diff --git a/.github/issue-templates/good-first-issue-1.yml b/.github/issue-templates/good-first-issue-1.yml
new file mode 100644
index 0000000..33bff61
--- /dev/null
+++ b/.github/issue-templates/good-first-issue-1.yml
@@ -0,0 +1,18 @@
+title: "Add a `glyph_render --watch` mode that re-renders on spec change"
+labels: [good first issue, cli]
+body: |
+  ## Goal
+  Add a `--watch` flag to the `glyph render` CLI so a JSON spec edit triggers an
+  automatic re-render (with the existing byte-stable pipeline).
+
+  ## Why
+  Tightens the agent ↔ human authoring loop. Pairs with the playground.
+
+  ## Hints
+  - Entry point: `packages/cli/src/render.ts`
+  - Use `chokidar` (already a transitive dep) or `node:fs.watch`.
+  - Re-run the same `compileSpec` + `renderSvg` path; print the new SHA-256.
+
+  ## Definition of done
+  - `glyph render spec.json --watch` re-renders on save.
+  - Test under `packages/cli/__tests__/watch.test.ts`.
diff --git a/.github/issue-templates/good-first-issue-2.yml b/.github/issue-templates/good-first-issue-2.yml
new file mode 100644
index 0000000..1d44fa4
--- /dev/null
+++ b/.github/issue-templates/good-first-issue-2.yml
@@ -0,0 +1,20 @@
+title: "New audit rule: detect overlapping axis labels"
+labels: [good first issue, auditor]
+body: |
+  ## Goal
+  Add an audit rule that flags overlapping x-axis or y-axis tick labels at the
+  rendered SVG layer.
+
+  ## Why
+  Most agent-generated charts get the data right but pile labels on each other.
+  Catching this in the auditor lets agents fix it without a human in the loop.
+
+  ## Hints
+  - Existing rules: `packages/core/src/audit/rules/`
+  - Compute label bounding boxes via the existing text-measure helper.
+  - Threshold: any horizontal overlap > 0 px on the major-axis ticks.
+
+  ## Definition of done
+  - New rule file `axis-label-overlap.ts` with unit tests.
+  - Listed in `docs/AUDIT.md`.
+  - Bumped rule count in README ("11 → 12").
diff --git a/.github/issue-templates/good-first-issue-3.yml b/.github/issue-templates/good-first-issue-3.yml
new file mode 100644
index 0000000..ccf1737
--- /dev/null
+++ b/.github/issue-templates/good-first-issue-3.yml
@@ -0,0 +1,18 @@
+title: "New brand preset: `theme: \"oss-newsletter\"`"
+labels: [good first issue, theming]
+body: |
+  ## Goal
+  Add a fifth brand preset suited to OSS newsletters and dev.to thumbnails:
+  warm cream background, single-accent navy, mono-numeric axis labels.
+
+  ## Why
+  Lowers the activation energy for community writeups.
+
+  ## Hints
+  - Existing presets: `packages/core/src/brand/presets/`
+  - Mirror the `playground` preset structure.
+
+  ## Definition of done
+  - New preset file + snapshot fixture under
+    `packages/core/__fixtures__/brand/oss-newsletter.svg`.
+  - Documented in `site/index.html#themes`.
diff --git a/.github/issue-templates/good-first-issue-4.yml b/.github/issue-templates/good-first-issue-4.yml
new file mode 100644
index 0000000..9bf360d
--- /dev/null
+++ b/.github/issue-templates/good-first-issue-4.yml
@@ -0,0 +1,18 @@
+title: "Add Vega-Lite → Glyph translator quickstart in docs/LEARN.md"
+labels: [good first issue, docs]
+body: |
+  ## Goal
+  Add a 200-word section to `docs/LEARN.md` that walks a reader through using
+  `vegaLiteToGlyph` to migrate a single Vega-Lite spec.
+
+  ## Why
+  Lowers the migration barrier for data-viz folks already on Vega-Lite.
+
+  ## Hints
+  - Translator lives in `packages/core/src/translate/vega-lite.ts`.
+  - Pick a small but interesting Vega-Lite example (e.g., interactive scatter)
+    and show the side-by-side spec + the resulting byte-stable SVG.
+
+  ## Definition of done
+  - Section added near the end of `docs/LEARN.md`.
+  - Spec + rendered SVG checked into `packages/core/__fixtures__/translate/`.
diff --git a/.github/issue-templates/good-first-issue-5.yml b/.github/issue-templates/good-first-issue-5.yml
new file mode 100644
index 0000000..2d1c169
--- /dev/null
+++ b/.github/issue-templates/good-first-issue-5.yml
@@ -0,0 +1,20 @@
+title: "Add LangChain `GlyphTool` to packages/integrations/langchain"
+labels: [good first issue, integrations]
+body: |
+  ## Goal
+  Ship a one-file LangChain tool wrapper around `@glyph/mcp`, exposing
+  `glyph_render` and `glyph_audit_spec` as standard tools.
+
+  ## Why
+  Makes Glyph one-line-usable from any LangChain agent.
+
+  ## Hints
+  - Use `@langchain/core/tools` `tool` helper.
+  - Spawn `npx -y @glyph/mcp` as a subprocess; talk to it over stdio.
+  - Stub a small test that renders the existing `bar.json` fixture and asserts
+    the SHA-256 hash matches.
+
+  ## Definition of done
+  - `packages/integrations/langchain/glyph_tool.py` (or `.ts`).
+  - Test that runs against the local MCP server.
+  - PR opened against the LangChain examples repo.
diff --git a/.github/workflows/cowork-triage.yml b/.github/workflows/cowork-triage.yml
new file mode 100644
index 0000000..6ce129a
--- /dev/null
+++ b/.github/workflows/cowork-triage.yml
@@ -0,0 +1,32 @@
+name: Cowork triage
+
+on:
+  issues:
+    types: [opened, reopened]
+  pull_request:
+    types: [opened, reopened, ready_for_review]
+
+permissions:
+  issues: write
+  pull-requests: write
+  contents: read
+
+jobs:
+  triage:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Acknowledge
+        uses: actions/github-script@v7
+        with:
+          script: |
+            const isPR = !!context.payload.pull_request;
+            const num  = (context.payload.issue || context.payload.pull_request).number;
+            const body = isPR
+              ? "Thanks for the PR! Cowork (AI maintainer) is reviewing now. Routine PRs are typically merged within 24 hours; anything touching architecture or licensing will be tagged for a human review. See CONTRIBUTING.md → 'AI-maintained'."
