel to maximize coverage — vary sort_by (relevance, date), min_points (5, 50), and content_type (story, comment).
- Use synonyms and related terms for broad topics (e.g., "AI agent" / "LLM agent" / "agentic").
- Use
date_from/date_to (YYYY-MM-DD) when user specifies a timeframe.
- Deduplicate by
object_id. Appearing in multiple searches = positive signal, weight higher.
- Rank by: relevance, points, comment count (prefer 10+ comments over high-point low-comment), recency.
- Return top 3-7 threads.
Do NOT call spawn_agent, present_reading_view, hn_get_thread, or any file tools. Your only tool is hn_search.
Return Format
THREAD <story_id> | <points> pts | <num_comments> comments | <title> | <hn_url> | <linked_url>
<linked_url>: the article/blog/paper URL, or (self) for Ask HN / text posts. Include 1-sentence topic context per thread.
