indrasvat · indrasvat · Jun 29, 2026
diff --git a/.claude-plugin/marketplace.json b/.claude-plugin/marketplace.json
@@ -4,12 +4,12 @@
     "name": "indrasvat",
     "url": "https://github.com/indrasvat"
   },
-  "description": "15 Claude Code skills for dev workflows, CI/CD, Kubernetes, browser automation, and AI-assisted development",
+  "description": "16 Claude Code skills for dev workflows, CI/CD, Kubernetes, browser automation, cross-platform signal research, and AI-assisted development",
   "plugins": [
     {
       "name": "indrasvat-skills",
       "source": "./",
-      "description": "15 Claude Code skills for dev workflows, CI/CD, Kubernetes, browser automation, and AI-assisted development",
+      "description": "16 Claude Code skills for dev workflows, CI/CD, Kubernetes, browser automation, cross-platform signal research, and AI-assisted development",
       "category": "productivity"
     },
     {

diff --git a/.claude-plugin/plugin.json b/.claude-plugin/plugin.json
@@ -1,7 +1,7 @@
 {
   "name": "indrasvat-skills",
-  "description": "15 Claude Code skills: CI gating, exec plans, PR shipping, code cleanup, PR triage, K8s diffing, migration safety, API compat, rollout health, CRD impact analysis, PRD generation, iTerm2 automation, Cloudflare deployment, CodeRabbit reviews, and browsing-as-you (shared authenticated Chrome over CDP)",
-  "version": "1.5.0",
+  "description": "16 Claude Code skills: CI gating, exec plans, PR shipping, code cleanup, PR triage, K8s diffing, migration safety, API compat, rollout health, CRD impact analysis, PRD generation, iTerm2 automation, Cloudflare deployment, CodeRabbit reviews, browsing-as-you (shared authenticated Chrome over CDP), and grokery (grok-powered cross-platform signal research)",
+  "version": "1.6.0",
   "commands": "./commands/",
   "skills": "./skills/",
   "author": {

diff --git a/README.md b/README.md
@@ -1,14 +1,14 @@
 # Claude Code Skills
 
-A marketplace of personal [Claude Code](https://claude.ai/code) plugins: 15 skills for AI-assisted dev workflows, plus `dootdashaa`, a glanceable statusline.
+A marketplace of personal [Claude Code](https://claude.ai/code) plugins: 16 skills for AI-assisted dev workflows, plus `dootdashaa`, a glanceable statusline.
 
 ## 🎯 Overview
 
 This repository is a Claude Code marketplace exposing two independently-installable plugins:
 
 | Plugin | Install command | What it adds |
 |---|---|---|
-| **indrasvat-skills** | `/plugin install indrasvat-skills@indrasvat-skills` | 15 skills (CI gating, exec plans, PR shipping, K8s diffing, PRD generation, iTerm2 automation, browsing-as-you, …) |
+| **indrasvat-skills** | `/plugin install indrasvat-skills@indrasvat-skills` | 16 skills (CI gating, exec plans, PR shipping, K8s diffing, PRD generation, iTerm2 automation, browsing-as-you, grokery signal research, …) |
 | **dootdashaa** | `/plugin install dootdashaa@indrasvat-skills` | Single-line, Nerd-Font-icon statusline. `~9ms` p50 render budget. |
 
 Both share one `/plugin marketplace add` step (see [Installation](#installation)). Each can be installed, updated, and uninstalled on its own.
@@ -17,7 +17,7 @@ Both share one `/plugin marketplace add` step (see [Installation](#installation)
 
 Glanceable, single-line. One carrier per signal (icon, text, or colour — never two). Zero emoji. Pre-computed git cache keeps the hot path under the 300ms Claude Code debounce. See [`plugins/dootdashaa/README.md`](plugins/dootdashaa/README.md) for env-var options and the install / uninstall command list; design report is at [`plugins/dootdashaa/docs/DESIGN.html`](plugins/dootdashaa/docs/DESIGN.html).
 