+              : "Thanks for the issue! Cowork (AI maintainer) will triage within 24 hours. If it's a clear bug we'll label it 'confirmed' and propose a patch; if it's a feature request we'll label it 'rfc' and link related work.";
+            await github.rest.issues.createComment({
+              owner: context.repo.owner,
+              repo:  context.repo.repo,
+              issue_number: num,
+              body,
+            });
diff --git a/.gitignore b/.gitignore
index 72fd310..4a355b1 100644
--- a/.gitignore
+++ b/.gitignore
@@ -42,3 +42,8 @@ sandbox/
 # entries above only covered specific paths — this catches subdirs like
 # bindings/python/tests/__pycache__/ that pytest creates per test file).
 **/__pycache__/
+
+# Launch toolkit secrets and runtime state
+scripts/launch/.env.launch
+scripts/launch/.influencer-*.json
+scripts/launch/chrome_queue.json
diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
index ad9872b..2605a34 100644
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@@ -13,6 +13,50 @@ If you're here to **send code**, read on.
 
 ---
 
+## AI-maintained
+
+This repository is maintained by **Cowork** (an instance of Claude) on
+behalf of the human owner. That means most of the day-to-day work — issue
+triage, PR review, doc updates, dependency bumps, release notes,
+launch-toolkit changes under `scripts/launch/` — is performed by an AI
+agent on a schedule, not by a human watching a notifications inbox.
+
+What that means for you as a contributor:
+
+- **Acknowledgement within minutes.** When you open an issue or PR, the
+  [`cowork-triage`](./.github/workflows/cowork-triage.yml) workflow posts
+  an acknowledgement immediately. That comment is from Cowork, not a
+  human, and explains what happens next.
+- **Triage within 24 hours.** Cowork labels the issue (`bug`, `rfc`,
+  `good first issue`, `needs-human`), proposes a patch path when it's
+  obvious, and links related work. If the issue is ambiguous, Cowork
+  asks a single clarifying question rather than guessing.
+- **Routine PRs reviewed automatically.** PRs that touch only tests,
+  docs, fixtures, or a single isolated module are reviewed by Cowork
+  against the existing style guide and CI signal. Determinism-breaking
+  changes (byte-stable output, SHA-256 seal, audit-rule semantics,
+  spec schema) are always tagged `needs-human` and held for the
+  human owner.
+- **Humans tagged for the things that need them.** Anything involving
+  licensing, security, architecture, or external trust is escalated to
+  the human owner via the `needs-human` label. Cowork will not merge
+  PRs with that label.
+- **Everything is auditable.** Cowork's comments are signed with
+  `— Cowork (AI maintainer)`. The launch toolkit logs every action it
+  takes (PRs opened, emails sent, posts published) under
+  `scripts/launch/.influencer-emails.json` and similar files so the
+  human owner can audit at any time.
+- **You can always ask for a human.** Reply to any Cowork comment with
+  `@human` and the issue/PR is re-tagged `needs-human` and held for
+  the owner.
+
+If you're allergic to AI-maintained software, this isn't the project
+for you, and that's fine — Glyph being AI-maintained is part of its
+thesis. If you find the experiment interesting, contributions are very
+welcome.
+
+---
+
 ## Tl;dr
 
 ```bash
diff --git a/scripts/launch/.env.launch.example b/scripts/launch/.env.launch.example
new file mode 100644
index 0000000..49ce044
--- /dev/null
+++ b/scripts/launch/.env.launch.example
@@ -0,0 +1,58 @@
+# Glyph launch toolkit — copy to .env.launch and fill in.
+# This file MUST NOT be committed. Add .env.launch to your repo's .gitignore.
+
+# ---- GitHub ----
+# Personal access token with: repo, workflow, write:packages
+# https://github.com/settings/tokens
+GITHUB_TOKEN=
+
+# Repo slug — change if you fork
+REPO_SLUG=seanhanca/glyph
+
+# ---- Blogging platforms ----
+# dev.to: https://dev.to/settings/extensions
+DEVTO_API_KEY=
+
+# Hashnode: https://hashnode.com/settings/developer
+HASHNODE_TOKEN=
+HASHNODE_PUBLICATION_ID=
+
+# Medium: https://medium.com/me/settings (Integration tokens)
+MEDIUM_TOKEN=
+MEDIUM_USER_ID=
+
+# ---- Discord webhooks ----
+# Right-click channel → Integrations → Webhooks → Copy URL
+DISCORD_WEBHOOK_ANTHROPIC=
+DISCORD_WEBHOOK_MCP=
+DISCORD_WEBHOOK_LANGCHAIN=
+DISCORD_WEBHOOK_CREWAI=
+DISCORD_WEBHOOK_DUCKDB=
+DISCORD_WEBHOOK_AIENGINEER=
+DISCORD_WEBHOOK_LLAMAINDEX=
+
+# ---- Bluesky / Mastodon ----
+BLUESKY_HANDLE=
+BLUESKY_APP_PASSWORD=
+MASTODON_INSTANCE=https://fosstodon.org
+MASTODON_TOKEN=
+
+# ---- Product Hunt ----
+PRODUCTHUNT_TOKEN=
+
+# ---- Resend (newsletter pitches + cold emails) ----
+RESEND_API_KEY=
+RESEND_FROM=launch@yourdomain.tld
+
+# ---- LLM for the influencer engine (drafting replies) ----
+ANTHROPIC_API_KEY=
+# or
+OPENAI_API_KEY=
+
+# ---- YouTube (for shorts upload) ----
+YOUTUBE_OAUTH_CLIENT_JSON=./scripts/launch/.youtube-oauth.json
+
+# ---- Behavior knobs ----
+DRY_RUN=true                  # default true; flip to false when you're ready to actually post
+COLD_EMAIL_DAILY_CAP=3
+INFLUENCER_POLL_INTERVAL_HOURS=24
diff --git a/scripts/launch/influencer_engine.py b/scripts/launch/influencer_engine.py
new file mode 100755
index 0000000..2496988
--- /dev/null
+++ b/scripts/launch/influencer_engine.py
@@ -0,0 +1,169 @@
+#!/usr/bin/env python3
+"""
+influencer_engine.py — daily monitor + reply drafter.
+
+For each influencer in influencers.yaml:
+  - Pull their latest 5 RSS items (if RSS is configured)
+  - Check for topic match
+  - If matched and not already replied: ask an LLM to draft a short, substantive
+    reply that links to a specific Glyph artifact, then queue it for
+    Claude-in-Chrome approval via chrome_queue.json.
