Autonomy Levels

Scraut offers three levels of agent autonomy. Choose based on your team's comfort with AI-driven automation.

---

supervised (recommended to start)

agents:
  autonomy_level: supervised

How it works:

• Agents propose sprint assignments → human SM approves before execution
• Agents implement tasks and open PRs → human review always required before merge
• Agents submit daily standups → human team reviews summaries
• Human checkpoint fires at every sprint boundary

Best for: Teams new to agent mode, high-stakes domains, regulated environments.

What requires human action:

• Approving the sprint plan each sprint
• Reviewing and merging every agent PR
• Resolving escalations

---

semi-auto

agents:
  autonomy_level: semi-auto

How it works:

• Agents claim issues and implement them without waiting for approval
• PRs still require human review before merge
• Human checkpoints only at sprint boundary and on escalation
• Agents can self-assign issues from the backlog based on specialty

Best for: Teams that have run supervised mode for 2+ sprints and trust the agents.

What still requires human action:

• Reviewing and merging every agent PR
• Sprint boundary approval
• Resolving escalations

---

full-auto

agents:
  autonomy_level: full-auto

How it works:

• Agents run continuously, claim issues, implement them, and request review
• Sprint planning happens automatically without waiting for SM input
• Humans receive weekly digests only
• Emergency checkpoints still fire (milestone slip, multiple escalations, repeated failures)

Best for: Very mature teams with well-defined acceptance criteria on every issue, and a high-trust relationship with the LLM provider.

Full-auto requires preparation

Full-auto mode is only effective when:

• Every backlog issue has clear, specific acceptance criteria
• CI is comprehensive and catches regressions reliably
• The definition_of_done in scraut.yml is strict
• The team reviews the weekly digest and acts on escalations promptly

---

Changing autonomy level

You can change the autonomy level between sprints. Edit workspace/scraut.yml:

agents:
  autonomy_level: semi-auto  # was supervised

Commit and push. The change takes effect on the next orchestrator cycle.

---

Comparison table

	supervised	semi-auto	full-auto
Sprint plan approval	Human required	Human required	Automated
Issue claiming	Human approves	Automated	Automated
PR merge	Human required	Human required	Human required
Sprint boundary	Human checkpoint	Human checkpoint	Auto
Escalation response	Human required	Human required	Human required
Weekly digest	Yes	Yes	Yes
Suitable for	All teams	Teams with 2+ sprint history	Mature, high-trust teams

---

Recommended progression

1. Sprints 1–2: Run with agents disabled — establish your baseline velocity and processes
2. Sprint 3: Enable agents in supervised mode with 1 agent and 1 low-risk issue
3. Sprint 5: Expand to more agents if supervised mode is working well
4. Sprint 8+: Consider semi-auto if you're consistently happy with agent PRs
5. Never rush to full-auto — supervised and semi-auto deliver most of the value at lower risk