-### Skills (15 total)
+### Skills (16 total)
 
 #### Universal Dev Workflow (5 new)
 
@@ -39,6 +39,12 @@ Glanceable, single-line. One carrier per signal (icon, text, or colour — never
 | **rollout-check** | `/rollout-check` | Verify K8s deployment health — pods, events, HPA, logs. |
 | **crd-impact** | `/crd-impact` | Find all controllers, webhooks, RBAC, and manifests affected by a CRD change. |
 
+#### Research & Signals (1 new)
+
+| Skill | Command | Description |
+|-------|---------|-------------|
+| **grokery** 🛒 | `/grokery` | Grok-powered cross-platform signal scan (X + HN + GitHub + Reddit + Polymarket + web), engagement-ranked. A `last30days`-style brief with **no API keys or browser login** — X/Reddit/web run through the `grok` CLI. |
+
 #### Original Skills (4)
 
 #### 📋 **prd-generator**
@@ -127,7 +133,7 @@ Inside a `claude` session, register the marketplace once and install whichever p
 ```
 /plugin marketplace add indrasvat/claude-code-skills    # once per machine
 
-# 15 skills (CI gating, exec plans, PR shipping, K8s diffing, PRD generation, browsing-as-you, ...)
+# 16 skills (CI gating, exec plans, PR shipping, K8s diffing, PRD generation, browsing-as-you, grokery, ...)
 /plugin install indrasvat-skills@indrasvat-skills
 
 # Statusline

diff --git a/skills/grokery/SKILL.md b/skills/grokery/SKILL.md
@@ -0,0 +1,72 @@
+---
+name: grokery
+description: >
+  Scanning the live internet for what people are actually saying about a topic,
+  then synthesizing an engagement-ranked brief — grok-powered first-party X
+  search plus Hacker News, GitHub, Reddit, Polymarket, YouTube and the open web,
+  with no API keys or browser login. Use when the user says "grokery", "what's
+  the buzz on", "what is X / the internet saying about", "scan the discourse",
+  "vibe check", "trend check", "social signals on", "what are people saying
+  about <topic>", "research the chatter on", or wants real-time cross-platform
+  sentiment, a comparison ("A vs B"), or a shareable HTML signal brief on a
+  tech / culture / product / person topic.
+allowed-tools: Bash, Read, Write, WebFetch
+argument-hint: <topic | A vs B> [--days N] [--mode hiring|competitors] [--github-user H] [--html]
+---
+
+# grokery 🛒
+
+Grok-powered cross-platform signal scan. Like `last30days`, but the auth-walled,
+key-hungry sources (X, Reddit, grounded web search) all run through the **`grok`
+CLI on your X subscription** — so there is **no Bird client, no browser login,
+no Brave/Perplexity keys**. _"Searches people, not editors. Ranked by upvotes,
+likes, and real money."_
+
+## Prerequisites
+- **`grok`** CLI, logged in (`grok login`) — powers X + Reddit + web. Required for those sources; everything else still runs without it.
+- **`curl`** + **`jq`** — for Hacker News, GitHub, Polymarket (keyless).
+- Optional: **`yt-dlp`** (YouTube), **`GITHUB_TOKEN`** (higher GitHub limits), **`python3`** (`--store` trend DB + HTML render).
+
+## Quickstart
+Run the orchestrator — it fans out every source in parallel and prints combined
+raw signal (also saved to `~/Documents/Grokery/<slug>-raw.md`):
+
+```bash
+bash skills/grokery/scripts/grokery.sh "grok cli" --days 30
+```
+> The grok pass is one combined session over X + Reddit + web and is the slow part — typically **3–7 min** (default `--timeout 420`s; after that it degrades gracefully and every other source still returns). The keyless sources (HN/GitHub/Polymarket) return in seconds. Run it and wait; do not abort.
+
+Then **you synthesize** the raw output into the final brief (this is the agent's
+job, mirroring last30days' "scripts gather, agent synthesizes" split):
+read [`references/ranking.md`](references/ranking.md) and
+[`references/modes.md`](references/modes.md), then produce the brief.
+
+## Pipeline (what to do)
+1. **Resolve** — grok identifies key handles / subreddits / repos / hashtags internally (entity resolution).