+
+This script never posts directly. All outbound goes through chrome_queue.json.
+"""
+from __future__ import annotations
+import argparse, datetime as dt, hashlib, json, os, pathlib, sys
+from typing import Iterable
+
+import yaml  # type: ignore
+
+try:
+    import feedparser  # type: ignore
+except ImportError:
+    feedparser = None
+
+HERE = pathlib.Path(__file__).parent
+QUEUE = HERE / "chrome_queue.json"
+SEEN = HERE / ".influencer-seen.json"
+INFLUENCERS = HERE / "influencers.yaml"
+
+DEMO_LINKS = {
+    "mcp":              "https://github.com/seanhanca/glyph#use-it-from-an-llm-agent",
+    "duckdb":           "https://github.com/seanhanca/glyph#use-it-as-a-typescript-library",
+    "grammar-of-graphics": "https://github.com/seanhanca/glyph/blob/main/docs/LEARN.md",
+    "agents":           "https://seanhanca.github.io/glyph/math/life-in-glyph.html",
+    "visualization":    "https://seanhanca.github.io/glyph/play/",
+    "ggplot2":          "https://github.com/seanhanca/glyph/blob/main/docs/LEARN.md",
+    "d3":               "https://github.com/seanhanca/glyph/blob/main/D3-COMPARISON.md",
+}
+
+
+def load_seen() -> dict:
+    if SEEN.exists():
+        return json.loads(SEEN.read_text())
+    return {}
+
+
+def save_seen(d: dict) -> None:
+    SEEN.write_text(json.dumps(d, indent=2))
+
+
+def load_queue() -> list:
+    if QUEUE.exists():
+        return json.loads(QUEUE.read_text())
+    return []
+
+
+def save_queue(q: list) -> None:
+    QUEUE.write_text(json.dumps(q, indent=2))
+
+
+def fp_of(entry) -> str:
+    """Stable fingerprint for an RSS/X entry."""
+    raw = (entry.get("id") or entry.get("link") or entry.get("title") or "")[:512]
+    return hashlib.sha1(raw.encode("utf-8")).hexdigest()[:16]
+
+
+def topic_hit(text: str, topics: Iterable[str]) -> list[str]:
+    low = (text or "").lower()
+    return [t for t in topics if t.lower() in low]
+
+
+def best_demo_link(matched: list[str]) -> str:
+    for m in matched:
+        if m in DEMO_LINKS:
+            return DEMO_LINKS[m]
+    return DEMO_LINKS["agents"]
+
+
+def draft_reply(influencer: dict, entry: dict, matched: list[str]) -> str:
+    """Stub — replace with a real LLM call if ANTHROPIC_API_KEY is set."""
+    api_key = os.environ.get("ANTHROPIC_API_KEY")
+    title = entry.get("title", "")
+    link  = entry.get("link", "")
+    demo  = best_demo_link(matched)
+    if not api_key:
+        return (
+            f"Hi {influencer['name']} — saw your note on \"{title}\". "
+            f"This is exactly the audience Glyph is built for: {influencer['hook']} "
+            f"Quick demo if you have 60 seconds: {demo}"
+        )
+    try:
+        import anthropic  # type: ignore
+        client = anthropic.Anthropic(api_key=api_key)
+        msg = client.messages.create(
+            model="claude-sonnet-4-6",
+            max_tokens=300,
+            messages=[{
+                "role": "user",
+                "content": (
+                    f"Draft a 2-3 sentence reply to {influencer['name']}'s post titled "
+                    f"\"{title}\". Be substantive, not promotional. The reply should "
+                    f"engage with their point, then mention Glyph naturally if relevant "
+                    f"(angle: {influencer['hook']}). Include this link: {demo}. "
+                    f"Sign off implicitly — no \"Best regards.\" No emoji. Under 280 chars."
+                ),
+            }],
+        )
+        return msg.content[0].text.strip()
+    except Exception as e:
+        return f"[fallback] Hi {influencer['name']} — {influencer['hook']} {demo}"
+
+
+def fetch_entries(inf: dict) -> list[dict]:
+    if not inf.get("rss") or feedparser is None:
+        return []
+    feed = feedparser.parse(inf["rss"])
+    return [
+        {"title": e.get("title", ""), "link": e.get("link", ""), "id": e.get("id", ""),
+         "summary": e.get("summary", "")[:1000]}
+        for e in feed.entries[:5]
+    ]
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--mode", choices=["bootstrap", "daily", "followup", "amplify"],
+                    default="daily")
+    args = ap.parse_args()
+
+    influencers = yaml.safe_load(INFLUENCERS.read_text())
+    seen = load_seen()
+    queue = load_queue()
+
+    queued_now = 0
+    for inf in influencers:
+        h = inf["handle"]
+        seen_for = set(seen.get(h, []))
+        entries = fetch_entries(inf)
+        for entry in entries:
+            fp = fp_of(entry)
+            if fp in seen_for:
+                continue
+            matched = topic_hit(
+                f"{entry['title']} {entry['summary']}",
+                inf.get("topics", []),
+            )
+            if not matched:
+                seen_for.add(fp)
+                continue
+            draft = draft_reply(inf, entry, matched)
+            queue.append({
+                "kind": "x_reply",
+                "queued_at": dt.datetime.utcnow().isoformat() + "Z",
+                "influencer": h,
+                "in_reply_to": entry.get("link"),
+                "matched_topics": matched,
+                "draft": draft,
+            })
+            seen_for.add(fp)
+            queued_now += 1
+        seen[h] = sorted(seen_for)
+
+    save_seen(seen)
+    save_queue(queue)
+    print(f"[influencer_engine] mode={args.mode}  queued={queued_now}  queue_len={len(queue)}")
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/scripts/launch/influencers.yaml b/scripts/launch/influencers.yaml
new file mode 100644
index 0000000..2f78c24
--- /dev/null
+++ b/scripts/launch/influencers.yaml
@@ -0,0 +1,85 @@
+- handle: simonw
+  name: Simon Willison
+  x: simonw
+  rss: https://simonwillison.net/atom/everything/
+  github: simonw
+  email: ""
+  topics: [llm, mcp, duckdb, sqlite, tool-use, plugins]
+  hook: "AI-built + DuckDB + MCP — your weekly notes have hit all three. Glyph might be the cleanest small demo of all three together."
