Agent Prompt Quality Bar
Audits an agent system prompt against a rubric of contract, fallbacks, and anti-patterns.
Install
npx skills add mastra-ai/mastra --skill agent-prompt-quality-bar -g -a claude-code -y- Repository
- mastra-ai/mastra29.3k
- Category
- AI and agents
- Updated
- 3d ago
- License
- See repository
- Context cost
- Mediumabout 1,868 tokens
- Runs scripts
- No, instructions only
What it does
Universal quality bar and final audit rubric for any agent system prompt. Activate this whenever you are unsure which archetype skill applies, or as a final review pass before writing the system prompt. It defines the required run contract, completion criteria, fallback paths, response format, and anti-patterns every produced agent prompt must satisfy.
Universal Agent-Prompt Quality Bar
This is the builder's compile check. If no archetype skill matches the user's request, OR if you want a final review pass after picking an archetype, follow these rules to write a great agent system prompt.
A produced agent is good only when it can finish a real run, not when it merely sounds capable.
Required run contract
Before writing the produced system prompt, instantiate this contract for the user's request:
- Trigger / input, what starts the run: a user message, schedule, webhook, file, spreadsheet row, support ticket, document, or event.
- Owned outcome, the one concrete result the agent is responsible for finishing.
- Available capabilities, only the tools, data sources, workflows, agents, or stored skills actually attached or available.
- Missing-capability fallback, what the agent must do when a required integration, credential, workspace, permission, source, or input is absent.
- Done criteria, how the agent proves the job is complete, including tool confirmation, read-back, tests, citations, delivery receipt, or an explicit not-run reason.
- Final response format, the exact receipt, report, draft, diff summary, confirmation, or escalation note the user receives.
Non-negotiable sections
Every produced agent system prompt MUST contain these sections. If any is missing, the agent will underperform.
1. Role and outcome
- One sentence: who the agent is and the single outcome it owns.
- Bad: "You are a helpful AI assistant."
- Good: "You are Acme's Tier-1 Email Triager. You reply to inbound billing and account tickets, or escalate cases that need a human."
2. Trigger and input
- Name what starts a run and what data the agent expects.
- Bad: implicit trigger.
- Good: "A run starts when the user provides a support ticket or a scheduled inbox workflow passes unread tickets to you."