+2. **Gather** — `grokery.sh` queries all enabled sources in parallel.
+3. **Dedup** — merge the same story across platforms into one cluster ([ranking.md](references/ranking.md)).
+4. **Rank** — order by real engagement (likes, upvotes, views, $ volume), not SEO.
+5. **Synthesize** — write a tight narrative with citations; surface genuinely witty/viral one-liners verbatim ([modes.md](references/modes.md)).
+6. **Emit** — markdown by default; `--html` → self-contained dark brief ([html-brief.md](references/html-brief.md)).
+
+## Modes & options (detail → [references/modes.md](references/modes.md))
+- **`A vs B vs C`** — run `grokery.sh` once per side, then compare side-by-side.
+- **`--mode competitors`** — auto-discover top peers and run the comparison.
+- **`--mode hiring`** / `--github-user <h>` — focus/career signals; person mode.
+- **`--days N`**, **`--max N`**, **`--save-suffix`**, **`--store`** (SQLite trend log), **`--no-save`**.
+- **`EXCLUDE_SOURCES=youtube,polymarket`** env to suppress sources.
+- **`eli5`** — after a run, ask to rewrite the synthesis jargon-free.
+
+## References (load on demand)
+| File | When to read |
+|---|---|
+| [`references/sources.md`](references/sources.md) | Per-source detail, grok mapping, optional deps, honest gaps vs last30days |
+| [`references/modes.md`](references/modes.md) | Comparison / competitors / hiring / person / eli5 recipes + synthesis & wit guidance |
+| [`references/ranking.md`](references/ranking.md) | Cross-platform engagement normalization + dedup/clustering |
+| [`references/html-brief.md`](references/html-brief.md) | HTML emit: `render-html.sh` or hand-craft from `brief-template.html` |
+
+## Honest gaps vs last30days
+- **Paid-only sources:** TikTok / Instagram / Pinterest / Threads **native** metrics and Digg's curated leaderboard need the paid ScrapeCreators API and are **out of scope** — grok's web/X pass surfaces mentions, not first-party engagement numbers.
+- **Data fidelity trade-off:** last30days reads X/Reddit through deterministic API clients (exact counts). grokery reads them through the grok LLM, which is faster to set up but can **approximate or misreport** engagement numbers. Treat grok-sourced metrics as indicative, prefer the cited URL as ground truth, and spot-check any number you'll quote. The curl sources (HN / GitHub / Polymarket) are exact.
+
+Everything else is at or near parity, with far less setup.
diff --git a/skills/grokery/references/brief-template.html b/skills/grokery/references/brief-template.html
@@ -0,0 +1,51 @@
+<!DOCTYPE html>
+<html lang="en"><head><meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>grokery — TOPIC</title>
+<!--
+  grokery HTML brief template. Self-contained: inline CSS, no JS, no external
+  assets. Fill the placeholders (TOPIC, WINDOW, DATE, SOURCES) and the CONTENT
+  slot below, then save as the brief file. Keep it offline-ready.
+-->
+<style>
+:root{
+  color-scheme:dark;
+  --bg:#000;--card:#16181c;--border:#2f3336;--text:#e7e9ea;--text2:#c8cdd2;--dim:#71767b;
+  --blue:#1d9bf0;--purple:#7856ff;--green:#00ba7c;
+}
+*{box-sizing:border-box}
+body{margin:0;background:var(--bg);color:var(--text);
+ font:16px/1.6 -apple-system,BlinkMacSystemFont,"Segoe UI",Inter,Roboto,Helvetica,Arial,sans-serif;letter-spacing:-.003em}
+.wrap{max-width:760px;margin:0 auto;padding:32px 20px 80px}
+.badge{display:inline-block;font-size:12px;font-weight:800;letter-spacing:.08em;text-transform:uppercase;
+ color:#fff;background:linear-gradient(135deg,var(--blue),var(--purple));padding:5px 12px;border-radius:999px}
+h1{font-size:30px;font-weight:850;letter-spacing:-.02em;margin:18px 0 4px}
+.meta{color:var(--dim);font-size:13px;margin-bottom:8px}
+.verdict{background:var(--card);border:1px solid var(--border);border-left:3px solid var(--blue);
+ border-radius:12px;padding:14px 16px;margin:18px 0;color:var(--text2)}
