Can a task pack include policies and environment state as well as records?
Task packs can carry structured policies, deliberate rule conflicts and starting state, across several domains, either as static files or as a simulated environment with tool actions.
**Can a task pack include policies as well as records?** Yes. Policies can be generated as structured rules with identifiers, conditions and outcomes, plus a plain-language version. The agent can be asked to cite the rule it applied, which makes decisions easier to check.
**Can the policies contradict each other?** A buyer can ask for deliberate conflicts, such as two routing rules matching one case with a tie-break defined elsewhere. Those cases test whether an agent follows the precedence order or guesses.
**Can the pack vary by domain?** The same structure can be adapted to insurance claim lodgement, patient registration for a clinic's administration team, retail supplier onboarding or professional services engagement requests. Field names and rules change; the pattern of capture, check and route stays.
**Does it include a simulated environment, or only files?** Either. A dataset can be files in JSONL or Parquet. A simulated environment adds tool actions, state that changes as the agent acts, and a reset. Which one fits depends on whether you are training on trajectories or evaluating live agent behaviour.
**Is any of it real?** No. Everything delivered is generated. It is informed by direct observation of how real businesses operate, but it is not a copy of real records. Inspect samples for grounding and disclosure risk before relying on them.
Policies and start state make decisions citable and gradable, and the same structure adapts across financial, insurance, utility and healthcare administration settings.
- Audience:
- Agent developers
- Content type:
- FAQ
- Buyer question:
- What can be generated
Content release: r-a220eb6e099b243b
Related questions
- How would a task pack encode a duplicate connection request?
- What should a generated intake release specify?
- What formats and keys can a generated intake release use?
- What records can MirrorFirm generate for an intake workflow?
- What could MirrorFirm generate for an intake environment?
- Can a release include policies, multiple snapshots and deliberately messy submissions?