# Skyflo

> Skyflo is the control plane for agentic engineering. It orchestrates the coding agents, models, and repositories you already use under one governed mission, keeps every approval and review on the record, and learns from each mission it completes.

Skyflo Desktop runs missions on a linked Mac. Preview and Planned capabilities are omitted from this chip-less file.

## Available capability claims

One objective, with state that survives the session.; A workspace is an ordered set of registered Git checkouts; one mission may span them.; Give each boundary to the right specialist.; Each coder specialist works in its own git worktree.; Plan mode blocks the turn until you approve the plan.; Code, browser, terminal, and automation in one loop.; Some engineering work should keep running after you close the mission.; Scoped; Cited; Forgettable; What one mission learns, the next one can find.; Skyflo learns from the missions it completes and keeps what the evidence supports.; A workflow proven in two repositories can become a reusable Skill once another agent or you approve it.; Skyflo can improve how it routes work, adopting a change only when it beats the current policy on past routes it was not built from.; Approvals, credentials, spending limits, and kill switches sit outside anything Skyflo can change about itself.; Every change Skyflo makes to itself is recorded with who judged it, and one Revert undoes it.; One approval policy on every runtime: external agents start at maximum-ask, and Skyflo decides who answers each request and records the decision.; An external runtime cannot start on a mission without an approved plan, and everything it is told about the mission comes from durable mission state.; External agents reach Skyflo's tools through a per-session MCP gateway, so every call passes the same policy and approval path.; Each mission keeps an append-only record on your Mac, with every event hash-linked to the one before it.; Four access modes decide who answers an agent's requests; Ask for approval is the default, and a missing or unreadable setting falls back to it.; In every access mode except Full access, external agents launch under a macOS Seatbelt profile scoped to their leased worktree and granted roots, and Skyflo refuses to launch unconfined.; Values in deterministically sensitive files, such as keys and env files, are masked before they become model-visible.; Secret-shaped values are redacted before anything is persisted, logged, projected, or reported.; Every linked Mac is listed in the account console and can be revoked, and a revoked device's access stops working.; An agent that writes nothing reviews the work.; Execution stays on your linked Mac.; Bring your own model keys. They stay in the macOS Keychain.

