Agents that start with context, not a cold prompt
When a Work Order runs, skaftorAI passes the agent an execution manifest: the repository, the branch, the relevant files, the acceptance criteria, and the memory of how similar work shipped. The agent opens the right workspace and starts inside it — the difference between a useful contribution and a plausible guess.
- Cursor and Claude Code driven over MCP
- A full execution manifest per run — repo, branch, files, criteria
- The workspace resolved and opened automatically
From execution to a pull request
The run produces a branch and an automated pull request, drafted with a summary of what changed and why. It lands in your normal review flow — the agent does the typing, your team keeps the judgment.
- Automated PRs with an AI-drafted change summary
- Delivery tracked from execution through merge
- Provider-agnostic — GitHub today, more to come
Orchestrated, and accountable
Every run is recorded — which agent, which model, on which branch, for how long, at what cost. That trail feeds cost visibility and organizational memory, so agent work is as auditable as any other change.