SYSTEM Cited by 5 sources

Rovo Dev¶

Rovo Dev is Atlassian's AI development agent — a coding-agent surface that integrates directly with Atlassian's developer products (Bitbucket, Bitbucket Pipelines, Jira). It is the agent that "built Fireworks in four weeks, entirely by LLMs" per the source post.

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

What it does¶

Rovo Dev is used in the canonical agentic coding loop shape: an agent that writes, tests, deploys, and iterates on code in a real environment. The source post emphasises two distinguishing properties:

Product-integrated end-to-end SDLC coverage. "It needs to be involved in all parts of the SDLC, not just code generation. [...] Raising PRs, spawning independent agents for self-review, addressing feedback, reading pipeline output, updating tickets."
Bitbucket + Pipelines built in. "The all-in-one access to Atlassian products is genuinely great. Having Bitbucket and Pipelines integration in the agent has been a game changer. The agent can raise PRs, read diffs, and monitor builds without leaving the conversation. It makes it a seriously compelling daily driver."

Features surfaced in the source¶

Skills — "Skills are useful for specific domains or common actions within your repo. Internally we've built lots of skills! Skills for PRs, using CLI, specific domains like Raft, gRPC." Skills are the Rovo Dev unit of codified workflow knowledge. See agent-orchestration-skill.
Meta-workflow / orchestration skills — a broader skill type that "gives the agent a set of 'golden path' loops for how to work on [a codebase] end-to-end." Not a single-tool binding; a multi-step runbook. Canonicalised as agent-orchestration-meta-skill.
Sub-agents / personas — "for review, have an adversarial persona subagent that spins up and reviews what the main agent has written." See adversarial-review-persona.
Prompt shortcuts — e.g. !review-pr fires an independent sub-agent with an adversarial review prompt.
PR-bot comments — the source post mentions "comments on PR from Rovo Dev PR bot" — Rovo Dev has a first-party reviewer role on the PR itself, not just the in-agent conversation.

Typical workflow¶

Per the source post, the developer's day-to-day pattern:

Three parallel git workspaces, one agent per workspace on its own branch (three-workspace-parallel-agent-workflow).
Each agent writes e2e tests, deploys to a dev shard, iterates on failures (ai-writes-own-e2e-tests).
Before human review, the developer runs !review-pr to spin up an adversarial sub-agent (patterns/specialized-agent-decomposition).
The agent reads Bitbucket Pipelines output and addresses issues before human review is requested (agentic-pr-triage).
The human focuses on "architecture, design intent, risk" rather than detail.

Seen in¶

sources/2026-04-24-atlassian-rovo-dev-driven-development — the production agent used to build Fireworks; full SDLC coverage with Bitbucket + Pipelines integration.
sources/2026-06-01-atlassian-how-we-cut-up-to-80-of-engineering-chores-using-ai-agents-in — Rovo Dev (likely; not explicitly named) as the KTLO-automation agent surface for the Jira repo. The source documents Atlassian's "AI agents in Jira" setup applied to KTLO chores (stale-flag cleanup, flaky-test triage / fix). Operational model: the Jira work item is the agent's structured prompt; a status transition triggers the agent with a custom system prompt; the agent dispatches a per-codebase or per-failure-category specialist skill via a fallback chain (agent-skill-with-fallback-chain, test-category-classifier-then-specialist-skill); the agent acts as first-pass investigator producing a draft PR for human review. Reported throughput: 500+ merged stale-flag-cleanup PRs in 70 days; ~80% reduction in flaky-test eng hours. The post does not explicitly name the agent as Rovo Dev — but the Bitbucket + Pipelines + Jira integration described in the Fireworks post and the workflow-transition agent feature (documented at Atlassian Support) align with Rovo Dev's surface.

Vulnerability-remediation instance (2026-08-28)¶

Rovo Dev is the coding runtime in Atlassian's dispatcher/coding-agent/closer vulnerability-remediation system: invoked non-interactively as a Bitbucket Agentic Pipeline step, with its prompt in .rovodev/*.md, its model/tool config in .rovodev/pipeline-config.yml (pinning modelId: claude-sonnet-4-6 and allowlisting Atlassian MCP tools like transitionJiraIssue / createPullRequest), and its codebase-specific knowledge in a fix-vulnerability decision-tree skill. Reported: 120+ vulnerabilities resolved, 55+ automated PRs merged, 95% first-run merge rate (May–July 2026). The thesis "the prompts are the system" comes from this post. (Source: sources/2026-08-28-atlassian-agentic-automation-in-practice-putting-standard-engineering-work-on-autopilot)

Feature-flag-cleanup instance (2026-09-24)¶

Rovo Dev is again the coding runtime in Atlassian's feature-flag cleanup Dispatcher → Coding Agent → Closer loop: invoked non-interactively as a Bitbucket Agentic Pipeline step (prompt in .claude/flag-cleanup-agent.md, config in .claude/pipeline-config.yml), dispatched per-ticket by a scheduled Rovo Studio rule. The agent reads the ticket (flag name + final value), loads a feature-flag-cleanup skill, verifies the flag's state in the codebase before editing (patterns/verify-before-changing-code), inlines the surviving branch, removes the dead path/gate/imports at each call site, updates tests, runs the repo's checks, and opens a PR only if they pass — then comments the PR link and labels the ticket. Reinforces the thin-prompt / rich-skill thesis: "almost none of the domain logic lives in the prompt itself. Everything lives in the skill." In production since April 2026. (Source: sources/2026-09-24-atlassian-how-we-automated-feature-flag-cleanup-with-agentic-pipelines)

systems/atlassian-fireworks — the platform Rovo Dev built
systems/bitbucket-pipelines — the CI surface Rovo Dev reads
systems/jira — the Jira-Cloud-native agent integration surface (work items + workflow transitions).
systems/model-context-protocol — scoped, permissioned access to Jira/Bitbucket/Confluence APIs in the pipeline.
systems/atlassian-teamwork-graph — cross-product knowledge graph that supplements work-item context.
concepts/agentic-development-loop
agent-orchestration-skill
adversarial-review-persona
work-item-as-agent-prompt — KTLO-axis substrate framing.
agent-as-first-pass-investigator — KTLO-axis operational model.
ktlo-engineering-chores — KTLO-axis work category.
agentic-vulnerability-remediation — the fix-loop workload class.
prompts-as-the-system — the versioned-prompt thesis.
jira-status-transition-triggers-agent-workflow — KTLO-axis trigger.
agentic-pr-triage — KTLO-axis upstream signal.
agent-skill-with-fallback-chain — KTLO-axis per-codebase skill dispatch.
test-category-classifier-then-specialist-skill — KTLO-axis per-failure-category skill dispatch.
dispatcher-coding-agent-closer — the three-part loop.
decision-tree-skill — the fix-vulnerability skill shape.
companies/atlassian

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

Atlassian used Rovo Dev as the authoring stage in a live feature loop—Rovo Dev → Bitbucket → Pipelines → production—on the same Unleash PWA that attendees used. Eight features or bugs shipped live during the event. The post documents the composition and outcome, not Rovo Dev's internal implementation or quality-gate policy. (Source: sources/2026-08-07-atlassian-building-a-real-time-pwa-on-atlassians-own-stack)