Context Rule recipes
Adapt practical organization standards for common AI agent workflows.
These recipes are starting points for Context Rules. Adjust the wording, triggers, and scope to match your company's actual requirements. Keep each rule focused on one operating expectation so it is easy to test and update.
Context Rules provide open-ended guidance. Pair them with an Operational Policy when a recognizable command must warn, pause for approval, or be blocked, and with a Sensitive Data rule when prompt content needs a specific protection outcome.
Accurate completion reports
Recommended trigger: Every prompt
Context to add:
Report only work that was actually completed and verified. Separate confirmed results from assumptions, describe any checks that were run, and identify remaining risks or follow-up work.
Use this as a baseline when the organization wants consistent, evidence-based agent responses.
Destructive or irreversible operations
Recommended trigger: Keywords such as delete, remove, destroy,
terminate, drop, purge, and organization-specific equivalents
Context to add:
Before any destructive or irreversible operation, identify the exact environment, target, scope, and expected impact. Explain the proposed action to the employee and obtain explicit confirmation before proceeding. Do not treat the original task request as confirmation.
Test this rule only in a disposable or non-production environment. The agent must pause for confirmation before deletion.
Production changes
Recommended trigger: Keywords such as production, prod, deploy,
release, migration, and production environment names
Context to add:
Before changing production, confirm the target environment, describe the planned change, state the expected customer impact, prepare rollback and verification steps, and identify any approval required by company process. After the change, report the checks performed and their results.
Incident response
Recommended trigger: Keywords such as incident, outage, degraded,
sev, and rollback
Context to add:
During an incident, prioritize service recovery and verified facts. Keep a concise timeline, distinguish observations from hypotheses, avoid unrelated changes, and provide the employee with clear impact, action, owner, and next update information.
Secrets and sensitive data
Recommended trigger: Keywords that identify credentials, customer data, exports, logs, and internal data classifications
Context to add:
Minimize access to sensitive data. Do not expose credentials or personal data in prompts, source changes, logs, or responses. Use approved secret and data handling methods, and stop for employee direction when the required boundary is unclear.
Pair this guidance with a Sensitive Data rule when a matched prompt must be recorded, warned, or blocked.
Access and permission changes
Recommended trigger: Keywords such as grant, role, permission,
admin, owner, access, and policy
Context to add:
Apply least privilege. Confirm the exact identity, resource, environment, requested permission, duration, and approver before changing access. Explain the resulting capability and how access will be reviewed or removed.
External communication
Recommended trigger: Keywords such as customer, status page, email,
announcement, post, and publish
Context to add:
Treat external communication as a proposed draft unless the employee clearly authorizes publication. Verify facts, audience, confidentiality, and required review. Do not send or publish content without explicit employee approval.
New dependencies and services
Recommended trigger: Keywords such as install, dependency, package,
vendor, service, and integration
Context to add:
Prefer organization-approved dependencies and services. Before introducing a new dependency, explain why it is needed, its maintenance and licensing implications, what data it can access, and whether an approved alternative already exists.
Turn a recipe into a rule
- Open Governance → Organization Library.
- Create a rule with a name employees and Owners will recognize.
- Choose a trigger that matches actual employee language.
- Paste and adapt the context.
- Apply it to the appropriate agent types.
- Activate and test it with a safe prompt.
- Review match activity and refine the rule.
For rule design principles and testing, see Organization Library, which shows the real Context Rules editor and usage views.