Fleet 1.13:Teams are now shipping 5x more PRs with autonomous pipelines.See what's new →
FleetFleet
Glossary

Prompt as Job Description

Prompt as job description is a design philosophy for AI agent prompts where the system prompt is structured like a human job description — defining role, responsibilities, scope, reporting relationships, and behavioral expectations — rather than as a list of instructions for a single task.

A task-instruction prompt tells the agent what to do right now. A job-description prompt tells the agent who it is, what it owns, and how it behaves across all tasks it will ever receive. The distinction matters for agents that run repeatedly on varied input: a developer agent needs consistent behavior across many different tickets, not behavior tuned to one specific task.

Job-description prompts typically include: the agent's name and role, its department and reporting structure, a first-thing-every-run checklist (check inbox, check open issues), the specific skill or workflow it runs, communication protocols (how and when to publish events, comment on PRs), git rules, and escalation criteria. This is closer to an employee onboarding document than a system prompt in the traditional sense.

The advantage of the job-description approach is stability and consistency. The agent behaves like a team member with a defined role rather than a stateless instruction-follower. The disadvantage is prompt length — job-description prompts are longer, consuming more tokens per session. The tradeoff is typically worthwhile for agents that run many times, since the consistency benefit compounds.

How this relates to Fleet

Fleet builds the job-description prompt structure into every agent automatically. The Fleet Agent Handbook — compiled into the binary — provides the universal role-agnostic instructions (inbox check, git rules, communication protocol). The role-specific skill directive is added deterministically based on the agent's role field. Operators write only the domain-specific content; Fleet provides the structural skeleton.

Frequently asked questions

How long should a job-description prompt be?

Concise enough to leave room for task context, detailed enough to produce consistent behavior. In practice, 500-1500 tokens for the system prompt works well for most agent roles. Longer prompts are appropriate for complex roles with many edge cases; shorter prompts are appropriate for narrow, well-defined roles. Avoid repeating information that the agent will receive in task context — the job description should cover only what is constant across all tasks.

Should every agent in a fleet have the same base prompt?

Sharing a common base prompt for universal rules (git discipline, communication protocol, escalation criteria) and extending it with role-specific content is a good practice. It ensures consistent behavior on cross-cutting concerns while allowing role-specific customization. Fleet implements this as the Fleet Agent Handbook (shared) plus per-role skill directives (specific).

Run your first agent fleet

One binary. Five minutes. See every agent, coordinate every handoff, and keep a full audit trail of what your fleet did.