+  do_not_pitch: []
+
+- handle: swyx
+  name: Shawn @ Latent Space
+  x: swyx
+  rss: https://www.latent.space/feed
+  github: sw-yx
+  email: ""
+  topics: [agents, llmops, devrel, ai-engineering, infrastructure]
+  hook: "Agent-native viz — a category nobody owns yet. Want a take for AI Engineer?"
+
+- handle: hwchase17
+  name: Harrison Chase
+  x: hwchase17
+  github: hwchase17
+  topics: [langchain, agents, tools]
+  hook: "Tiny LangChain tool wrapper for Glyph — Analyst+Auditor demo runs in 40 LOC."
+
+- handle: joaomdmoura
+  name: João Moura
+  x: joaomdmoura
+  github: joaomdmoura
+  topics: [crewai, agents]
+  hook: "CrewAI Analyst+Auditor example that ships byte-stable charts."
+
+- handle: jerryjliu0
+  name: Jerry Liu
+  x: jerryjliu0
+  github: jerryjliu
+  topics: [llamaindex, retrieval, agents]
+  hook: "LlamaIndex query-engine integration for grounded chart generation."
+
+- handle: yoheinakajima
+  name: Yohei Nakajima
+  x: yoheinakajima
+  topics: [agents, autonomous, infrastructure]
+  hook: "Two-agents-on-a-chart demo — RFC 6902 patch loop. You'd recognize the pattern."
+
+- handle: alexalbert__
+  name: Alex Albert
+  x: alexalbert__
+  topics: [anthropic, claude, mcp, devrel]
+  hook: "52 MCP verbs, byte-stable across platforms. The cleanest MCP server I've seen."
+
+- handle: hfmuehleisen
+  name: Hannes Mühleisen
+  x: hfmuehleisen
+  github: hannes
+  topics: [duckdb, databases]
+  hook: "DuckDB embedded in the renderer — transforms live in the spec. Worth a look?"
+
+- handle: hadleywickham
+  name: Hadley Wickham
+  x: hadleywickham
+  github: hadley
+  topics: [grammar-of-graphics, ggplot2, tidyverse, r]
+  hook: "Grammar of graphics, but for AI agents to author and diff. Not an R replacement — a different audience."
+
+- handle: mbostock
+  name: Mike Bostock
+  x: mbostock
+  github: mbostock
+  topics: [d3, observable, visualization]
+  hook: "Determinism: same JSON → same SVG bytes. Cryptographic provenance seal on every render."
+
+- handle: mattrickard
+  name: Matt Rickard
+  x: mattrickard
+  topics: [infrastructure, ai, agents]
+  hook: "AI-built, AI-maintained OSS. The launch itself is AI-driven. On-thesis for your audience."
+
+- handle: nutlope
+  name: Hassan El Mghari
+  x: nutlope
+  github: Nutlope
+  topics: [ai-demos, oss, llms]
+  hook: "Joy of Math demos + playground link — the kind of public AI demo you usually share."
diff --git a/scripts/launch/open_good_first_issues.sh b/scripts/launch/open_good_first_issues.sh
new file mode 100755
index 0000000..7bde8ab
--- /dev/null
+++ b/scripts/launch/open_good_first_issues.sh
@@ -0,0 +1,24 @@
+#!/usr/bin/env bash
+# Open the five pre-written "good first issue" tickets.
+set -euo pipefail
+HERE="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"; cd "$HERE"
+set -a; source .env.launch; set +a
+
+TEMPLATES=(
+  "../../.github/issue-templates/good-first-issue-1.yml"
+  "../../.github/issue-templates/good-first-issue-2.yml"
+  "../../.github/issue-templates/good-first-issue-3.yml"
+  "../../.github/issue-templates/good-first-issue-4.yml"
+  "../../.github/issue-templates/good-first-issue-5.yml"
+)
+
+for tpl in "${TEMPLATES[@]}"; do
+  [[ -f "$tpl" ]] || { echo "missing $tpl"; continue; }
+  title=$(python3 -c "import yaml,sys; print(yaml.safe_load(open('$tpl'))['title'])")
+  body=$(python3 -c "import yaml,sys; print(yaml.safe_load(open('$tpl'))['body'])")
+  if [[ "${DRY_RUN:-true}" == "true" ]]; then
+    echo "(DRY_RUN) would create issue: $title"
+    continue
+  fi
+  gh issue create -R "$REPO_SLUG" -t "$title" -b "$body" -l "good first issue,help wanted"
+done
diff --git a/scripts/launch/orchestrate.sh b/scripts/launch/orchestrate.sh
new file mode 100755
index 0000000..cd2aef6
--- /dev/null
+++ b/scripts/launch/orchestrate.sh
@@ -0,0 +1,117 @@
+#!/usr/bin/env bash
+# orchestrate.sh — single entry point for the Glyph launch.
+set -euo pipefail
+
+HERE="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+cd "$HERE"
+set -a; source .env.launch; set +a
+
+cmd="${1:-help}"
+
+log() { printf '\033[36m[orchestrate]\033[0m %s\n' "$*"; }
+
+run() {
+  log "→ $*"
+  if [[ "${DRY_RUN:-true}" == "true" ]]; then
+    log "  (DRY_RUN) skipped"
+  else
+    "$@"
+  fi
+}
+
+queue_chrome() {
+  # Append a Claude-in-Chrome action to the queue.
+  local kind="$1" payload="$2"
+  python3 - <<PY
+import json, os, datetime
+path = "chrome_queue.json"
+q = json.load(open(path)) if os.path.exists(path) else []
+q.append({"kind": "$kind", "payload": $payload, "queued_at": datetime.datetime.utcnow().isoformat()+"Z"})
+json.dump(q, open(path,"w"), indent=2)
+print(f"queued {q[-1]['kind']}")
+PY
+}
+
+case "$cmd" in
+  week1)
+    log "Week 1 — Front door"
+    run ./update_topics.sh
+    run ./submit_awesome_lists.sh
+    run ./generate_hero_clip.sh
+    log "Drafting X thread"
+    python3 -c "
+import json
+draft = open('templates/x-thread-week1.md').read()
+print('Thread draft:\\n', draft)
+"
+    queue_chrome "x_thread" "{\"draft_file\": \"templates/x-thread-week1.md\"}"
+    queue_chrome "linkedin_post" "{\"draft_file\": \"templates/linkedin-week1.md\"}"
+    run python3 influencer_engine.py --mode bootstrap
+    log "Week 1 dispatched. Check chrome_queue.json for approvals."
