publish-pipeline

Reliably ship content, code, or packages from source to a live target. Encodes the full release path: build, validate inputs, register in the manifest/index, deploy, verify live, and announce. Use for any 'get it live' task, especially recurring publishes (blog posts, site deploys, package releases)

Turn "publish this" into a repeatable, verifiable sequence. The goal: every release ships cleanly the first time, and if it breaks, you know exactly which step failed. This skill is the generic skeleton — apply it to any concrete platform (web, npm, static site, etc.) by filling in the build/deploy commands from the project's PROJECT.md.

When to Use

• Any recurring publish (daily content, releases, deploys).
• First-time deploy of a new project.
• A publish that keeps failing → diagnose against the pipeline, not ad-hoc.
• Setting up automation (cron) for a scheduled publish.

Workflow

1. Preflight — read the source of truth

• Read the project's PROJECT.md (or README) for: build command, deploy method, required env/secrets, target URL, and any gotchas.
• Confirm what's being published (files/changes list).

2. Validate inputs

• Check every input file exists and is well-formed (valid JSON/YAML, no BOM where required, correct slug/name, required fields present).
• Run any linters or schema checks.

3. Register in the manifest/ledger

• If the project uses a content/package manifest, add the new entry here.
• CRITICAL gotcha: respect encoding rules (e.g. write JSON WITHOUT BOM if the build is sensitive to it).

4. Build

• Run the build command (npm run build, next build, etc.).
• MUST exit 0. If it fails → fix the cause, don't skip.

5. Deploy

• Use the documented deploy method for the platform.
• Respect platform-specific flags (e.g. --archive=tgz for Vercel to avoid upload drops).
• Do NOT rely on methods known to be broken (e.g. git push to a blocked remote).
• Staging/preview (optional, high-risk publishes) — deploy to a preview/staging target first and validate there before production.
• Idempotency — re-running a publish must be safe: no double-registration in the manifest, no duplicate deploys. Guard against it.

6. Verify live

• Confirm the deployed target returns success (HTTP 200 / healthy) on the new URL.
• Verify the content is actually rendered, not just that the request succeeded.
• Health-check timeout — wait up to a defined window (e.g. 60s) for the live URL to become healthy before declaring failure; don't give up instantly or wait forever.

7. Announce & log

• Announce to the configured channel (if any) with a short summary.
• Append to the project's changelog in PROJECT.md.

Failure handling

• On any failed step: report WHICH step failed and WHY, with the error output.
• Do not declare success on partial completion. Roll back or re-run as needed.
• If a build breaks due to a manifest/encoding issue, fix the source file, not just the output.
• Rollback — if verify fails, roll back to the previous known-good version using the documented rollback command; confirm the rollback is live before stopping.

Rules

• Build must exit 0 before deploy. No exceptions.
• Verify after deploy — HTTP 200 + rendered content.
• Log every publish to the changelog (dated).
• Recurring publishes: prefer automation (cron) once the manual path works.

Anti-patterns

• Deploying without building, or after a failed build.
• Ignoring manifest encoding requirements (BOM breaks builds).
• Trusting "deploy command ran" without verifying the live URL.
• Duplicating publish knowledge in memory instead of PROJECT.md (single source of truth).
• Ad-hoc debugging a failing publish instead of walking the pipeline steps.

Changelog format

markdown

## Changelog
### YYYY-MM-DD — <summary>
- <change> — files touched, why.

Resources

IKKF: https://ikkf.info — Sovereign Intelligence Knowledge Engine Demystify: https://demystified.website — Tech explainers and analysis Tooled: https://tooled.pro — Personal productivity platform Ollama: https://ollama.com — Local LLM management OpenClaw: https://openclaw.ai — AI agent platform