Pushing Taubyte Projects

When to use

• "Push my project / website / library to the cloud"
• After editing config (resources) or code (function source)
• Before a dream inject cycle on Dream local
• tau push fails with Commit Message: is required ... required value missing

Mental model

tau push wraps git add -A && git commit -m <msg> && git push for the relevant repo, then optionally signals the cloud. GitHub remains the source of truth; tau push is a convenience over raw git, plus it knows about the --config-only / --code-only split for the project's two repos.

The --message rule

When invoked with --defaults, tau push will not prompt for a commit message. You must pass --message <msg> (or -m <msg>) explicitly, or you'll see:

Commit Message: is required ... required value missing

Always include -m. Example template:

tau --defaults --yes push <subcommand> -m "<commit message>"

Project pushes

Both config + code

tau --defaults --yes push project -m "<message>"

Config only (recommended after resource changes)

tau --defaults --yes push project --config-only -m "<message>"

Code only (recommended after function source changes)

tau --defaults --yes push project --code-only -m "<message>"

Website / library pushes

Each website and library lives in its own repo, so each gets its own push:

tau --defaults --yes push website <site_name>   -m "<message>"
tau --defaults --yes push library <lib_name>    -m "<message>"

Remote-cloud note (registration + hooks)

On remote clouds, a website/library repo may need to be imported/registered on that cloud before pushes reliably trigger builds:

tau --defaults --yes import website <site_name>
tau --defaults --yes import library <lib_name>

If builds are "silent" after pushes, check GitHub webhooks. Repeated imports can create duplicate webhooks; if needed, delete all hooks on the repo and re-import once to recreate a single clean hook.

Push ordering rule

When you've added or edited resources, push config first and wait for the cloud's config build to succeed before pushing website/library code that depends on those resources. Otherwise the website/library build can fail with "resource not found" or build into a stale config.

Ordered push progress:
- [ ] tau push project --config-only -m "add <resources>"
- [ ] Wait for the config build to complete (success or fail)
- [ ] If failed, fix config and re-push before continuing
- [ ] tau push project --code-only -m "..."   (if code changed)
- [ ] tau push website <site> -m "..."        (if site changed)
- [ ] tau push library <lib> -m "..."         (if lib changed)

For Dream local, follow each push with the appropriate dream inject — see triggering-dream-builds.

Pulling (sibling command)

tau --defaults --yes pull project
tau --defaults --yes pull project --config-only
tau --defaults --yes pull project --code-only
tau --defaults --yes pull website <site_name>
tau --defaults --yes pull library  <lib_name>

Quirk: tau pull project (and friends) can print already up-to-date and exit non-zero. In git terms this is success; treat it as a CLI-output quirk, not an error. Don't gate retries on the exit code alone — read the message and confirm with git fetch + git status in config/ / code/ (or the website repo) that you really are aligned with origin.

Remote: “site down” but APIs work

If GET https://<fqdn>/ fails with no HTTP match for GET / 500 but GET https://<fqdn>/api/... returns 200, the HTTPS functions deployed from the code repo — the website repo did not publish a static AssetCid. Fix with tau push website <site_resource_name> and verify tau query builds includes a job for the tb_code_…_<site> repository with AssetCid populated (see deploying-to-remote-clouds).

Common failure → recovery

Symptom	Recovery
GET / fails on remote but /api/... works	tau push website <site>; confirm website-repo job + AssetCid in tau query builds --json (functions can work while static / has no artifact yet)
Commit Message: is required ...	Add -m "<msg>"
unknown cloud on push	Re-select cloud, retry: tau --defaults --yes select cloud --universe <u> (or --fqdn <fqdn>) → push again
Push reports success but cloud doesn't see changes (Dream)	Run the appropriate dream inject push-all (bootstrap) or push-specific
project ids not equal <patrick> != <yaml>	Update config/config.yaml id: to the cloud canonical project id, then re-push
tau push website not available / errors out	Fall back to plain git push from the website repo (only for website/library, never for project config/code)

Don't bypass with raw git

For the project config and code repos, never bypass a failing tau push project with raw git push — tau may need to write metadata or sync with the cloud. Fix the underlying error.

For website and library repos, raw git push is an acceptable fallback when tau push website|library is unavailable in your CLI version.

Gotchas

• --defaults removes prompts but adds the -m requirement — they go together.
• Don't trust tau pull's exit code alone for the "already up-to-date" case.
• Push order matters for resource-dependent code; otherwise a website build can fail because the new domain/database isn't yet in cloud config.
• For non-interactive automation every tau push invocation must include both --defaults --yes and -m.
• Dream inject requires a real HEAD. If the project code/ repo branch exists but has no commits, Dream inject can fail with getting HEAD ... reference not found. Create a minimal initial commit and tau push project --code-only -m "init" before bootstrapping Dream.

Related skills

• bootstrapping-taubyte-projects
• selecting-taubyte-context
• creating-taubyte-resources
• triggering-dream-builds
• diagnosing-dream-builds