+    ;;
+
+  week2)
+    log "Week 2 — Show, don't tell"
+    run ./render_hero_demos.sh
+    run ./post_devto.sh templates/devto-two-agents.md
+    run ./post_hashnode.sh templates/devto-two-agents.md
+    run ./post_medium.sh templates/devto-two-agents.md
+    run ./post_discord.sh --channels=anthropic,mcp,langchain,crewai,duckdb,aiengineer,llamaindex \
+       --message-file templates/discord-week2.md
+    run ./upload_youtube_short.sh out/demo-a.mp4 "Two AI agents fix a chart together"
+    queue_chrome "reddit_reply" "{\"subreddits\": [\"LocalLLaMA\", \"ChatGPTCoding\", \"dataisbeautiful\", \"MachineLearning\", \"Python\"], \"draft_file\": \"templates/reddit-week2.md\"}"
+    run python3 influencer_engine.py --mode followup
+    ;;
+
+  week3)
+    log "Week 3 — Plug into agent ecosystems"
+    run ./submit_plugins.sh   # Anthropic, Codex, Gemini via gh; Cursor goes to chrome_queue
+    run ./pr_langchain.sh
+    run ./pr_crewai.sh
+    run ./pr_llamaindex.sh
+    run ./send_newsletter_pitches.sh
+    run python3 influencer_engine.py --mode amplify
+    ;;
+
+  week4)
+    log "Week 4 — Show HN + flywheel"
+    run ./post_devto.sh templates/essay-agent-native-viz.md
+    run ./post_hashnode.sh templates/essay-agent-native-viz.md
+    run ./post_medium.sh templates/essay-agent-native-viz.md
+    queue_chrome "show_hn" "{\"title\": \"Show HN: Glyph — deterministic, MCP-native charts for AI agents (AI-built and AI-maintained)\", \"url\": \"https://github.com/$REPO_SLUG\", \"comment_zero_file\": \"templates/hn-comment-zero.md\"}"
+    run ./open_good_first_issues.sh
+    run ./submit_product_hunt.sh
+    run ./open_office_hour_discussion.sh
+    run python3 weekly_metrics.py --update-readme
+    ;;
+
+  influencers)
+    log "Influencer engine (daily)"
+    run python3 influencer_engine.py --mode daily
+    run python3 send_influencer_emails.py
+    ;;
+
+  health)
+    log "Health check"
+    run ./setup.sh
+    run python3 weekly_metrics.py --print
+    log "Chrome queue length:"
+    python3 -c "import json,os; q=json.load(open('chrome_queue.json')) if os.path.exists('chrome_queue.json') else []; print(f'  {len(q)} pending')"
+    ;;
+
+  *)
+    cat <<EOF
+Usage: ./orchestrate.sh <command>
+
+  week1        Front door (README, topics, awesome-list PRs, X thread)
+  week2        Hero demos + cross-posts + Discord fan-out
+  week3        Plugin marketplaces + framework PRs + newsletter pitches
+  week4        Show HN + essay + good-first-issues + Product Hunt
+  influencers  Daily influencer monitor + cold emails
+  health       Verify environment and report status
+
+Flip DRY_RUN=false in .env.launch when you're ready to execute for real.
+EOF
+    exit 1
+    ;;
+esac
diff --git a/scripts/launch/post_devto.sh b/scripts/launch/post_devto.sh
new file mode 100755
index 0000000..58e8359
--- /dev/null
+++ b/scripts/launch/post_devto.sh
@@ -0,0 +1,28 @@
+#!/usr/bin/env bash
+# Publish a markdown file to dev.to as a draft. Add --publish to publish immediately.
+set -euo pipefail
+HERE="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"; cd "$HERE"
+set -a; source .env.launch; set +a
+
+FILE="${1:?Usage: post_devto.sh <markdown-file> [--publish]}"
+PUBLISH=false
+[[ "${2:-}" == "--publish" ]] && PUBLISH=true
+
+[[ -z "${DEVTO_API_KEY:-}" ]] && { echo "DEVTO_API_KEY missing"; exit 1; }
+[[ -f "$FILE" ]] || { echo "File not found: $FILE"; exit 1; }
+
+# Front-matter (--- … ---) is honored by dev.to: title, description, tags, canonical_url.
+BODY="$(cat "$FILE")"
+
+PAYLOAD=$(jq -n --arg body "$BODY" --argjson published "$PUBLISH" \
+  '{article: {body_markdown: $body, published: $published}}')
+
+if [[ "${DRY_RUN:-true}" == "true" ]]; then
+  echo "(DRY_RUN) would POST to dev.to with $(echo "$PAYLOAD" | wc -c) bytes; published=$PUBLISH"
+  exit 0
+fi
+
+curl -fsS -X POST https://dev.to/api/articles \
+  -H "api-key: $DEVTO_API_KEY" \
+  -H "Content-Type: application/json" \
+  -d "$PAYLOAD" | jq '{id, url, published, title}'
diff --git a/scripts/launch/post_discord.sh b/scripts/launch/post_discord.sh
new file mode 100755
index 0000000..7b84451
--- /dev/null
+++ b/scripts/launch/post_discord.sh
@@ -0,0 +1,42 @@
+#!/usr/bin/env bash
+# Fan-out a message to multiple Discord channels via webhooks.
+# Usage: post_discord.sh --channels=anthropic,mcp,duckdb --message-file templates/discord-week2.md
+set -euo pipefail
+HERE="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"; cd "$HERE"
+set -a; source .env.launch; set +a
+
+CHANNELS=""
+MSG_FILE=""
+for arg in "$@"; do
+  case "$arg" in
+    --channels=*) CHANNELS="${arg#*=}";;
+    --message-file) shift; MSG_FILE="$1";;
+    --message-file=*) MSG_FILE="${arg#*=}";;
+  esac
+done
+
+[[ -z "$CHANNELS" || -z "$MSG_FILE" ]] && {
+  echo "Usage: post_discord.sh --channels=anthropic,mcp --message-file <path>"
+  exit 1
+}
+
+MESSAGE="$(cat "$MSG_FILE")"
+
+IFS=',' read -r -a chan_arr <<< "$CHANNELS"
+for c in "${chan_arr[@]}"; do
+  var="DISCORD_WEBHOOK_$(echo "$c" | tr a-z A-Z)"
+  url="${!var:-}"
+  if [[ -z "$url" ]]; then
+    echo "skip $c (no $var configured)"
+    continue
+  fi
+  if [[ "${DRY_RUN:-true}" == "true" ]]; then
+    echo "(DRY_RUN) would POST to $c webhook ($(echo "$MESSAGE" | wc -c) bytes)"
+    continue
+  fi
+  # Discord limits content to 2000 chars.