## Canonical pages

- [Skyflo: The Control Plane for Agentic Engineering](https://skyflo.ai): Codex, Claude Code, Cursor Agent, and your models under one governed mission: plan approval, isolated worktrees, independent review, and memory that learns.
- [AI Agent Orchestration for Software Engineering](https://skyflo.ai/how-it-works): How Skyflo orchestrates coding agents: an approved plan, isolated specialists, one approval policy on every runtime, independent review, and a learning loop that carries each mission into the next.
- [Multi-Repo AI Coding Agents and Engineering Missions](https://skyflo.ai/missions): Skyflo coordinates AI coding agents across multiple repositories in isolated worktrees, with plan approval, independent review, and source-linked memory.
- [Persistent Memory for AI Coding Agents](https://skyflo.ai/knowledge): Local, source-linked memory that learns from completed missions, turns proven workflows into Skills, and stays reviewable and revertible, so your coding agents stop starting from zero.
- [Skyflo Pricing: Control Plane Plans for Teams and BYOK](https://skyflo.ai/pricing): Compare Skyflo plans from Free to Enterprise. The full control plane on every plan, your own model keys, local execution, and 20% off annually. No card required on Free.
- [Download Skyflo: Control Plane for Coding Agents on macOS](https://skyflo.ai/download): Skyflo Desktop runs on your Mac: Codex, Claude Code, Cursor Agent, and more under one mission, with isolated worktrees, browser and terminal work, memory that learns, and your own model keys (BYOK).
- [Capability Status: What Ships Today](https://skyflo.ai/status): Every registered Skyflo capability with its current Available, Preview, or Planned status and the code, test, or documentation evidence behind it.
- [Skyflo Desktop Release Notes and Changelog](https://skyflo.ai/release-notes): Every Skyflo Desktop release that reached the stable update channel, with what changed for the person using it: new capabilities, improvements, fixes, and upgrade notes.
- [Book a Demo: Skyflo for Engineering Teams](https://skyflo.ai/demo): Bring a real engineering objective and see the current Skyflo Desktop build run it on the agents your team already uses: plan approval, isolated specialists, independent review, and what Skyflo learns.
- [Skyflo Enterprise: The Control Plane for Agentic Engineering](https://skyflo.ai/enterprise): For CTOs and platform teams: how Skyflo governs the coding agents your engineers already run, where code and credentials live, what is recorded, and what ships today.
- [AI Coding Agent Control Planes and Harnesses | Skyflo](https://skyflo.ai/compare): Compare Skyflo with T3 Code, Omnigent, Claude Code, Codex, Cursor, and other coding-agent control planes and harnesses across execution, policy, review, and memory.
- [AI Coding Agent Use Cases for Multi-Repo Engineering](https://skyflo.ai/use-cases): Where an AI engineering harness earns its keep: coordinated change across repositories, and migrations that outlive a session, with what each mission learns carried into the next.
- [Cross-Repository Changes with AI Coding Agents](https://skyflo.ai/use-cases/multi-repo-development): When one change lands in four codebases: registered Git checkouts, a pinned interface contract, a specialist per repository, and a reviewer that cannot edit.
- [AI Agent for Framework and Dependency Migrations](https://skyflo.ai/use-cases/framework-migrations): Use AI coding agents for a framework or dependency migration across repositories: one approved plan, bounded specialist work, and context that survives each pass.
- [What Is an AI Engineering Harness?](https://skyflo.ai/concepts/engineering-harness): An engineering harness is the layer around AI coding agents that holds the objective: the plan, delegation, execution surfaces, review, and memory between runs.
- [AI Coding Agents, Orchestration, and Agent Memory](https://skyflo.ai/blog): Essays and build notes on AI agent orchestration, multi-repo coding agents, persistent engineering memory, self-improving agents, and independent code review.
- [Skyflo Kubernetes Agent, open source](https://skyflo.ai/open-source/kubernetes-agent): The Skyflo Kubernetes agent is a separate Apache-2.0 open-source project for cluster operations, with its own documentation and product boundary.
- [Introduction](https://skyflo.ai/open-source/kubernetes-agent/docs): Documentation for the separate Apache-2.0 Kubernetes agent.
- [Kubernetes agent capabilities](https://skyflo.ai/open-source/kubernetes-agent/capabilities): Capabilities of the separate open-source Skyflo Kubernetes agent, including typed Kubernetes, Helm, Argo Rollouts, and Jenkins operations.
- [Kubernetes agent use cases](https://skyflo.ai/open-source/kubernetes-agent/use-cases): Use cases for the separate open-source Skyflo Kubernetes agent across troubleshooting, CI/CD, incidents, and progressive delivery.
- [Kubernetes agent integrations](https://skyflo.ai/open-source/kubernetes-agent/supported-tools): Integrations documented for the separate open-source Skyflo Kubernetes agent, including Kubernetes, Helm, Argo Rollouts, and Jenkins.
- [Kubernetes agent community](https://skyflo.ai/open-source/kubernetes-agent/community): Community links and contribution paths for the separate Apache-2.0 Skyflo Kubernetes agent.

