digest

<!-- autoresearch: variation B — curatorial discipline (filter → distill → structure → sanity-check) folded with direct-curl xAI + web/RSS inputs and memory-aware dedup; RSS feed-reading + item-selection absorbed from rss-digest as an additional source class -->

> **${var}** — Selects the digest's topic and which source classes feed it. Grammar:
> - `""` (empty) → **digest's default sources** (WebSearch + xAI/Grok + aggregators), no topic filter — a broad daily digest.
> - `"<topic>"` → topic-focused digest on the default web sources, filtered to `<topic>` (e.g. `"solana"`, `"AI agents"`, `"rust"`).
> - `"rss"` → **RSS-only**: pull feeds from `memory/feeds.yml`, no topic filter.
> - `"rss: <topic>"` → RSS-only, filtered to `<topic>` (e.g. `"rss: rust"`).
> - `"<topic> +rss"` → default web sources **and** RSS feeds combined, both filtered to `<topic>` (e.g. `"AI agents +rss"`).

Today is ${today}. Generate and send a daily **${var}** digest.

The whole point of a digest is **signal, not volume**. A reader skimming for 60 seconds should walk away with three things they didn't know that morning and one of them should change a decision they'd make this week. Anything that doesn't clear that bar gets cut.

## Preamble — orient and parse the selector

1. Read `memory/MEMORY.md` for high-level context and tracked topics, and scan the last 3 days of `memory/logs/` so you can dedup against anything already reported.
2. Parse `${var}` into **`{topic, sources}`**:
   - Strip a trailing `+rss` → adds RSS to the default web sources. Remainder is the `topic`.
   - A leading `rss:` (or the bare token `rss`) → **RSS-only** source set; text after the colon is the `topic` (empty = no filter).
   - Otherwise the whole string is the `topic` and `sources = default web` (empty string = default web, no topic filter).
   - Resulting `sources` is one of: `web` (default), `rss` (RSS-only), or `web+rss` (both).
3. The `topic`, when non-empty, is a filter applied to **every** source class — web queries are scoped to it and RSS items must match it (title/description/tags). Empty topic = keep all relevant items.

## Config (RSS source)

When `sources` includes `rss`, this skill reads feed URLs from `memory/feeds.yml`. If the file doesn't exist yet, create it (write mode) with the shape below, or — if you have nothing to seed it with — log a one-line note and treat the RSS source as empty for this run.

```yaml
# memory/feeds.yml
feeds:
  - name: Example Feed
    url: https://example.com/rss
  - name: Another Feed
    url: https://example.com/atom.xml
```

## Phase 1 — Gather (cast a wide net)

Pull from the source classes selected by `${var}`. Never rely on a single one — if `sources = web`, use at least two of the web classes below; if `sources = web+rss`, RSS counts as one class and you still want a second.

### Web sources (active when `sources` is `web` or `web+rss`)

1. **WebSearch** (built-in) — run 2 distinct queries:
   - `"${topic}" news ${today}` (broad). If `topic` is empty, run a general query for the day's notable stories in the operator's tracked areas (from `memory/MEMORY.md`).
   - One narrower query you choose based on `${topic}` (e.g. for "solana" → `"solana" launches OR funding OR exploit ${today}`; for "AI agents" → `"agent framework" OR "agentic" release ${today}`).
2. **xAI x_search via Grok** — pulls the X/Twitter signal layer. `XAI_API_KEY` is injected into this skill's environment (declared in `requires:`) and is the **primary** path; see **Fetching the X signal** below for the full contract (attempt the curl before any fallback, set the Bash tool `timeout` ≥180000, record the true failure reason).

   **Path A — X.AI API (primary):** a direct `curl` to the Responses API. First confirm the key with `[ -n "$XAI_API_KEY" ] && echo KEY_PRESENT || echo KEY_UNSET`; if `KEY_PRESENT` (it will be), this path is required. When you run the curl, set the Bash tool's `timeout` to at least `180000`.
   ```bash
   FROM_DATE=$(date -u -d "yesterday" +%Y-%m-%d 2>/dev/null || date -u -v-1d +%Y-%m-%d)
   TO_DATE=$(date -u +%Y-%m-%d)
   PROMPT="Search X for substantive, recent posts about: ${topic:-the most notable technology, AI, and crypto stories today}. Date range: $FROM_DATE to $TO_DATE. Return up to 10 high-signal posts — prioritize verifiable claims, launches, funding, releases, exploits, or hard data over hot takes. For EACH post return: @handle, the full text, date posted, exact engagement counts (likes, retweets, replies; 0 if unknown), and the direct link https://x.com/handle/status/ID. Return a numbered list."
   jq -n --arg p "$PROMPT" --arg fd "$FROM_DATE" --arg td "$TO_DATE" \
     '{model:"grok-4.6", input:[{role:"user",content:$p}], tools:[{type:"x_search",from_date:$fd,to_date:$td}]}' \
     > /tmp/xai-digest-payload.json
   HTTP=$(./secretcurl -s -o /tmp/xai-digest.json -w '%{http_code}' --max-time 150 -X POST "https://api.x.ai/v1/responses" \
     -H "Content-Type: application/json" -H "Authorization: Bearer {XAI_API_KEY}" -d @/tmp/xai-digest-payload.json)
   echo "xai http=$HTTP bytes=$(wc -c </tmp/xai-digest.json)"
   ```
   On `HTTP=200` with a non-empty body, parse it with `jq -r '.output[] | select(.type == "message") | .content[] | select(.type == "output_text") | .text'` and feed each post (handle, text, engagement, permalink) into the web-candidate pool. A slow curl is **not** a missing key — do not treat a timeout as key-unavailable.

   **Path B — WebFetch/WebSearch fallback (last resort, lower quality):** only if the key is `KEY_UNSET`, or Path A returned a non-2xx / empty body / timeout. Attempt a WebFetch to a public X search URL like `https://x.com/search?q=${topic}&f=live`, or a `site:x.com "<topic>" after:${FROM_DATE}` WebSearch; extract a few top posts and prefer results within the last 48h. Record the **true reason** (`key-unset` | `http-<code>` | `empty` | `timeout`) in the log — never "XAI_API_KEY unavailable" when the key was set. If this also returns nothing, skip the