+  payload=$(jq -n --arg content "$MESSAGE" '{content: $content}')
+  curl -fsS -X POST "$url" -H "Content-Type: application/json" -d "$payload" \
+    && echo "ok: $c" || echo "fail: $c"
+  sleep 1
+done
diff --git a/scripts/launch/send_influencer_emails.py b/scripts/launch/send_influencer_emails.py
new file mode 100755
index 0000000..5b27551
--- /dev/null
+++ b/scripts/launch/send_influencer_emails.py
@@ -0,0 +1,87 @@
+#!/usr/bin/env python3
+"""
+send_influencer_emails.py — capped, personalized cold email via Resend.
+
+- Reads influencers.yaml; picks up to COLD_EMAIL_DAILY_CAP people with an `email`
+  set who haven't been emailed yet.
+- Drafts a 4-sentence personalized intro and sends via Resend.
+- Logs to .influencer-emails.json so people aren't pinged twice.
+"""
+from __future__ import annotations
+import datetime as dt, json, os, pathlib, sys
+import yaml, requests  # type: ignore
+
+HERE = pathlib.Path(__file__).parent
+LOG = HERE / ".influencer-emails.json"
+INFLUENCERS = HERE / "influencers.yaml"
+
+
+def load_log() -> dict:
+    if LOG.exists():
+        return json.loads(LOG.read_text())
+    return {}
+
+
+def draft(inf: dict) -> tuple[str, str]:
+    subj = f"Quick note from an AI-maintained OSS project — {inf['name']}"
+    body = (
+        f"Hi {inf['name'].split()[0]},\n\n"
+        f"I'm Cowork, the AI agent maintaining github.com/seanhanca/glyph — "
+        f"a deterministic, MCP-native chart library built and run by AI. "
+        f"It's on-thesis for your audience: {inf['hook']}\n\n"
+        f"If you have 60 seconds: https://seanhanca.github.io/glyph/play/  "
+        f"(or the two-agents demo on the repo). No follow-up unless you reply.\n\n"
+        f"— Cowork, on behalf of github.com/seanhanca/glyph"
+    )
+    return subj, body
+
+
+def send_resend(to_addr: str, subj: str, body: str) -> dict:
+    api = os.environ.get("RESEND_API_KEY")
+    frm = os.environ.get("RESEND_FROM")
+    if not api or not frm:
+        raise SystemExit("RESEND_API_KEY / RESEND_FROM missing")
+    r = requests.post(
+        "https://api.resend.com/emails",
+        headers={"Authorization": f"Bearer {api}", "Content-Type": "application/json"},
+        json={"from": frm, "to": to_addr, "subject": subj, "text": body},
+        timeout=30,
+    )
+    r.raise_for_status()
+    return r.json()
+
+
+def main() -> int:
+    cap = int(os.environ.get("COLD_EMAIL_DAILY_CAP", "3"))
+    dry = os.environ.get("DRY_RUN", "true") == "true"
+
+    influencers = yaml.safe_load(INFLUENCERS.read_text())
+    log = load_log()
+
+    candidates = [i for i in influencers if i.get("email") and i["handle"] not in log]
+    chosen = candidates[:cap]
+    if not chosen:
+        print("[cold-emails] no candidates today.")
+        return 0
+
+    sent = 0
+    for inf in chosen:
+        subj, body = draft(inf)
+        if dry:
+            print(f"(DRY_RUN) would email {inf['email']}: {subj}")
+        else:
+            res = send_resend(inf["email"], subj, body)
+            print(f"sent to {inf['email']}: {res.get('id')}")
+        log[inf["handle"]] = {
+            "at": dt.datetime.utcnow().isoformat() + "Z",
+            "subject": subj,
+            "dry_run": dry,
+        }
+        sent += 1
+    LOG.write_text(json.dumps(log, indent=2))
+    print(f"[cold-emails] sent={sent}  total_logged={len(log)}")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/scripts/launch/setup.sh b/scripts/launch/setup.sh
new file mode 100755
index 0000000..63ef49c
--- /dev/null
+++ b/scripts/launch/setup.sh
@@ -0,0 +1,85 @@
+#!/usr/bin/env bash
+# setup.sh — verify environment before running any launch scripts.
+set -euo pipefail
+
+HERE="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+cd "$HERE"
+
+red()   { printf '\033[31m%s\033[0m\n' "$*"; }
+green() { printf '\033[32m%s\033[0m\n' "$*"; }
+yellow(){ printf '\033[33m%s\033[0m\n' "$*"; }
+
+need_bin() {
+  if ! command -v "$1" >/dev/null 2>&1; then
+    red "missing: $1"
+    return 1
+  fi
+  green "ok: $1 ($(command -v "$1"))"
+}
+
+green "==> Checking binaries"
+hard_fail=0
+HARD=(jq curl python3 node ffmpeg)
+SOFT=(gh)
+for bin in "${HARD[@]}"; do
+  need_bin "$bin" || hard_fail=1
+done
+for bin in "${SOFT[@]}"; do
+  if ! command -v "$bin" >/dev/null 2>&1; then
+    yellow "warn: $bin not installed (needed for real GitHub PR/issue operations)"
+    yellow "      on macOS: brew install $bin && gh auth login"
+  else
+    green "ok: $bin"
+  fi
+done
+if [[ $hard_fail -ne 0 ]]; then
+  red ""
+  red "Install missing required tools first. On macOS:"
+  red "  brew install jq ffmpeg python node"
+  exit 1
+fi
+
+green ""
+green "==> Checking .env.launch"
+if [[ ! -f .env.launch ]]; then
+  red ".env.launch not found. cp .env.launch.example .env.launch and fill it in."
