yt-briefing 0.15.0 → 0.15.1

package/.claude/skills/yt/SKILL.md

@@ -36,6 +36,7 @@ while true:
 - **A.** Take `out.summary` (markdown) and `out.pending` (metadata).
 - **B.** In the same turn, as **your chat text** (NOT command output — the UI does not show it), paste `summary` **verbatim**: no paraphrase, no shortening, no comment, no "see above". The user must see it before the popup. B is **unconditional** — it does not depend on there being anything else to write, so an iteration that returns no `skipped` line still opens with the pasted summary, never straight with the tool call. _(If the user says "I don't see the summary" — you skipped B. In Claude Code a PreToolUse gate refuses to record the rating in step D when the pending video's id is missing from your chat text; if it fires, paste the summary, ask for the rating again, then record it.)_
 - **C.** In the same message call `AskUserQuestion` — **1 call, 1 question** (everything in one step), phrased in `output_lang`:
+  - **Put the summary into the question text too**, opening with the channel, title and video id — e.g. `@channel — "title" (videoId, type)` followed by the summary points. Some clients don't show chat text above an open popup, so the popup is what the user actually reads; it is also what the gate in step D accepts as proof when the chat text reaches the transcript late.
   - The question (e.g. "Rating?") with three options whose descriptions explain: **OK** = neutral (no effect on the filter), **Weak** = worthless (teach the filter to skip such titles), **Research** = break the loop and dig into this video's content together. The digits never appear in the popup — internally map to `--rating`: **OK → 1, Weak → 0**; Research maps to no digit, it exits the loop (see Research mode). There is **no positive rating** — keeping the channel is the implicit positive; you only down-rate noise (`0`) or steer with a comment.
   - **Other** is the comment / stop / research channel (no second question): the user types free text. If it equals `stop` (case-insensitive, trimmed) — or the popup is dismissed (✕) — **end the loop**. If it **starts with `?`**, the text after the `?` is a **research question** → Research mode. Otherwise it is a **comment**: **distill** the user's raw text into a clean, generalizable rule, and infer the rating — clearly negative → `0`, otherwise → `1`.
 - **D.** Act on the answer:

package/dist/lib/summary-gate.js

@@ -19,9 +19,19 @@ export function isRatingWrite(payload) {
     return command.includes(RATING_SCRIPT) && command.includes('--rating');
 }
 /**
- * Did the agent actually paste the summary? The summary carries the video's watch URL, so its id
- * appears verbatim in the pasted text. Only the assistant's own text blocks count: the id also
- * travels through tool calls and their results, and neither is shown to the user.
+ * Did the user actually see the summary? The summary carries the video's watch URL, so its id
+ * appears verbatim wherever it was shown. Two surfaces count:
+ *
+ * - the assistant's own text blocks (step B, the paste above the popup);
+ * - the `AskUserQuestion` call itself — the popup's question text is what the user reads while
+ *   rating, and some clients don't show chat text above an open popup at all.
+ *
+ * The popup is also the only surface that is reliably on disk in time: the harness can persist an
+ * assistant's text blocks late — in background sessions only once the turn ends — while the
+ * popup's tool_use entry is written as soon as the call is made. Gating on text alone blocked
+ * ratings whose summary the user had plainly read (measured 2026-10-04: the popup entry carrying
+ * the id was in the transcript, the text block with the same id was not, across several messages
+ * of one turn). Other tool calls and all tool results still don't count — the user sees neither.
  */
 export function summaryWasPasted(transcript, videoId) {
     for (const line of transcript.split('\n')) {
@@ -36,8 +46,16 @@ export function summaryWasPasted(transcript, videoId) {
         }
         if (entry.message?.role !== 'assistant' || !Array.isArray(entry.message.content))
             continue;
-        if (entry.message.content.some((b) => b.type === 'text' && b.text?.includes(videoId)))
+        if (entry.message.content.some((b) => shownToUser(b, videoId)))
             return true;
     }
     return false;
 }
+function shownToUser(block, videoId) {
+    if (block.type === 'text')
+        return block.text?.includes(videoId) ?? false;
+    if (block.type === 'tool_use' && block.name === 'AskUserQuestion') {
+        return JSON.stringify(block.input ?? {}).includes(videoId);
+    }
+    return false;
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "yt-briefing",
-  "version": "0.15.0",
+  "version": "0.15.1",
   "description": "A self-learning YouTube briefing engine: it sweeps the channels you follow, filters noise in two stages (title, then transcript), summarizes the rest in your language, and adapts to your ratings — one video at a time.",
   "type": "module",
   "bin": {