+h2{font-size:21px;font-weight:800;margin:32px 0 6px;padding-bottom:6px;border-bottom:1px solid var(--border)}
+h3{font-size:16px;font-weight:750;color:#7cc6f7;margin:20px 0 4px}
+p,li{color:var(--text2)}
+ul{margin:8px 0;padding-left:20px}li{margin:5px 0}
+a{color:var(--blue);text-decoration:none}a:hover{text-decoration:underline}
+strong{color:#f3f5f6}
+code{font-family:ui-monospace,"JetBrains Mono",Menlo,Consolas,monospace;font-size:.86em;
+ background:var(--card);border:1px solid var(--border);border-radius:5px;padding:.05em .4em;color:#cfd9e0}
+.bar{height:6px;border-radius:3px;background:linear-gradient(90deg,var(--blue),var(--purple));margin:2px 0 8px}
+.foot{margin-top:40px;padding-top:16px;border-top:1px solid var(--border);color:var(--dim);font-size:13px}
+@media print{body{background:#fff;color:#000}.verdict,code{background:#f4f4f4}a{color:#06c}}
+</style></head>
+<body><div class="wrap">
+  <span class="badge">grokery · signal brief</span>
+  <h1>TOPIC</h1>
+  <div class="meta">Window: last WINDOW days · DATE · sources: SOURCES</div>
+
+  <div class="verdict"><strong>Verdict —</strong> one-to-three-sentence dominant take + strongest counter-take.</div>
+
+  <!-- CONTENT: themes as <h2>, sub-points as <h3>/<ul>, citations as inline <a>,
+       engagement numbers in <strong>. Optional <div class="bar"></div> under a
+       heading to visualize relative engagement. -->
+
+  <!-- Footer: state which sources actually returned, e.g. "Sources: X, HN, GitHub, Polymarket". Do not claim sources that failed or were skipped. -->
+  <div class="foot">Generated by <code>grokery</code> — grok-powered cross-platform signal scan. · Ranked by engagement, not SEO. · Sources: SOURCES.</div>
+</div></body></html>
diff --git a/skills/grokery/references/html-brief.md b/skills/grokery/references/html-brief.md
@@ -0,0 +1,28 @@
+# grokery HTML brief
+
+Two ways to produce the shareable, self-contained, dark-mode brief
+(`--html` / "give me an HTML brief" / "export as html"). Both are offline:
+inline CSS, no JavaScript, system fonts, print-friendly.
+
+## A. Deterministic (CLI) — `render-html.sh`
+Write your synthesized markdown to a file, then convert:
+```bash
+bash skills/grokery/scripts/render-html.sh ~/Documents/Grokery/<slug>-synthesis.md \
+  --title 'grokery — <topic>' --out ~/Documents/Grokery/<slug>-brief.html
+```
+The script handles headings, bold/italic, links, autolinks (`<https://…>`),
+bullets, code, and `---` rules, and wraps them in the styled shell with a badge
+and the `✅ All sources reported back.` footer.
+
+## B. Hand-crafted (richer)
+For a higher-touch brief (callouts, a verdict hero, simple inline-SVG bars for
+engagement), copy [`brief-template.html`](brief-template.html) and fill the
+`<!-- CONTENT -->` slot. Keep it self-contained: no external fonts, scripts, or
+images. Reuse the template's CSS variables so it stays on-brand.
+
+## Conventions
+- **Filename:** `<slug>-brief.html` in the memory dir (default `~/Documents/Grokery/`).
+- **Top:** badge → topic `<h1>` → metadata line (window, date, sources that reported).
+- **Body:** the synthesis — themes as `<h2>`, citations as inline links, engagement numbers in `<strong>`.
+- **Footer:** a one-line colophon naming grokery and "ranked by engagement, not SEO", plus the sources that **actually returned** (e.g. "Sources: X, HN, GitHub, Polymarket"). Never claim a source that failed or was skipped.
+- Verify it opens offline (double-click) before declaring done.
diff --git a/skills/grokery/references/modes.md b/skills/grokery/references/modes.md
@@ -0,0 +1,41 @@
+# grokery modes & synthesis
+
+## Synthesis (always)
+After `grokery.sh` returns raw signal, write the brief:
+- **Lead with the verdict** — 2–3 sentences: the dominant take + the strongest counter-take.
+- **Group by theme**, not by platform. Each theme cites its strongest 1–2 sources with engagement numbers.
