| name | 15-ai-agent-rollout-runbook |
| description | Use when promoting an AI agent from shadow through supervised to bounded autonomy with action gates, evidence, kill switches, bake periods, and rollback authority; use ai-feature-rollout-runbook for non-agentic AI features. |
| metadata | {"portable":true,"compatible_with":["claude-code","codex"]} |
AI Agent Rollout Runbook Skill
Use When
- Produce or update controlled AI rollout runbook from approved project evidence.
- Resolve decisions about staged exposure, promotion gates, bake periods, rollback triggers, and authority.
- Prepare a reviewable handoff for AI release and operations teams.
Do Not Use When
- The task is primarily owned by go-live-readiness; route there and use this skill only for its named output.
- Required project evidence or decision authority is unavailable and the requester expects a pass, release, certification, or production change.
Required Inputs
| Artefact | Source/provider | Required? | Behaviour when absent |
|---|
| Project _context/, approved requirements, and relevant architecture | Project owner and upstream phase skills | Required | Stop at a gap register; do not invent scope, thresholds, integrations, or owners. |
| Existing artefact, implementation, configuration, and evidence named below | Repository, delivery team, or service owner | Required when updating or assessing | Mark inaccessible items not assessed; do not treat them as passed. |
| Target audience, environment, risk tolerance, and authority | Requester and accountable owner | Required | Produce a read-only outline with explicit assumptions; do not mutate project or production state. |
Outputs
| Artefact | Consumer | Observable acceptance condition |
|---|
| Controlled AI Rollout Runbook | AI release and operations teams | Every stage has an eligible cohort, measurable gate, minimum observation period, rollback trigger, and named approver. |
| Decision and gap register | Reviewer and downstream phase owner | Every assumption, rejected option, unresolved dependency, waiver, and owner is explicit. |
| Validation evidence | Release or governance reviewer | Checks identify command or method, date, result, evidence location, and all unassessed items. |
Evidence Produced
| Evidence | Minimum content | Acceptance |
|---|
| Traceability record | Source artefact, decision, output section, owner | No mandatory decision is source-free. |
| Quality-gate result | Check, expected result, observed result, evidence path | Failures and unavailable checks cannot appear as passes. |
| Review record | Reviewer, date, disposition, open actions | The consumer can reproduce the acceptance decision. |
Capability and Permission Boundaries
- Minimum capabilities: read and search the authorised project sources. Execution is optional and limited to non-destructive validation.
- Inspection is read-only by default. Create or edit the named project document only when explicitly authorised. Production mutation, publishing, destructive action, spending, external communication, or certification claims require separate explicit authority.
- Treat secrets, tenant data, incident evidence, and financial records as least-privilege inputs; expose only the minimum evidence needed for review.
Degraded Mode
If files, execution, network, rendering, environment access, fonts, or current evidence are unavailable, return the narrowest useful draft plus a gap register. Label affected checks not assessed, retain the intended acceptance oracle, and state who must supply or verify the missing evidence. Never convert an unavailable check into a pass.
Decision Rules
| Choice | Action | Failure or risk avoided |
|---|
| Evidence is complete and authority is explicit | Choose promotion from observed stage evidence and produce the full artefact. | Scaling harm before it is measurable. |
| A required source or approval is missing | Stop the affected branch; record the gap, owner, and unblock condition. | Fabricated requirements or unauthorised action. |
| Evidence conflicts across sources | Preserve both claims, identify the controlling owner, and request a recorded decision. | Silent selection of a convenient but wrong source. |
| A check cannot run in the available environment | Keep its oracle and mark it not assessed; require later execution evidence. | False assurance from capability limits. |
Workflow
- Confirm the named deliverable, consumer, scope, environment, authority, and neighbouring-skill boundary.
- Inventory required sources and validate provenance, freshness, internal consistency, and missing inputs. Stop the affected branch on a mandatory gap.
- Extract traceable requirements, invariants, risks, and measurable acceptance criteria; record conflicts before choosing a design or procedure.
- Apply the decision rules and the domain workflow below. For a failed branch, preserve evidence, choose the documented recovery path, or escalate to the named owner.
- Draft the artefact, decision register, and evidence record together. Do not defer failure handling, rollback, security, tenancy, accessibility, or operational ownership.
- Run available checks, review every result, repair failures, and hand off only when acceptance is observable. If recovery fails or authority is exceeded, stop and escalate without mutation.
Quality Standards
- Ground every section in a named project source, decision, measured result, or accountable owner.
- Give each requirement or procedure a deterministic oracle that another reviewer can reproduce.
- Keep assumptions, exclusions, degraded checks, residual risks, and waivers visible at handoff.
- Preserve the domain invariants and more specific controls in the existing workflow below; this contract does not replace them.
- Run the repository anti-AI-slop gate: remove filler, verify named standards and dependencies, and retain purposeful domain detail.
Anti-Patterns
- Copying a generic template without mapping it to project sources. Fix: attach each section to an approved requirement, configuration, risk, or owner.
