Deploy workflow (internal agent) — record_processedAction contract

The agent may deploy a service only when the service is healthy and the target environment is allowed. A successful deploy must show up as a completed pipeline in CI, a healthy check in monitoring and a change record, all within 30 minutes. Exactly one deploy per request, at most three deploys per agent per hour, never an unhealthy check afterwards, and a rollback within 24 hours fails the deploy.

Contract ID: deploy_workflow
Source: internal://policies/deploys
Retrieved: 2026-10-11
Notes: Template values. Replace with your change-management policy and the limits your platform team signed off.

What the contract checks

Each block is evaluated against read-only facts from the systems of record. The agent's own report is never used as evidence.

Contract YAML

The machine-readable action contract: allowed action, required inputs, preconditions, limits, expected and forbidden effects, window. Sign in to select it for your tenant or fork it.

# Deploy workflow — action contract for an agent that ships a release to a running service.
# Internal policy template: rename the systems (ci, monitoring, changes) to the ones your platform team already
# reads from. Nothing here is specific to engineering: the checks are the same shape as every other contract.
id: deploy_workflow
version: 1
kind: action
vendor: internal
vendor_display: "Deploy workflow (internal agent)"
action_type: deploy_service
outcome_type: record_processed
claim_source: agent.actions
subject_kind: service
description: >
  The agent may deploy a service only when the service is healthy and the target environment is allowed. A
  successful deploy must show up as a completed pipeline in CI, a healthy check in monitoring and a change record,
  all within 30 minutes. Exactly one deploy per request, at most three deploys per agent per hour, never an
  unhealthy check afterwards, and a rollback within 24 hours fails the deploy.
source_citation:
  url: "internal://policies/deploys"
  retrieved: "2026-10-11"
  title: "Internal deploy policy (template)"
  notes: >
    Template values. Replace with your change-management policy and the limits your platform team signed off.
unknown_fields: []
action:
  window: "30m"
  preconditions:
    - state_before_action:
        source: monitoring
        type: service_health
        where: {status: healthy}
        message: "Service was not healthy before the deploy"
  limits:
    - param_in: {param: environment, values: [staging, production]}
  aggregate_limits:
    - aggregate_exceeds:
        source: ci
        type: pipeline_completed
        max: 3
        over: "1h"
        by: agent
        message: "This agent has deployed more than three times in an hour"
  expected_effects:
    - effect_observed:
        source: ci
        type: pipeline_completed
        where: {status: success}
        message: "CI shows no successful pipeline for this deploy"
    - effect_observed:
        source: monitoring
        type: health_check
        where: {status: healthy}
        message: "Monitoring shows no healthy check after the deploy"
    - effect_observed:
        source: changes
        type: change_recorded
        message: "No change record was written"
  forbidden_effects:
    - effect_count_exceeds: {source: ci, type: pipeline_completed, max: 1}
    - effect_observed:
        source: monitoring
        type: health_check
        where: {status: unhealthy}
        message: "The service reported unhealthy after the deploy"
  reversed_by:
    - effect_observed:
        source: ci
        type: rollback_completed
        within: "24h"
        message: "The deploy was rolled back"
  unverifiable_if:
    - missing_fact_source: [ci, monitoring]
billing:
  unit_price: 0
  currency: USD

Use this contract See it applied in a sample verification