Agent Prompt Reviewer is a free, MIT-licensed agent skill that audits an AIML.chat agent system prompt and rewrites the weak parts — checking for grounding, citations, escalation, and safe handling of write actions. Paste your prompt and get a sharper, safer one back with reasons.
What it does
The skill reviews a system prompt against the practices that make an agent reliable: does it require answering from context only, does it cite sources, does it escalate instead of guessing, does it treat mutating actions as needing approval, and is the tone right for the audience. It flags missing guardrails — a prompt that lets the model answer from general knowledge, or one that promises actions without confirmation — and rewrites them. The output is an improved prompt plus a short list of what changed and why, so you learn the patterns rather than just accepting an edit.
The skill
name: agent-prompt-reviewer
Use when you have an agent prompt and want it audited and improved.
Steps:
1. Read the existing system prompt and the agent's intended job.
2. Check grounding, citation, escalation, tone, and write-safety rules.
3. Flag missing guardrails and rewrite the weak sections.
4. Output the improved prompt plus a changelog of what changed and why.
How to use it
Copy the SKILL.md into your agent framework's skills folder and run it on any prompt, or apply the result in AIML.chat under Agents. It pairs with the builder skills that generate a first draft — see the support agent builder — and the configure-agent skill to wire the rest. The agent overview explains the behaviors it checks for.
Ready to tighten your agent? Start free, paste your prompt, and ship a better one in minutes — see pricing for limits.
What does it check for?+
Grounding, citations, escalation, tone, and whether mutating actions are gated — the rules that make an agent reliable and safe.
Will it explain its edits?+
Yes — it returns a changelog of what changed and why, so you learn the patterns, not just the result.
Is the skill free?+
Yes — it is MIT-licensed and free to copy. Running the agent counts toward your plan's limits; see pricing.