Skip to content

SYSTEM Cited by 5 sources

Bitbucket Pipelines

Atlassian's hosted CI/CD product, integrated directly with Bitbucket. On the sysdesign-wiki, the relevant surface is its role as an automated quality gate that the Rovo Dev agent reads output from and iterates against.

(Source: sources/2026-04-24-atlassian-rovo-dev-driven-development)

Role in the agentic workflow

Per the source post, the Fireworks team runs this CI surface on every PR:

"CI pipeline as quality gate: Every PR runs lint, vet, tests, and Helm validation. The agent reads pipeline output and addresses failures before requesting review."

Canonicalised as agentic-pr-triage. Key properties that make Bitbucket Pipelines fit this role:

  • First-class agent read access. The Rovo Dev agent has first-party access to pipeline output "without leaving the conversation" — the agent doesn't have to scrape a UI or invoke a CLI; the output is available as agent context.
  • Scope: lint + vet + tests + Helm validation. Covers the correctness surface end-to-end: static analysis (lint, vet), the test suite, and deployment-config validation (Helm).
  • PR-scoped. Every PR gets validated — the automated gate is always between the agent's output and the human review surface.

Seen in

  • sources/2026-04-24-atlassian-rovo-dev-driven-development — the CI surface Rovo Dev reads and iterates against during the Fireworks build.
  • sources/2026-04-29-atlassian-inside-atlassians-merge-queues — the execution substrate for the merge-queue pipeline; a dedicated pre-merge pipeline defined via the merge-queues: section in bitbucket-pipelines.yml, orthogonal to the post-merge pipeline. Atlassian's Jira team composes this with three parallel parent-child pipelines so the merge-queue pipeline validates three product distributions in wall-clock rather than additively.

Role in Merge Queues (2026-04-29)

Bitbucket Pipelines is also the execution substrate underneath Bitbucket Merge Queues. Two load-bearing configuration surfaces:

  • merge-queues: section in bitbucket-pipelines.yml — lets teams run a different, faster validation suite pre-merge without changing the post-merge pipeline that builds release artefacts. The merge-queue pipeline and the post-merge pipeline are deliberately distinct; bundling them would turn queue latency into a ceiling on release-pipeline cost.
  • Parent-child pipelines for merge-queue parallelism — Jira's merge-queue pipeline fans out to three parallel parent-child pipelines (one per product distribution). Pattern canonicalised as parent-child-pipelines-for-ci-parallelism.

On the agent-side, Bitbucket Pipelines also functions as the CI surface Rovo Dev reads output from and addresses before requesting review (see agentic-pr-triage). Together these two roles make Pipelines a dual quality gate: a defensive checkpoint for human-authored merges via the merge queue, and a feedback channel for agent-authored PRs via direct agent integration.

Live-event delivery instance (2026-08-07)

Bitbucket Pipelines was the deployment stage in the live Rovo Dev → Bitbucket → Pipelines → production loop for the Atlassian Unleash PWA. A scheduled upstream infrastructure change temporarily paused this path mid-event; an incident-channel escalation restored deployment. The post does not disclose the pipeline configuration or the underlying incident mechanism. (Source: sources/2026-08-07-atlassian-building-a-real-time-pwa-on-atlassians-own-stack)

Agentic Pipelines instance (2026-08-28)

The vulnerability-remediation post markets Pipelines under the "Bitbucket Agentic Pipelines" framing: pipeline steps that define AI agents (via a definitions.agents: block pointing at a .rovodev/*.md prompt + pipeline-config.yml) and invoke Rovo Dev non-interactively, "triggered on a schedule, by an event, or programmatically." Each step declares its own OAuth-style scopes (read:repository, write:repository, write:pullrequest), giving each of the dispatcher/coding-agent/closer stages a least-privilege boundary. The pipeline definition, prompts, model/tool config, and skills all live in-repo and are reviewed like code — the "prompts are the system" principle. (Source: sources/2026-08-28-atlassian-agentic-automation-in-practice-putting-standard-engineering-work-on-autopilot)

Feature-flag-cleanup Agentic Pipeline instance (2026-09-24)

Bitbucket Pipelines is again the execution substrate for the coding agent in Atlassian's feature-flag-cleanup Dispatcher → Coding Agent → Closer loop. A scheduled Rovo Studio rule triggers one pipeline run per eligible stale-flag ticket (ticket key passed as context), fanning out in parallel. The excerpted bitbucket-pipelines.yml shows a definitions.agents.flag-cleanup block pointing at a .claude/flag-cleanup-agent.md prompt + .claude/pipeline-config.yml, and a single step whose auth.system.scopes declare exactly read:repository, write:repository, write:pullrequest — the same per-step least-privilege scoping as the vulnerability-remediation instance, plus branch restrictions preventing direct pushes to main. Checks must pass before the step opens a PR. (Source: sources/2026-09-24-atlassian-how-we-automated-feature-flag-cleanup-with-agentic-pipelines)

Last updated · 766 distilled / 2,225 read