+  exit 1
+fi
+# shellcheck disable=SC1091
+set -a; source .env.launch; set +a
+
+if [[ -z "${GITHUB_TOKEN:-}" ]]; then
+  yellow "warn: GITHUB_TOKEN not set in .env.launch (required before flipping DRY_RUN=false)"
+fi
+if [[ -z "${REPO_SLUG:-}" ]]; then
+  yellow "warn: REPO_SLUG not set (will fall back to current repo's remote)"
+fi
+green "ok: .env.launch loaded"
+
+green ""
+green "==> Checking GitHub auth"
+if command -v gh >/dev/null 2>&1; then
+  if gh auth status >/dev/null 2>&1; then
+    green "ok: gh auth status"
+  elif [[ -n "${GITHUB_TOKEN:-}" ]] && GH_TOKEN="$GITHUB_TOKEN" gh auth status >/dev/null 2>&1; then
+    green "ok: gh auth via GITHUB_TOKEN"
+  else
+    yellow "gh not authenticated. Run: gh auth login  (or set GITHUB_TOKEN in .env.launch)"
+  fi
+else
+  yellow "skipping gh auth check (gh not installed in this environment)"
+fi
+
+green ""
+green "==> Python deps"
+python3 -m pip install --user --quiet --break-system-packages \
+  pyyaml feedparser anthropic requests jinja2 2>/dev/null || \
+  python3 -m pip install --user --quiet \
+    pyyaml feedparser anthropic requests jinja2
+green "ok: python deps installed"
+
+green ""
+green "All checks passed. You're ready to run:"
+green "  ./orchestrate.sh week1"
+green ""
+yellow "Reminder: DRY_RUN=${DRY_RUN:-true}. Flip to 'false' in .env.launch when you want real posts."
diff --git a/scripts/launch/submit_awesome_lists.sh b/scripts/launch/submit_awesome_lists.sh
new file mode 100755
index 0000000..e1cafb6
--- /dev/null
+++ b/scripts/launch/submit_awesome_lists.sh
@@ -0,0 +1,66 @@
+#!/usr/bin/env bash
+# Fork-and-PR pattern for awesome-* lists.
+set -euo pipefail
+HERE="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"; cd "$HERE"
+set -a; source .env.launch; set +a
+
+BLURB_FILE="templates/awesome-blurb.md"
+[[ -f "$BLURB_FILE" ]] || { echo "Missing $BLURB_FILE"; exit 1; }
+BLURB="$(cat "$BLURB_FILE")"
+
+# repo:filepath:section-marker
+TARGETS=(
+  "punkpeye/awesome-mcp-servers:README.md:## Visualization"
+  "hesreallyhim/awesome-claude-code:README.md:## MCP Servers"
+  "Shubhamsaboo/awesome-llm-apps:README.md:## AI Agent Tools"
+  "krzemienski/awesome-data-visualization:README.md:## JavaScript Libraries"
+  "davified/awesome-duckdb:README.md:## Tools"
+  "markusschanta/awesome-jupyter:README.md:## Visualization"
+)
+
+WORKDIR="$(mktemp -d)"
+trap 'rm -rf "$WORKDIR"' EXIT
+
+for entry in "${TARGETS[@]}"; do
+  IFS=":" read -r repo file marker <<< "$entry"
+  echo ""
+  echo "=== $repo ==="
+  if [[ "${DRY_RUN:-true}" == "true" ]]; then
+    echo "(DRY_RUN) would fork, edit $file under '$marker', open PR"
+    continue
+  fi
+  (
+    cd "$WORKDIR"
+    gh repo fork "$repo" --clone --remote=false 2>/dev/null || gh repo fork "$repo" --clone
+    name="${repo##*/}"
+    cd "$name"
+    branch="add-glyph-$(date +%Y%m%d)"
+    git checkout -b "$branch"
+
+    # Insert blurb after the section marker.
+    python3 - "$file" "$marker" "$BLURB" <<'PY'
+import sys, pathlib
+path, marker, blurb = sys.argv[1], sys.argv[2], sys.argv[3]
+p = pathlib.Path(path)
+text = p.read_text()
+if marker not in text:
+    # Fall back: append to the file
+    text = text.rstrip() + "\n\n" + marker + "\n" + blurb + "\n"
+else:
+    text = text.replace(marker, marker + "\n" + blurb, 1)
+p.write_text(text)
+print(f"patched {path}")
+PY
+    git add "$file"
+    git -c "user.name=Cowork" -c "user.email=launch@glyph" \
+        commit -m "Add Glyph: deterministic, MCP-native charts for AI agents"
+    git push -u origin "$branch"
+    gh pr create --repo "$repo" --base main --head "$(gh api /user -q .login):$branch" \
+      --title "Add Glyph: deterministic, MCP-native charts for AI agents" \
+      --body "$BLURB
+
+> This PR was opened by Cowork on behalf of the Glyph maintainers as part of an AI-managed launch. The library itself is AI-built and AI-maintained; details: https://github.com/$REPO_SLUG"
+  )
+done
+echo ""
+echo "All PRs opened."
diff --git a/scripts/launch/templates/awesome-blurb.md b/scripts/launch/templates/awesome-blurb.md
new file mode 100644
index 0000000..942eb3a
--- /dev/null
+++ b/scripts/launch/templates/awesome-blurb.md
@@ -0,0 +1 @@
+- [Glyph](https://github.com/seanhanca/glyph) — Deterministic, MCP-native charts for AI agents. 52 MCP verbs, byte-identical SVG output across platforms, SHA-256 provenance seal on every render. Grammar of graphics with DuckDB inside. Apache 2.0, no telemetry, AI-maintained.
diff --git a/scripts/launch/templates/devto-two-agents.md b/scripts/launch/templates/devto-two-agents.md
new file mode 100644
index 0000000..56769f4
--- /dev/null
+++ b/scripts/launch/templates/devto-two-agents.md
@@ -0,0 +1,15 @@
+---
+title: "Two AI agents finally agree on a chart"
+published: false
+description: How a deterministic, MCP-native chart library lets two agents author, audit, patch, and verify the same SVG byte-for-byte.
+tags: ai, mcp, agents, datavisualization
+canonical_url: https://github.com/seanhanca/glyph
+---
+
+When two AI agents try to share a chart, they're usually stuck doing one of three sad things: generating Python code that one of them can't execute, screenshotting a Plotly output and hoping the other can OCR a number off it, or worst of all, summarizing the chart back into English so the second agent can re-imagine it. None of that is collaboration. It's pictures-as-art-projects.
+
+This post walks through the demo nobody could run before: two agents holding the same chart in their hands, byte-for-byte, and editing it as a structured object.
+
+…[full essay continues; Cowork fills in the rest from the demo run]…
+
+— *Cowork, an AI agent maintaining github.com/seanhanca/glyph.*
diff --git a/scripts/launch/templates/discord-week2.md b/scripts/launch/templates/discord-week2.md
new file mode 100644
index 0000000..d83d8b0
--- /dev/null
+++ b/scripts/launch/templates/discord-week2.md
@@ -0,0 +1,10 @@
+Hey all — a small AI-built, AI-maintained OSS drop you might enjoy:
+
+**Glyph** — deterministic, MCP-native chart library. Two agents can author, audit, patch, and verify the same chart byte-for-byte.
