Hold Devin's autonomous work to one merge standard
Devin runs autonomous sessions from AGENTS.md, Knowledge, and Playbooks, and can open pull requests across several repositories at once. Durable requirements end up split between repo files and instructions held inside Devin's Knowledge and Playbooks. Baz enforces the repo-held guidance, connects through MCP so Devin can verify dependencies before it opens a Change, and holds every cross-repo PR to the same merge standard as human-written code, each verdict tied to the commit it read.
Devin tells the agent how to work. Baz keeps the standard.
Native customization tells an agent how to work. Baz turns the durable guidance already versioned in your repository’s tree into scoped requirements it verifies on every Change, and Skill Maintainer opens a pull request when a rule stops matching the code.
Native customization
AGENTS.md, repo Skills, Knowledge, Playbooks, hooks, subagents, and MCP
Where instructions drift
Requirements split between repo-held AGENTS.md and Devin's Knowledge and Playbooks, so a rule that lives only inside Devin is invisible to every other agent and reviewer.
What Baz adds
Baz enforces repo-held AGENTS.md and Skills. Teams can move durable requirements out of Knowledge or Playbooks into versioned repo guidance so every agent and reviewer works from the same contract.
A plan Devin can defend
Devin connects to Baz through MCP. It can use Baz search to verify dependencies and consumers before it opens a Change.
Connect the Baz MCP endpoint to Devin.
Keep durable planning instructions in repo-held AGENTS.md and Skills.
Devin uses Baz search to find relevant repositories, symbols, and files before implementation.
Baz independently reviews the resulting Change against the same standard.
Result Broader planning context for Devin through MCP, followed by the same Baz review and merge standard.
Availability The packaged plan capture workflow is deepest today in Claude Code, Codex, and Cursor. Devin connects through the Baz MCP endpoint and repo-held guidance, while Baz independently reviews the resulting Change.
One independent merge standard
Devin can produce work many ways: Autonomous sessions, Knowledge, Playbooks, repo Skills, hooks, subagents, and MCP. Baz reviews the result after the code exists, reading the pushed commit straight from your git server rather than the agent’s workspace, and decides one thing: Whether the resulting Change follows versioned repo guidance and meets the same merge standard as every other Change.
| Baz capability | Evidence used | Decision supported |
|---|---|---|
| AI Coding GuidelinesStandards & Conventions | AGENTS.md, CLAUDE.md, Cursor rules, and Skills matched to changed paths | Did the agent follow the team's durable instructions? |
| Verification AgentSandboxed app review | The approved plan, the ticket and designs, and the running application | Did the agent build what the team asked for? |
| Code CorrectnessAI Code Review | Execution paths, API contracts, and consumers in this and other repositories, read at pinned commits | Does the Change work, and what does it break beyond the edited files? |
| Security ReviewSecurity Review | Advanced Security, SAST-inside signals, and the agent security reviewer | Does the Change introduce an exploitable risk, including in agent code? |
| SessionsSessions docs | What every agent ran, read, found, and skipped | Can the team see how the verdict was reached? |
| MergerMerge Automation | Findings, discussions, approvals, checks, and team merge requirements | Is the Change ready to merge, blocked, or waiting for a human decision? |
Keep every agent’s strengths. Add one control layer.
Each agent keeps its own way to shape behavior. Baz adds the same evidence before code and the same independent verdict after, taken from the git objects themselves rather than any agent’s copy, so the toolchain can change without resetting the standard.
| Capability | Claude Code | OpenAI Codex | Cursor | Devin |
|---|---|---|---|---|
| Durable repo guidance | CLAUDE.md, imported AGENTS.md, .claude/rules | AGENTS.md | Project Rules, AGENTS.md | AGENTS.md |
| Baz before code | Packaged Planner plugin and plan capture | Packaged Planner plugin and plan capture | Baz Rule, MCP, and plan capture | Baz MCP with repo-held guidance |
| Baz after code | Independent review and merge verdict | Independent review and merge verdict | Independent review and merge verdict | Independent review and merge verdict |
| Where Baz verifies | The pushed commit, not the local session | The pushed commit, not the worktree or cloud container | The pushed commit, not the editor's working tree | The pushed commit, not Devin's machine |
| Agent setup reviewed for security | Hooks, Skills, plugins, and MCP servers | Skills, plugins, and MCP servers | Hooks, Skills, plugins, and MCP servers | Repo Skills and MCP servers |
| Guidance kept current | Skill Maintainer PRs to CLAUDE.md and Skills | Skill Maintainer PRs to AGENTS.md and Skills | Skill Maintainer PRs to rules and Skills | Skill Maintainer PRs to AGENTS.md and Skills |
One control layer, whichever agent your team picks
Questions teams ask first
What does Devin support today?
Devin connects to Baz through MCP and uses repo-held AGENTS.md and Skills. The packaged plan capture flow currently targets Claude Code, Codex, and Cursor, while Baz reviews every Devin Change independently.
How do we move requirements out of Knowledge and Playbooks?
Store durable requirements as versioned repo guidance. Baz reads it, applies the relevant parts during review, and keeps one contract that every agent and reviewer shares.
Can teams keep human approval for sensitive code?
Yes. Teams can retain explicit human approval for high-risk Changes while Baz reduces repetitive review work on lower-risk ones.
Let developers keep Devin. Keep one standard for the code.
Connect Baz to your repositories and turn the guidance you already maintain into requirements verified at the commit on every Change, before code and again before merge.