From event to outcome
Four stages, in order. Nothing here is a black box: each stage names the action that performs it.
Trigger
A webhook, cron, slash command, @mention, or manual run fires the workflow.
Identify repository
e-volv resolves the relevant repos from the event context or explicit config.
Agent runsOpt-in
The right agent executes with tools, memory, and optional MCP extensions.
Execute & deliver
Results are written back to git, tickets, or CI as the workflow defines.
What goes in, what comes out
What can start a run, and what a run can do. Every entry is a node type or an event the engine already handles.
Triggers
- Pull requestsopened, merged, synchronized, labeled, reviewed
- Issuesopened, closed, labeled, commented
- CI eventsworkflow completed, status changed
- Scheduledcron expressions, recurring checks
- Interactive/review slash command, @evolve mention
- Manualdashboard button, API call
Actions
- git.cloneShallow clone a repository for code context.
- repo.identifyResolve relevant repos from trigger or config.
- agent.runRun a single-purpose AI agent with a toolset.
- agent.sessionCoordinate multiple agents around a complex task.
- ci.executeWorkflowTrigger external CI and poll for completion.
Worked examples
Four workflows teams actually run, each assembled from the node types above.
AI code review
On every PR, gather context, run a Code Reviewer agent, and post inline comments plus a verdict.
Fix failing CI
When CI fails, diagnose logs, generate a patch, push a commit, and monitor the re-run.
Ticket to PR
When a ticket is marked ready, break it down, generate a technical design, and open an implementation PR.
Run one against your own repository
Eighty-nine playbooks ship ready to run. Open one, read the graph and the prompt, then point it at a repository.