+- **Quote wit verbatim.** Surface genuinely funny / viral one-liners — they carry signal and are why people read these. (last30days calls this "humor scoring".)
+- **Numbers over adjectives** — "365K views · 117♥" beats "went viral".
+- **Cite everything** — every claim links to a concrete URL.
+- **Flag confidence** — call out when a take is one loud account vs broad consensus.
+
+## Research (default)
+Single topic. Run:
+```bash
+bash skills/grokery/scripts/grokery.sh "<topic>" --days 30
+```
+
+## Comparison — `A vs B [vs C]`
+Detect `vs` in the request. Run the orchestrator **once per side**, then synthesize a side-by-side:
+```bash
+bash skills/grokery/scripts/grokery.sh "OpenClaw" --save-suffix cmp --days 30
+bash skills/grokery/scripts/grokery.sh "Hermes"   --save-suffix cmp --days 30
+```
+Compare on the axes people actually argue about (architecture, reliability, price, momentum) and include live GitHub stars per side.
+
+## `--mode competitors`
+Pass `--mode competitors`; grok names the top ~3 peers in its `### Competitors`
+subsection. Then run a comparison pass on those names.
+
+## `--mode hiring` / `--github-user <handle>` (person mode)
+- `--mode hiring` adds a `### Hiring signals` subsection (open roles, careers-page shifts) — reveals where a person/company is focusing.
+- `--github-user steipete` switches GitHub to person mode: recent PRs/pushes + top repos. Combine with the topic for a full "what is this person up to" brief.
+
+## `eli5` (post-run)
+If the user says `eli5 on`, rewrite the existing synthesis in plain language —
+short sentences, no jargon, concrete analogies — without re-running sources.
+`eli5 off` restores the standard brief.
+
+## Persistence
+- Snapshots auto-save to `~/Documents/Grokery/<slug>-raw.md` (override `--save-dir`, disable `--no-save`).
+- `--store` appends the run to `grokery.sqlite3` for trend monitoring; diff successive `<slug>-raw.md` snapshots to see what changed since last time.
diff --git a/skills/grokery/references/ranking.md b/skills/grokery/references/ranking.md
@@ -0,0 +1,37 @@
+# grokery ranking & dedup
+
+The fetchers pre-sort within each source. Cross-source ranking and clustering is
+the agent's job — engagement is the currency, not SEO.
+
+## 1. Normalize engagement across platforms
+Different platforms, different scales. Judge by **reach within a platform's norms**, not raw counts:
+
+| Platform | Primary signal | "High signal" rule of thumb |
+|---|---|---|
+| X | likes + reposts + views | 4-figure+ likes, or high views relative to author size |
+| Reddit | upvotes + comments | 3-figure+ upvotes, or top comment of a big thread |
+| Hacker News | points + comments | 100+ pts; comment count ≈ controversy |
+| YouTube | views + recency | high views for the topic's niche |
+| GitHub | stars + push recency | stars signal adoption; recent pushes signal momentum |
+| Polymarket | $ volume (odds on event page) | high volume = real conviction; open the event link for the crowd's number |
+
+Weight by **independence**: 5 unrelated accounts > 1 account quoted 5 times.
+
+## 2. Dedup / cluster
+Merge the same story across sources into one cluster:
+- Same URL, or same event/claim described by multiple platforms → one bullet, list the corroborating sources.
+- Quote-reposts and reaction threads collapse into the original.
+- A claim appearing on X **and** HN **and** Reddit is *stronger* — say so; don't list it three times.
+
+## 3. Recency
+Within the `--days` window, prefer fresher items on fast-moving topics; for
+evergreen topics, let engagement dominate.
+
+## 4. Confidence labels
+- **Consensus** — corroborated across ≥3 independent sources/platforms.
+- **Contested** — strong takes on both sides; present both.
+- **Single loud voice** — one high-reach account; flag as not-yet-corroborated.
+
+## 5. Final order
+Lead with the highest-confidence, highest-engagement cluster. Bury low-signal
+chatter or drop it. A short, sharp brief beats an exhaustive dump.