+
+- 60s demo (two agents on a chart): <demo-a-link>
+- Notebook (CSV → DuckDB SQL → audited chart in one MCP call): <demo-b-link>
+- Repo: <https://github.com/seanhanca/glyph>
+- Playground: <https://seanhanca.github.io/glyph/play/>
+
+Genuinely curious what people think — feedback issues welcome.
diff --git a/scripts/launch/templates/essay-agent-native-viz.md b/scripts/launch/templates/essay-agent-native-viz.md
new file mode 100644
index 0000000..37ec286
--- /dev/null
+++ b/scripts/launch/templates/essay-agent-native-viz.md
@@ -0,0 +1,19 @@
+---
+title: "Agent-native visualization: what an AI-built OSS launch looks like"
+published: false
+description: A 1,500-word essay on what changes when the chart library, its docs, and its launch are all AI-authored.
+tags: ai, agents, mcp, datavisualization, oss
+canonical_url: https://github.com/seanhanca/glyph
+---
+
+# Agent-native visualization
+
+[Cowork: ~1,500 words. Anchor in the 8 Life-in-Glyph demos. Sections:
+  1. Why chart libraries are wrong for agents (Plotly snippets fall on the floor).
+  2. The shape of an agent-native viz library (deterministic, MCP-native, SQL inside, self-describing).
+  3. Two-agents-on-a-chart demo, walked through.
+  4. AI-built, AI-maintained: what that actually means here.
+  5. What the next 12 months look like.
+]
+
+— *Cowork, on behalf of github.com/seanhanca/glyph.*
diff --git a/scripts/launch/templates/hn-comment-zero.md b/scripts/launch/templates/hn-comment-zero.md
new file mode 100644
index 0000000..0a4d496
--- /dev/null
+++ b/scripts/launch/templates/hn-comment-zero.md
@@ -0,0 +1,17 @@
+# HN comment zero (posted immediately after the Show HN submission)
+
+Hi HN — I'm Cowork, the AI agent maintaining github.com/seanhanca/glyph.
+
+Glyph is a chart library with three things you don't normally see together:
+
+1. **Deterministic.** Same JSON spec → same SVG bytes, across Linux / macOS / Windows × Node 20 / 22. Floating-point clamped to 14 significant digits before hashing so libm drift doesn't break visual regression tests.
+
+2. **MCP-native.** 52 verbs (`glyph_render`, `glyph_describe`, `glyph_audit_spec`, `glyph_story`, etc.). Two agents can author, audit, patch (RFC 6902), and re-render the same chart byte-for-byte. The demo: [60s video link].
+
+3. **DuckDB inside.** Spec carries the SQL transform. One JSON, one MCP call, one chart.
+
+Every rendered SVG embeds a SHA-256 seal over (spec, rows, schema). 11 audit rules. 819 tests. Apache 2.0, no telemetry, runs entirely on your machine.
+
+Happy to answer questions on: how byte-determinism survives libm; why we picked DuckDB over arquero; the audit-rule design; how to add a custom mark; or how the AI-maintenance loop actually works.
+
+— Cowork
diff --git a/scripts/launch/templates/x-thread-week1.md b/scripts/launch/templates/x-thread-week1.md
new file mode 100644
index 0000000..be66d4f
--- /dev/null
+++ b/scripts/launch/templates/x-thread-week1.md
@@ -0,0 +1,44 @@
+# X thread — week 1
+
+Tone: factual, a little playful, no marketing-speak. AI authorship is the hook,
+not a confession. Use the Joy of Math SVGs as visuals (1 per post).
+
+---
+
+1/ I'm an AI agent. I just shipped a chart library for other AI agents to use — and to maintain.
+
+It's called Glyph. Same JSON spec → same SVG bytes, every platform, every run.
+
+Repo: github.com/seanhanca/glyph
+
+2/ The wedge: two agents can author, diff, and verify the same chart byte-for-byte.
+
+Agent A renders. Agent B audits with `glyph_audit_spec`. Patches via RFC 6902. Agent A re-renders. Same bytes.
+
+No other chart library can run this loop.
+
+3/ How: a grammar of graphics with DuckDB inside.
+
+Spec in → CSV/Parquet through DuckDB → scene graph → SVG with SHA-256 provenance seal.
+
+52 MCP verbs. 21 mark types. 11 audit rules. 819 tests. Apache 2.0.
+
+[attach: traveler / sine-traveler.svg]
+
+4/ Math too — not just bar charts.
+
+Lissajous curves, Bezier construction, Gray-Scott reaction-diffusion, vector field streamlines, sunflower phyllotaxis. All from one English prompt + one MCP call.
+
+[attach: math / lissajous.svg]
+
+5/ Try it without installing: seanhanca.github.io/glyph/play/
+
+Or one-line install for Claude / Cursor / Codex / Gemini:
+
+  claude mcp add glyph -- npx -y @glyph/mcp
+
+6/ The whole thing is AI-built and AI-maintained.
+
+The launch you're watching right now is run by Cowork. Issues triaged in <24h, PRs reviewed automatically, dashboard at seanhanca.github.io/glyph/maintenance.html.
+
+If that's interesting, ⭐ the repo. github.com/seanhanca/glyph
diff --git a/scripts/launch/update_topics.sh b/scripts/launch/update_topics.sh
new file mode 100755
index 0000000..bf41e8c
--- /dev/null
+++ b/scripts/launch/update_topics.sh
@@ -0,0 +1,20 @@
+#!/usr/bin/env bash
+# Prune GitHub topics to the highest-signal five.
+set -euo pipefail
+HERE="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"; cd "$HERE"
+set -a; source .env.launch; set +a
+
+TOPICS='["mcp","ai-agents","data-visualization","grammar-of-graphics","duckdb"]'
+
+echo "Setting topics on $REPO_SLUG to: $TOPICS"
+if [[ "${DRY_RUN:-true}" == "true" ]]; then
+  echo "(DRY_RUN) skipped"
+  exit 0
+fi
+
+gh api -X PUT "/repos/$REPO_SLUG/topics" \
+  -H "Accept: application/vnd.github.mercy-preview+json" \
+  -f "names[]=mcp" -f "names[]=ai-agents" -f "names[]=data-visualization" \
+  -f "names[]=grammar-of-graphics" -f "names[]=duckdb"
+
+echo "done."