Buildline for developers

You own the technical design, the plan, the gates and the merge.

#Daily loop

/buildline                       # what's in flight, what's next
/buildline-plan 012              # turn an approved spec into tasks
/buildline-build 012             # one task at a time: test first, gates, commit
/buildline-verify 012            # independent check in a clean context
/buildline-review 012            # findings + PR body
/buildline-ship 012              # PR, changelog, update context

Or unattended, at autonomy L3+:

npx buildline run 012              # opens Claude Code with a /goal condition
npx buildline run 012 --headless   # claude -p, for CI or a background terminal
npx buildline run 012 --print      # just print the /goal line

#Gates are the contract

Agents believe whatever the gates say, so make them honest:

• Keep them fast enough to run after every task (put slow suites in an optional gate).
• Make sure test actually fails when behaviour breaks.
• Add an e2e gate (Playwright, Cypress) as soon as you have UI acceptance scenarios.

npx buildline doctor warns when no gates are configured.

#How build sessions stay on track

Every build session starts the same way: git status and git log, progress.md, tasks, spec, then gates on the untouched baseline. It ends with a progress.md entry. That's what lets a fresh session (or a teammate's agent) resume mid-feature without re-explaining anything.

Commits are tagged [012/T3], so git log --grep 012/ shows a feature's history.

#Parallel work

Tasks marked "parallel": true can go to separate sessions. In Claude Code, use worktrees (claude --worktree) or background agents, one task each, and merge in order.

#CI

--ci adds .github/workflows/buildline.yml:

• gates on every PR,
• review: Claude reviews the PR against its factory spec (needs ANTHROPIC_API_KEY or CLAUDE_CODE_OAUTH_TOKEN secret; skipped otherwise),
• @claude mentions in PR comments.

#Other agents

Codex and Copilot read AGENTS.md; Cursor gets .cursor/rules/buildline.mdc; Gemini CLI gets a GEMINI.md import. They follow the same playbooks in .buildline/stations/. Claude-only extras: skills, subagents, the acceptance-lock hook, /goal.

Edit this page on GitHub