## Comparisons

- [Skyflo vs Claude Code: engineering harness or coding agent?](https://skyflo.ai/compare/claude-code): A source-dated comparison of Skyflo and Claude Code across execution, repository scope, approvals, review, automation, and memory.
- [Skyflo vs Cursor: engineering harness or agentic editor?](https://skyflo.ai/compare/cursor): A source-dated comparison of Skyflo and Cursor across execution, repository scope, approvals, review, automation, and memory.
- [Skyflo vs OpenAI Codex: engineering harness or coding agent?](https://skyflo.ai/compare/openai-codex): A source-dated comparison of Skyflo and OpenAI Codex across local and cloud execution, worktrees, automation, review, and memory.
- [Skyflo vs Devin: local engineering harness or cloud agent?](https://skyflo.ai/compare/devin): A source-dated comparison of Skyflo and Devin across execution environment, repository coordination, approvals, review, and knowledge.
- [Skyflo vs Factory: engineering mission or agent platform?](https://skyflo.ai/compare/factory): A source-dated comparison of Skyflo and Factory across environments, agents, repository coordination, approvals, review, and memory.
- [Skyflo vs Conductor: mission harness or parallel agent workspace?](https://skyflo.ai/compare/conductor): A source-dated comparison of Skyflo and Conductor across local and cloud workspaces, agent harnesses, repository coordination, review, and memory.
- [Skyflo vs Warp: engineering mission or agent platform?](https://skyflo.ai/compare/warp): A source-dated comparison of Skyflo and Warp across local and cloud execution, agent harness choice, parallel work, schedules, and memory.
- [Skyflo vs Cline: engineering mission or coding-agent task board?](https://skyflo.ai/compare/cline): A source-dated comparison of Skyflo and Cline across editor tasks, Agent Teams, Kanban, persistent rules, approvals, and review.
- [T3 Code vs Skyflo: plan mode, orchestrator, harness](https://skyflo.ai/compare/t3-code): T3 Code's legacy plan mode, its orchestrator and orchestrator v2, whether it is an agent harness, and how it differs from Skyflo, from sources read 2026-09-18.
- [Skyflo vs Omnigent: AI engineering harnesses](https://skyflo.ai/compare/omnigent): Compare Omnigent's agent meta-harness with Skyflo's mission-level control plane for agentic engineering across models, policy, isolation, collaboration, review, and memory.

## Published articles

- [What Is a HyperAgent? Self-Improving Engineering Agents With Boundaries](https://skyflo.ai/blog/skyflo-hyperagent-improves-how-it-improves): A HyperAgent improves how it improves. How Skyflo learns from completed missions, turns proven workflows into Skills, and tunes routing inside limits it cannot change.
- [Why AI Coding Agents Need an Engineering Harness](https://skyflo.ai/blog/what-is-an-engineering-harness): Coding agents improved file-level work. This essay defines the layer that holds the objective, plan, review, execution evidence, and accepted memory around them.
- [Why Multi-Agent Systems Should Not Use One Model for Every Role](https://skyflo.ai/blog/model-independent-orchestration-right-model-per-role): Missions mix deep implementation, broad research, and high-volume checks. Per-role model routing with BYOK, and why the orchestration layer is the durable one.
- [Why AI Coding Agents Need Memory Across Sessions](https://skyflo.ai/blog/engineering-memory-ai-agents-that-remember): Every agent session starting from zero is an organizational tax. What memory across sessions has to record, and why it only works when it is a by-product of the work.
- [How to Review and Verify AI-Generated Code](https://skyflo.ai/blog/evidence-backed-outcomes-verification): AI coding agents report success whether or not the work landed. Why review has to be structural: an agent that can read the change but has no authority to alter it.
- [Why AI Coding Agents Struggle Across Multiple Repositories](https://skyflo.ai/blog/cross-repository-missions-coordinated-change): Real changes span repositories, infrastructure, pipelines, and runtime. What coordinated multi-repository work requires beyond separate agent sessions.
- [Kubernetes Observability and AI Operations: Closing the Loop](https://skyflo.ai/blog/kubernetes-observability-ai-powered-operations): Observability surfaces signals. Operations still require a bounded decision, an approval gate, typed execution, and verification.
- [Why Approval Gates Must Be Architectural in DevOps AI Agents](https://skyflo.ai/blog/approval-gates-architectural-devops-ai-agents): Approval gates in DevOps AI agents cannot be UI toggles or confirmation prompts. They must be enforced at the execution engine level, below the model.

- [Expanded snapshot](https://skyflo.ai/llms-full.txt): Documentation text and the complete published article index.