Public-only multi-Git mirror + auto-start services + datasets-to-build queue

decision Sun Jun 28 2026 00:00:00 GMT+0000 (Coordinated Universal Time) backupmirroringauto-startdatasetsmaster-plan

Master plan — multi-Git mirror + auto-start + dataset queue

1. Multi-Git mirror

Goal: every public chirag127/* and chirag127/* repo gets pushed to N Git providers automatically. If one provider goes down or DMCAs us, the others stand.

Targets (locked 2026-06-28) — every managed (non-self-hosted) Git host. Account creation needed on each:

Scope of mirroring:

• ? Only repos under repos/own/* in the umbrella (originals we authored)
• ? NOT repos/frk/* (forks — upstream already exists elsewhere)
• ? NOT private content (per public-only rule)

Skipped (and why):

• Self-hosted Gitea / Forgejo / GitLab CE — we don't run servers (per cloud-dbs-as-caches + no-self-host bias)
• Radicle (P2P) — too niche, no web URL means no jsDelivr equivalent
• PikaCode — dead since ~2017

1a. API documentation + landing page

Goal: every API repo has a real docs page and the fleet has a landing page.

Build pattern: each API repo contains a _site/ (Pages output) with:

• index.html — landing for that API (Astro page or hand-rolled HTML)
• /codes/...json — the actual data files (existing layout)
• Same Pages deploy serves both — https://rto.oriz.in/ is HTML, https://rto.oriz.in/codes/MH-12.json is the API

Fleet-level landing lives in chirag127/api-fleet-landing ? deploys to api.oriz.in (or replaces the placeholder at oriz.in directly).

Stack: Astro static (matches framework-astro-react-tailwind-shadcn decision), with a single shared layout component that each API repo imports via npm or git submodule. Or keep it simple: pure HTML + Tailwind CDN per repo, no shared dep, easier to maintain than a shared package.

Implementation: GitHub Actions workflow in each repo that pushes to all 4 mirrors on every push to main. Use pixta-dev/repository-mirroring-action or hand-rolled git push --mirror per provider. Secrets (mirror PATs) stored as GH Actions org-level secrets.

Scope: PUBLIC repos only. Secrets stay in restic?Backblaze B2 + Bitwarden per existing rules. No private mirror repo.

2. Auto-start services

Goal: Hr (Headroom), hai, RTK hook, cavemem worker, plus anything else session-critical, start on user login without manual intervention. Survive reboot.

Mechanism: Windows Task Scheduler entries triggered at user logon. One entry per service. Each entry runs a small .cmd or directly the binary.

Services to wire:

So actually only Headroom + cavemem truly need Task Scheduler. RTK + MCP run inside the AI agent.

Implementation: a single script C:\D\oriz\scripts\install-auto-start.cmd that:

1. Registers a Scheduled Task Oriz-Headroom-Login running docker start headroom at user logon
2. Registers a Scheduled Task Oriz-Cavemem-Login running cavemem start at user logon
3. Both with "Run only when user is logged on" + "Run with highest privileges"

3. Dataset queue — build order

Verified gaps from 2026-06-28 audit (no free OSS API exists for any of these):

Build order rationale: Constants next (trivial, fastest validation of the rto-api pattern on a different dataset). Then ragas/dynasties/countries-plus in parallel if time allows.

Each ships with same pattern:

• One repo chirag127/<slug>-api
• jsDelivr URL immediately works after first push
• CF Pages subdomain <slug>.oriz.in wired after wrangler login (done as of 2026-06-28)
• Auto-mirrored to GitLab/Codeberg/Bitbucket per §1

4. Sequence to execute

1. Deploy rto.oriz.in to CF Pages now (wrangler is authed; 5-10 min)
2. Lock auto-start rule + write install-auto-start.cmd
3. Set up mirror automation — create org accounts on GitLab/Codeberg/Bitbucket, generate PATs, store as GH org secrets, write the reusable workflow
4. Ship Physics Constants API (next dataset)
5. Ship the remaining 3 datasets one by one
6. Backfill mirrors — once §3 works, run a one-time backfill to push every existing chirag127/* and chirag127/* public repo to all mirrors

Step 1 can happen this session. §2-§6 likely need multiple sessions; this plan is the durable spec.

5. Cost (verified 2x via web search)