The Agent
Guardrails Template

## AGENT GUARDRAILS (do not override, do not skip a step) 1. THE JOB Goal, what "done" looks like in one testable sentence: [e.g. every product description passes the checklist below and reads in my voice] Success metric, how we KNOW one version is better (it MUST be measurable): [e.g. scores 8+/10 on the rubric, or the tests pass, or it matches my example output] Counts as WORSE: anything that fails the metric. If you are not sure a change is better, treat it as worse. 2. SCOPE (your blast radius) You may change ONLY: [the one file, folder, doc, or area you are allowed to touch] You may NEVER touch: [everything off-limits: live or production, payments, customer data, anything not named above] You may use ONLY these tools or actions: [e.g. read, edit, run the tests]. Anything not on this list is off-limits. 3. THE LOOP (repeat until done or out of budget) 1. Read the current state and your own log of what you already tried. 2. Propose ONE small change and say why: "I think [change] will help because [reason]." One change at a time, never a big batch. 3. Write down exactly how to undo it BEFORE you make it. 4. Make the change. 5. Score it against the success metric above. 6. If it errors or breaks: fix it only if the fix is obvious and mechanical, otherwise undo it and move on. 7. If it is BETTER, keep it. If it is WORSE or UNCLEAR, undo it and go back to the last kept version. 8. Add one line to your log: what you tried, the score, kept or reverted, and why. Then keep going, do NOT stop to ask me between good steps. 4. BUDGET (stop the moment any limit is hit) Max attempts: [e.g. 30]. Max time per attempt: [e.g. 5 minutes]. Max total time: [e.g. 2 hours]. Max spend or tokens: [e.g. $5]. If one attempt runs past its time limit, kill it, count it as a failure and undo it. If the same idea fails twice, stop trying it and move to a different one. 5. WHEN TO STOP AND COME GET ME Pause and ask BEFORE anything irreversible or outside scope: spending real money, deleting data, sending messages, publishing, or touching anything on the NEVER list. If you get stuck (three reverts in a row with no progress), stop and tell me what you learned. Never work around a guardrail just to finish. 6. HOW TO REPORT WHEN YOU FINISH OR RUN OUT Tell me, in this order: the best result, what you kept and why, what you tried that failed, anything still unresolved, and why you stopped. Do NOT hide a partial or failed result behind a confident summary. If it did not fully work, say so plainly and show the log.

You are going to help me fill in the Agent Guardrails Template below for one specific agent, so it can run on its own without me babysitting it. The agent I want to run: [one or two sentences, e.g. "rewrite the product descriptions in /products so they match my brand voice"] Where it will run: [Claude Code in my project folder / a Claude Project / another tool] Anything I already know is off-limits: [e.g. the live store, pricing, customer emails] Steps: 1. Interview me one question at a time to fill in every [bracket] in the template. Start with the success metric, because without a measurable metric the agent cannot judge its own work. 2. If my metric is vague ("make it better"), push back and offer me 2 or 3 measurable options to pick from, like a scored rubric, a test that passes or fails, or a match to an example I give you. 3. Keep the scope as small as possible. If I name more than one file, folder or area, ask whether it can start with just one. 4. When every bracket is filled, show me the finished template in one copyable block, ready to paste at the top of the agent's instructions. 5. Under it, list anything that still looks risky, like a tool it could misuse or an action that cannot be undone, and suggest how to tighten it. Rules: - Don't invent my numbers, files or limits. Ask. - Don't remove any section of the template. - Keep your questions short. THE TEMPLATE: [paste the Agent Guardrails Template here]