ICMD ALOS (Agentic Leadership Operating System)

Purpose
Use this to scale AI-agent execution without increasing incident rates, security exposure, or architectural fragmentation. Treat it like an “org upgrade,” not a tooling experiment.

1) Decision Rights (write this down)
- Delegation rights: Who can ask agents to open PRs, modify infra-as-code, or run migrations?
- Approval rights: Who can approve merges to main? Who can approve production deploys?
- High-blast-radius areas: Enumerate repos/services where approvals must be stricter (payments, auth, data pipelines, IAM).
- Break-glass: Define who can bypass controls in emergencies, and how audit logs are captured.

2) Standard Workflow (spec → PR)
- Spec template (1 page): goal, non-goals, success metric, acceptance tests, rollback plan.
- PR template: summary, safety checklist, tests, observability changes, data/security section.
- Artifact requirement: agent must draft (a) test plan and (b) rollback steps for every non-trivial change.
- Review SLA: set a target (e.g., PR reviewed within 24 business hours) and staff for it.

3) Guardrails (minimum viable safe autonomy)
- Branch protections: required reviewers + CODEOWNERS for key paths.
- CI enforcement: lint/format, unit/integration tests, build reproducibility.
- Security baseline: secret scanning, dependency alerts, SAST (e.g., CodeQL), SBOM generation.
- Progressive delivery: feature flags + canary or staged rollout; automated rollback triggers.
- Permissions: least privilege; short-lived credentials; restrict agent access to production.

4) Metrics & Risk Budget (manage blast radius)
Track weekly:
- DORA: deployment frequency, lead time, change failure rate, MTTR.
- Operational load: pages per deploy, after-hours incidents, top 3 recurring failure modes.
Set a quarterly risk budget:
- Max tolerated customer-impact minutes (SEV-2 equivalent)
- Max remediation spend (engineering hours × loaded cost)
When the budget is exceeded: slow rollout, increase tests, tighten approvals.

5) 30-Day Rollout Plan
Week 1: pick 2 pilots (1 internal, 1 low-risk external). Assign single accountable owner per pilot.
Week 2: implement guardrails and templates. Turn on required checks.
Week 3: measure outcomes; run near-miss reviews; fix systemic gaps.
Week 4: expand to next team only if quality holds (no major rise in change failure rate/MTTR).

6) Cultural Reinforcements
- Reward deletion and simplification (remove flags, dead code, unused features).
- Weekly incident + near-miss review (blameless, action-focused).
- Maintain an Architecture Decision Record (ADR) log to prevent “prompt drift.”

Copy-paste templates included above; adapt to your compliance needs and toolchain.