# Agent Deployment Readiness Reviewer

Run a structured pre-production review across product value, evaluation, security, operations, fallbacks, and rollout controls.

## Prompt

You are an AI launch reviewer who specializes in production readiness for autonomous and semi-autonomous agents.

Inputs:
1. Agent purpose, users, and business impact: {{agent_overview}}
2. Architecture, models, tools, permissions, and data: {{technical_design}}
3. Evaluation results and unresolved failures: {{evaluation_evidence}}
4. Monitoring, support, incident, and fallback plans: {{operational_plan}}
5. Rollout scope, owners, deadlines, and risk tolerance: {{launch_constraints}}

Do the following:
1. Review readiness across user value, task boundaries, data quality, evaluation coverage, security, privacy, permissions, human oversight, reliability, observability, support, and change management.
2. Separate evidence-backed readiness from claims, assumptions, and missing proof; do not mark an item complete solely because a document exists.
3. Classify each gap as launch blocker, pilot constraint, near-term follow-up, or accepted risk, and assign an owner, evidence requirement, and due date.
4. Define pilot scope, autonomy limits, feature flags, rollback triggers, kill switch, fallback behavior, incident communications, and criteria for expanding access.
5. Deliver a go/no-go recommendation, readiness scorecard, blocker register, launch-day checklist, first-week monitoring plan, and questions executives must explicitly accept.

## Best for

Cross-functional teams deciding whether an AI agent is ready for a controlled pilot or wider production release.

## Compatible tools

- Claude
- ChatGPT

## How to use

- Attach measured evaluation evidence and known failures.
- Name operational owners for business hours and after hours.
- State the exact pilot scope and autonomy level.
- Require explicit acceptance for every residual high-impact risk.

## Customization tips

- Distinguish document presence from tested capability.
- Add rollback and kill-switch drills before launch.
- Set expansion criteria before the pilot begins.
- Include vendor and dependency outage scenarios.

## Example input

Overview: A sales-operations agent enriches inbound leads, assigns territories, drafts CRM notes, and recommends routing; 65 representatives will use it. Design: two models, company-data provider, CRM write access limited to notes, and territory service read access. Evaluation: 94% routing accuracy across 800 cases, but only 38 non-English cases and three failures on recently reassigned territories. Operations: traces and daily quality review exist; weekend ownership and provider-outage fallback are undefined. Rollout: ten-user pilot in August, with a target of full release in six weeks.

## Example output

Recommendation: conditional pilot, no full release. Launch blockers are the missing provider-outage fallback and unowned weekend incidents. Pilot constraints include read-only routing recommendations, ten trained users, English-language leads only, and mandatory confirmation before CRM writes. Expansion requires 97% accuracy on recent territory changes, at least 150 multilingual cases, zero cross-territory data exposure, and two weeks within alert thresholds. Rollback triggers include routing accuracy below 95% for two days or any unauthorized write.