Skip to content

LAUNCH_CHECKLIST.md

Latest commit

 

History

History
138 lines (113 loc) · 6.33 KB

LAUNCH_CHECKLIST.md

File metadata and controls

138 lines (113 loc) · 6.33 KB
Raw

ghostloop v1.0 launch checklist

The work is done. The artifacts below are ready; what remains is the credentialed steps (uploading, submitting, posting). Each item links to the file with full instructions.

Order matters — earlier items unblock later ones.

Day 1 — distribution foundations (now automated)

PyPI Trusted Publishing + HuggingFace deploy are wired through GitHub Actions. Once the one-time setup below is done, every release happens by pushing a git tag.

  • One-time PyPI Trusted Publishing setup (Joe, ~5 min)

    1. Visit https://pypi.org/manage/account/publishing/ → "Add a new pending publisher".
    2. Project name: ghostloop. Owner: joemunene-by. Repo: ghostloop. Workflow: publish-pypi.yml. Environment name: pypi.
    3. Save. PyPI is now ready to accept OIDC pushes from .github/workflows/publish-pypi.yml.
    4. (You said you've already wired this — verify the form fields match the workflow file exactly. Mismatched workflow filename or env name is the most common cause of the OIDC handshake failing.)

  • One-time HF token paste (Joe, ~2 min)

    1. Generate a Write-access token at https://huggingface.co/settings/tokens.
    2. In the ghostloop repo on GitHub: Settings → Secrets and variables → Actions → New repository secret. Name: HF_TOKEN. Value: paste.
    3. (Optional) Add a repo variable HF_USERNAME if your HF account isn't joemunene-by. Same screen, "Variables" tab.

  • Publish v1.0.0 to PyPI (one command)

    git tag -a v1.0.0 -m "v1.0.0 — production release"
git push origin v1.0.0

    The publish-pypi.yml workflow runs: version-tag-match check, full test suite, sdist + wheel build, twine check, OIDC-authenticated PyPI upload. Status visible at github.com/joemunene-by/ghostloop/actions. (We already pushed the v1.0.0 tag earlier this session — that tag predates the workflow file; for the first publish, push a temporary v1.0.1 tag with the same artifacts after fixing any version mismatch, OR delete + re-tag v1.0.0 after merging this commit.)

  • Publish the HuggingFace Space (no command needed) The first push to main that touches spaces/ghostloop-demo/** (or any change to the workflow file itself) triggers publish-hf-space.yml. The workflow creates the Space if it doesn't exist and uploads the directory. After this commit lands, the Space is live at huggingface.co/spaces/Ghostgim/ghostloop-demo within ~3 minutes (HF build time on the free tier).

Day 2 — the demo

  • Record the demo video (marketing/DEMO_VIDEO_SCRIPT.md) Five-cut storyboard, two pre-written prompts, shot list, posting format guide per platform. Goal: 60 seconds vertical for X / LinkedIn, 16:9 horizontal for the README hero + YouTube.

    Setup needed: ghostloop installed locally with [mcp,mujoco], Claude Desktop wired to the MCP server, MuJoCo viewer with a Franka model loaded. Recording tools: QuickTime / OBS / Cleanshot.

Day 3 — visibility

  • Submit MCP marketplace listings (marketing/MARKETPLACE_LISTINGS.md) Eight channels in priority order:

    1. PyPI (already done by Day 1)
    2. HuggingFace Space (already done by Day 1)
    3. modelcontextprotocol.io community list — one PR
    4. Smithery.ai — needs smithery.yaml at repo root
    5. Awesome-list PRs (5 each):
      • awesome-mcp-servers
      • awesome-robotics-libraries
      • awesome-llm-tooluse
      • awesome-embodied-ai
    6. Tweet the demo video tagging @AnthropicAI, @deepmind, @MuJoCoSim
    7. LinkedIn post — long form, lead with the missing-middle thesis
    8. Show HN post — wait until video has 10k views to maximise lift

  • Publish the launch blog post (blog/2026-05-10_v1_launch.md) ~1500 words, ready to paste into a Substack / personal blog / Complex Developers site. Includes the eleven-release recap, the novel-vs-standard breakdown, the honest "what was hardest" section.

Week 2 — the deep play

  • Submit the arXiv preprint (paper/SUBMIT_TO_ARXIV.md) LaTeX + bibliography written, 6-8 pages, MIT contribution positioned correctly vs LangChain / AutoGen / MCP / OpenVLA / π0. Need:

    1. arXiv endorsement for cs.RO (one researcher's email + 2-line summary). The OWASP CheatSheetSeries credit + the GhostLM / ghostloop bodies of work make this easy.
    2. ORCID (60 seconds at orcid.org).
    3. Build the PDF (Overleaf or local pdflatex) and submit at arxiv.org/submit.

    Once the arXiv URL exists, update README + Space description + blog post + LinkedIn post to reference it.

Week 3+ — sustained reputation

  • Run a real OpenVLA reproduction through VLABenchmarkSuite. ghostloop's bench harness already ships pre-populated with published baselines (pick_place_widowx, manipulation_bridge, reach_target). Running an actual OpenVLA-7B checkpoint, recording the violation-rate gap under your safety pipeline, publishing the numbers as a follow-up arXiv submission — that's a top-tier conference paper.

  • Weekly publishing rhythm — one post per release / milestone. The work is the bottleneck no longer; the narration is.

What lands where

Audience Artifact URL
ML researchers arXiv preprint arxiv.org/abs/2605.NNNNN
Python users PyPI listing pypi.org/project/ghostloop
Curious clickers HuggingFace Space huggingface.co/spaces/Ghostgim/ghostloop-demo
MCP power users Smithery.ai listing smithery.ai/server/ghostloop
robotics generalists Awesome-list entries five PRs
Twitter/X demo video platform-native
LinkedIn demo video + long-form post platform-native
HN crowd Show HN post (week 2) news.ycombinator.com
Personal brand launch blog post personal site

What I'm NOT doing here

  • The actual recording. I can't operate a camera; you do that.
  • The actual submissions. I don't have your PyPI / HF / arXiv credentials and you wouldn't want me to.
  • The actual posting. Voice + cadence is yours.
  • The OpenVLA reproduction. That requires GPU time and a real OpenVLA checkpoint — separate project, separate sprint.

Everything gating those steps — drafts, scripts, manifests, configurations, tests, validation — is shipped.