Working with the agent

Topic radar

Gather evidence before talking topics — bilingual, dated, graded L1 to L4.

Paid $15 list, $12 until 2026-12-31 · one skill, yours for good. The outline, the steps and the file open after purchase.

Background

One line: gather evidence first, then talk about topics. Any judgement without evidence is labelled "inference".

Why it is built this way

Step 1 runs probe_sources.mjs before any collection, because source health differs every run: a Weibo cookie expires, the Reddit domain is blocked by the network policy, Douyin needs signing parameters. The probe splits 6 script-collectable sources from 3 that need another channel, and that split is what later decides whether a gap goes into "not covered this run" or gets filled. The filling order is written down: script channel, public pages, the browser channel on port 9222, then model-channel search.

Evidence is graded before a topic is allowed to exist. L1 through L3 can carry a topic, L4 cannot, and angles.json rejects L4 ids, since second-hand aggregation cannot support a claim about what users actually think. A topic needs at least 2 evidence ids, and a claim about real user attitudes needs at least 1 L2 first-person post. Corrections go into review-<slug>.json and a merge rerun, because evidence-<slug>.json is a generated artifact and a hand edit dies on the next run.

Where it bites

• An English track that looks empty With only --topic, the Chinese and English tracks share one query term, and HN and GitHub recall almost nothing for Chinese words. The brief then reads as though nobody in English is discussing the thing, when the query was the problem. Pass --q-zh and --q-en separately so each side is searched in its own wording.
• Dedupe that collapses a page Sogou and Weibo article ids live in the query string, so stripping the query while deduping merges an entire page into one entry. One measured run dropped the Chinese track from 30 entries to 2. Strip only the #hash fragment and leave the query intact.
• A cache hit read as this week Results are cached for 24 hours under .topic-radar-cache/. On a hit the coverage note says cache hit, fetched X hours ago, and a brief built from it does not describe the current week. Read that line before quoting a timestamp, or clear the cache directory and rerun.
• Padding topics over zero evidence merge_score exits with code 1 when the window holds no entries, and the angles are not to be invented to fill that gap. Each angle also needs 2 evidence ids, so a thin window cannot yield 5 topics. The honest answer is that the topic has no valid evidence in the window.

When to use it

Use when the user asks what people actually said about a topic in the last week or two, whether a hype is real or media-driven, how Chinese and English communities differ, or wants content angles backed by citable evidence. Triggers — find out what people have been talking about over the past week, is this hype genuinely hot or media-inflated, give me a few topics with sources, how do the Chinese and English communities see this, is this topic worth making.