npx skills add https://github.com/getcolors/once --skill package-once-blueSKILL.md
ONCE with Blue
Use this skill in the user's current directory. Read references/configuration.md before creating or changing desired state and before a real create or delete, and references/github-deploy.md before writing or changing a GitHub Actions workflow.
Safety
- Never request or print a secret, private key, token, password, or application value.
- Secrets use
COLORS_PAR_*, the one namespace every colour shares, and never belong incolors.yml, generated files, commands, or logs. - Read only an approved public
.pubfile; never inspect a private key. - Do not overwrite
blueorcolors.ymlwithout explicit approval. - Default to
buildandcreate --dry-run; require explicit confirmation for real create/delete. - Build and dry-run do not validate credentials.
- Delete is blocked until
COLORS_PAR_COMPUTE_PREVENT_DESTROY=falseor itsCOLORS_PAR_*alias is set. - Never edit
.colors/. It is shared with Green and Red; never run implementations concurrently against it.
Initialize
Gather profile, applications, providers, selected providers' non-secret
settings. Deploy keys are not among them: ONCE generates one per repository
named under github on every create and publishes it to a GitHub environment
named after the profile, so ask only for COLORS_PAR_GITHUB_TOKEN. Never
gather secret values. When an application names a repository, also ask whether
the user wants continuous deployment — see below.
After confirmation:
- Copy this skill's bundled
blueto./blueand make it executable. The PEP 723 metadata must pin bothpackage-once-blueandblueto immutable Git commits. - Write
colors.ymlfrom the reference withworkdir: .colors. - Add
.colors/and private environment files to.gitignorewithout replacing unrelated entries. - Run
./blue buildand./blue create --dry-run. - Report required variable names only and do not provision automatically.
Continuous deployment
create publishes SSH_PRIVATE_KEY, SERVER_IP, SERVER_USER, and
SSH_KNOWN_HOSTS into an Actions environment named after the profile; nothing
reads them until a workflow does. So whenever an application names github,
ask whether to also add the workflow that builds the image and pings the server
on every push to main. Do not ask when no application names a repository —
without it nothing is published.
On yes, adapt references/github-deploy.md:
substitute the profile for PROFILE and the application host for the
environment URL. The workflow belongs in the application's repository, often
not the directory holding colors.yml. Compare the origin remote to
owner/repo: on a match write .github/workflows/deploy.yml, asking before
overwriting an existing one; otherwise show the YAML and name its path rather
than writing into the wrong repository. State that the repository needs its own
Dockerfile, that the published image must equal the application's image, and
that the first run works only after a real create.
On no, nothing further is needed: ssh -T deploy@SERVER_IP </dev/null is the
whole deploy interface, and auto_update: true updates without a ping.
Operate
./blue build
./blue create --dry-run
./blue create
./blue describe
./blue delete --dry-run
./blue delete
Use -f|--file for another desired-state file. Check credential presence only
before a real operation and let the launcher perform final validation.