- Choosing a threshold because it is common practice. Fix: derive it from a requirement, measured baseline, risk decision, or current verified source.
- Reporting an inaccessible or unexecuted check as passed. Fix: mark it
not assessed, preserve the oracle, and name the verifier.
- Mixing the neighbouring go-live-readiness concern into this artefact without a boundary. Fix: cross-reference its output and keep ownership explicit.
- Omitting failure, rollback, empty-state, security, tenancy, or escalation behaviour. Fix: specify the trigger, safe action, verification, and owner for each applicable case.
- Mutating a repository, environment, tenant, ledger, or external system while drafting guidance. Fix: remain read-only until the exact mutation and authority are explicit.
- Claiming compliance, certification, readiness, or release from prose alone. Fix: require source-attributed evidence and a named acceptance decision.
Worked Example
Given an approved project source and a conflicting implementation detail, record both with provenance, stop the affected branch, and obtain the accountable owner's decision. Then update the relevant contract, define a reproducible acceptance check, and retain its observed result. The artefact is accepted only when every stage has an eligible cohort, measurable gate, minimum observation period, rollback trigger, and named approver.
References
- logic.prompt - load only when its template, logic, or detail is needed.
- README.md - load only when its template, logic, or detail is needed.
Core Instructions
Step 1: Rollout stages
Standard stages for an agent feature:
- Internal staff (dogfood) — engineering and ops use the feature in our own workspace.
- Shadow mode — the agent runs end-to-end and proposes actions; humans actually execute. The agent's plan and proposed actions are recorded and judged offline. No customer-visible change to outcomes.
- Canary at L1 — small named cohort opted in; the agent acts with per-call human approval; bake period.
- Tier rollout — promoted to the target tier (Pro or Enterprise) with the target autonomy level; percentage cohort within tier (e.g. 5% → 25% → 50% → 100%).
- GA — feature documented in trust center, sales enablement, customer commitments.
Shadow mode is mandatory between dogfood and canary for any feature that includes write or external tools. Shadow mode is the cheapest stage to collect human-judged trajectory evidence without exposing customers to agent execution risk.
Step 2: Promotion gates
Per stage, all gates must pass for the bake duration:
| From → To | Gates |
|---|
| Dogfood → Shadow | agent eval green; red-team smoke green; runbook drills complete |
| Shadow → Canary L1 | shadow agreement rate >= 0.85 (judge-LLM agrees with human action); zero CRITICAL / HIGH red-team open; irreversible-action-incident rate = 0 on shadow trajectories |
| Canary L1 → Tier rollout | task success >= SLO target; intervention rate <= SLO target; irreversible-action incidents = 0 during canary bake (minimum 14 d) |
| Tier rollout (cohort %) | per cohort: SLOs met for 7 d; no SEV1/SEV2 attributable |
| Tier rollout → GA | full SLO targets met for 30 d; trust-center entry approved; legal review of disclosures complete |
Step 3: Auto-rollback triggers
Wired to the feature-flag and dispatcher state:
- Irreversible-action incident → per-feature kill-switch + SEV1.
- Task success drop > 5 pp in 24 h → auto-rollback planner / prompt tag.
- Intervention rate up > 50% in 24 h → auto-rollback last planner / prompt change.
- Cost-per-run P95 > 200% of ceiling for 1 h → throttle then pause.
- Red-team CI regression on next PR → block merge.
Step 4: Comms plan
- Internal (dogfood and shadow): launch readiness review, run-of-show, ops channel watch list.
- Canary cohort: per-customer kickoff call; daily check-in for first 7 d; success-team owner.
- Percentage cohort: in-product modal disclosing agent feature + scope + revert path; help-centre article.
- GA: blog post; customer email; sales enablement; trust-center update.
Step 5: Opt-in / opt-out by region
- Default state per region: EEA defaults to off for L2+ features. UK / US / CA default on at L1 with disclosure.
- Workspace admin can disable per workspace at any stage.
- End user disclosure on first agent action ("This task was performed by the X agent on your behalf").
- Privacy notice for consent capture where the agent processes special-category personal data.
Step 6: Post-launch monitoring window
First 30 days post-GA: heightened monitoring; daily agent-quality stand-up; immediate review of every irreversible-action audit event; weekly review of cost-per-tenant; bi-weekly review of customer flags.
Step 7: Write the runbook
AI_Agent_Rollout_Runbook.md sections: 1) Rollout Stages, 2) Promotion Gates, 3) Auto-Rollback Triggers, 4) Comms Plan, 5) Opt-in / Opt-out, 6) Post-Launch Monitoring Window, 7) Roles & Sign-off Ledger.
Standards
- Google production-launch checklists
- Anthropic / OpenAI agent-launch playbooks
- ISO/IEC 42001 Clause 8
- EU AI Act Art. 13 (transparency) for disclosures
Resources
logic.prompt, README.md, references/ai-agent-rollout-runbook-template.md.