Orcaflow
Autonomous coding task orchestrator for VS Code.
Orcaflow uses Copilot Chat in agent mode to investigate bugs, implement fixes, self-review code, create Azure DevOps pull requests, handle automated code review loops, resolve PR comments, and generate learnings — all with minimal human intervention.
Supports skill-based workflows, priority queuing, ADO work item linking, webhook notifications, and multi-cycle review automation.
✨ Key Features
- Autonomous Task Pipeline — 7-stage workflow: Created → Working → Review → PrCreation → CodeReview → ResolveComments → Completed
- Copilot Chat Integration — Uses
@orcaflow in agent mode with tool calls (advanceStage, finishTurn, notifyUser)
- ADO Integration — Creates branches, draft PRs, links work items, monitors pipelines, manages PR threads
- Automated Code Review — Detects review bot comments, feeds them to the agent, auto-resolves threads, supports multi-cycle review loops with configurable limits
- Skill System — Define reusable fix patterns. Import/export across teams. The agent auto-matches skills to tasks.
- Priority Queue — P0–P3 priority levels, auto-sorted. Swap-in higher priority tasks mid-flight.
- Webhook Notifications — Push task updates to Power Automate, Teams, or any HTTP endpoint
- Rich Sidebar — Tree views for active task, queued tasks, and skills with inline actions
- Detail Panels — Full task dashboard with stage timeline, logs, PR links, and review cycle tracking
⚡ Quick Start
- Install the extension
- Follow the setup toast (ADO token, org URL, project)
- Press
Ctrl+Shift+T to create your first task
- Start the orchestrator — Orca handles the rest
⌨️ Commands
| Command |
Shortcut |
| Orcaflow: Create Task |
Ctrl+Shift+T |
| Orcaflow: Show Sidebar |
Ctrl+Shift+O |
| Orcaflow: Start Orchestrator |
|
| Orcaflow: Stop Orchestrator |
|
| Orcaflow: Set ADO Token |
|
| Orcaflow: Resume Task |
|
| Orcaflow: Create Skill |
|
| Orcaflow: Import / Export Skills |
|
| Orcaflow: Send Test Notification |
|
| Orcaflow: Set Task Stage/State (Debug) |
|
⚙️ Settings
| Setting |
Default |
Description |
orcaflow.ado.organizationUrl |
"" |
ADO organization URL |
orcaflow.ado.project |
"" |
ADO project name |
orcaflow.ado.defaultRepository |
"" |
Default repository name |
orcaflow.ado.personalAccessToken |
"" |
ADO PAT (prefer secret storage via setup wizard) |
orcaflow.reviewBot.userId |
"" |
Review bot user ID for comment detection |
orcaflow.reviewBot.maxCycles |
2 |
Max automated review cycles before holding |
orcaflow.reviewBot.autoTriggerComment |
"copilot: review" |
Comment text that triggers a review bot pass |
orcaflow.notifications.enabled |
false |
Enable webhook notifications |
orcaflow.notifications.powerAutomateUrl |
"" |
Webhook endpoint URL |
📋 Task Lifecycle
Created → Working → Review → PrCreation → CodeReview ↔ ResolveComments → Completed
Each stage is handled autonomously. The orchestrator monitors ADO for PR status, review bot comments, and pipeline results. Human intervention is only needed for PR approval.
📂 .orcaflow/ Workspace Folder
Orcaflow creates a .orcaflow/ folder in your workspace root (auto-added to .gitignore). This is where all local state lives:
.orcaflow/
config.json ← project-level settings
run.log ← action log (append-only, human-readable)
status.json ← tool call audit log
skills/
{id}.json ← individual skill definitions
tasks/
{id}/
task.json ← persisted task data & stage history
context.md ← context file sent to Copilot Chat
status.md ← human-readable status summary
learnings.md ← auto-generated after task completion
config.json defaults
{
"branchNameFormat": "<username>/orcaflow/<slug>",
"stageTimeoutMinutes": 60,
"reviewWaitMinutes": 20,
"reviewMaxCycles": 2,
"holdTimeoutHours": 2,
"defaultModel": "",
"powerAutomateWebhookUrl": "",
"notificationsEnabled": true,
"emailRecipients": [],
"adoOrganizationUrl": "https://dev.azure.com/<org>",
"adoProject": "<project>",
"adoDefaultRepository": "<repo>"
}
Edit this file directly or use the first-run setup wizard. VS Code Settings (orcaflow.ado.*, orcaflow.reviewBot.*) override values in config.json.
run.log
Append-only log of every orchestrator action, stage transition, tool call, and error. Open it anytime to debug what Orca did, or check the Orcaflow Output Channel in VS Code.
🔧 Force Stage / State (Debug)
Stuck on a stage? Need to skip CodeReview or retry Working?
Run command: Orcaflow: Set Task Stage/State (Debug)
- Pick the task from the list
- Choose "Stage" or "State"
- Select the target value from the dropdown
This bypasses normal transition guards. The orchestrator then runs the stage-entry logic for the new stage (opens chat, sends continuation prompt, etc.)
Examples:
- Force to
Working — restart implementation from scratch
- Force to
CodeReview — re-trigger review bot polling
- Force to
Completed — mark done without merging
- Force state to
queued — re-queue a held/errored task
📦 Requirements
- VS Code 1.95+
- GitHub Copilot Chat extension
- Azure DevOps organization (for PR/work item features)
- ADO Personal Access Token with scopes: Code R/W, Work Items R/W, Build R/E, PR Threads R/W, Identity Read, Policy Read
License
MIT
