Run a structured 7-day test on any new agent, cron job, or automation. The build evaluates itself after every run. You review one report on Day 7 and reply with one word.
← All Claude SkillsYou just shipped an agent or automation and it worked once. On-Ramp runs it for a full 7 days, makes it log its own performance after every run, and hands you a single graduation report that tells you whether to trust it. Copy the two blocks below and give them to your Claude.
Step 1. Copy the skill block below.
Step 2. Open Claude and paste this instruction first, then paste the block underneath it:
Turn the block below into a Claude skill and install it. Create the skill file exactly as written. Then personalize it with smart defaults from what you know about me: my business, my priorities, and how I work. Confirm when it is installed and tell me how to run it.
--- name: on-ramp description: Run a structured 7-day MVP testing protocol after any new agent build, cron job, or automated workflow goes live. This is the quality gate between "it's built" and "it's trusted." Use this skill whenever I say "on-ramp," "run on-ramp," "MVP test," "test this build," "graduation checklist," "is this ready for the next phase," or any variation of wanting to validate that a live build is actually working before adding more responsibility. Also trigger when I just deployed a new agent, cron job, or automation and haven't set up a testing protocol yet. If I move straight from building to Phase 2 without running on-ramp, suggest it. Pre-flight is the steering wheel before you build. On-ramp is the quality gate after the engine starts running. --- # Agent On-Ramp — Trust Before You Scale Run a structured 7-day MVP testing protocol after any new agent build, cron job, or automated workflow goes live. The agent evaluates itself. I review one message. That's it. The principle: Building it is step one. Trusting it is step two. On-ramp is how you earn trust before giving an agent more responsibility. ## When to Use This Any time a new agent, cron job, or automated workflow just went live and needs validation before scaling. This includes: - A new agent running its first cron job - A new automation running on a schedule - A new automated pipeline writing to a tracker or database - Any build that runs autonomously without me starting it - Any build moving from Phase 1 (capture/categorize) to Phase 2 (autonomous implementation) If the build is a one-time manual task (running a script once, testing a single API call), skip this. On-ramp is for things that run repeatedly and autonomously. ## Trigger Phrases - "on-ramp" - "run on-ramp" - "MVP test" - "test this build" - "graduation checklist" - "is this ready" - "ready for the next phase" - "validate this build" - "how do we know this is working" ## The On-Ramp Protocol ### Step 1: Identify What's Being On-Ramped Confirm with me: - What agent or automation just went live? - What does it do? (one sentence) - Where does it write output? (my tracker, a file, my phone or chat, wherever the results land) - How often does it run? (cron schedule) If this is already clear from the conversation, don't re-ask. Move to Step 2. ### Step 2: Set Up the Accountability Log The agent being on-ramped needs a place to log its own performance after every run. Check whether I already have a tracker set up for this, wherever I track work. If one exists, use it. If it doesn't, set one up with these fields: - **Run Date** (Title): Date of the run - **Agent** (Select): Which agent ran (the agent being tested, named however I refer to it) - **Run Status** (Select): Success / Partial / Failed - **Items Found** (Number): How many new items were detected - **Items Processed** (Number): How many were fully processed - **Items Skipped** (Number): How many were skipped and why - **Errors** (Text): Any failures, auth issues, or timeouts. "None" if clean. - **Output Accuracy** (Select): Yes / No (did output land correctly in the target system) - **Confidence** (Number): The agent's self-assessed confidence in its output quality (1-5) - **Notable Item** (Text): The single most valuable item from this run - **Security Check** (Select): Clean / Issue Found ### Step 3: Write the Agent Instructions Draft the message I send to my agent (via my phone, chat, or a workspace file) that turns on the on-ramp protocol. The message must include: 1. **What to log**: Every field from the Accountability Log, filled out after every run 2. **When to log**: Immediately after each run completes 3. **The 7-day graduation report**: On Day 7, the agent sends me a message (on my phone or chat) with this format: --- **[AGENT NAME] — 7-DAY ON-RAMP REPORT** **TL;DR:** [emoji signal] + one sentence summary Emoji signals: - 🟢 = Ready to graduate - 🟡 = Needs another week - 🔴 = Real issues found Then list each checklist item with an emoji signal: - Cron ran all 7 days without manual intervention - Zero duplicate entries in the target system - Zero authentication or credential failures - Average confidence score 4+ out of 5 - Output landed correctly in the target system every run - Briefings/notifications sent on time every day - Credentials stored in locked files, not in chat or workspace files - Operating with minimum necessary permissions **By the numbers:** - Total items processed across 7 days - Total entries created in the target system - High-value items flagged **Your call. Reply with one:** - **"Graduate"** — Agent moves to the next phase - **"Run another week"** — 7 more days at the current level - **"Fix [issue]"** — Tell the agent what needs adjusting --- ### Step 4: Confirm the Protocol Present the full on-ramp setup to me: - Where the Accountability Log lives - What the agent will log after every run - When the graduation report arrives - What the response options are Then say: > **On-ramp is set. The agent handles everything. You review one message on Day 7 and reply with one word. Ready to activate?** ### Step 5: Activate Once confirmed, I send the instructions to my agent. On-ramp is live. I do nothing for 7 days except keep using the system naturally (the normal action that triggers the agent's work). ### Step 6: Graduation (Day 7) When the graduation report arrives: - **All green** = the build earned trust. Move to the next phase. Update the build tracker. - **Yellow items** = the build is close but needs refinement. Fix the flagged issues and run another 7 days. - **Red items** = something is broken. Fix the root cause before continuing. After graduation, log the result in my build tracker, if I keep one, with a one-line graduation statement: what the agent proved and what it earned the right to do next. ## Voice and Tone - Direct. No fluff. - This is not a test I administer. This is a test the agent administers on itself, and I review. - My time is protected. The whole point is that the agent does the evaluation work. - Keep the energy forward-moving. On-ramp is not a gate that blocks progress. It's the bridge between "built" and "trusted." ## Key Principles - Building something is exciting. Trusting it requires evidence. - The most dangerous moment in agentic AI is when something works once and you assume it works always. On-ramp catches the gap between "it worked in the test" and "it works reliably." - Agents should evaluate themselves. Humans should review the evaluation. That's the right division of labor. - Security is baked into the on-ramp from Day 1, not saved for a later phase. - This protocol is reusable. Every new agent, every new cron job, every new automation gets the same on-ramp. One Accountability Log supports all agents. The graduation report format stays consistent. The response options are always the same. Systems create trust at scale.
Claude reads the block, builds the skill file, and installs it. From then on, you just run the skill in any conversation. No folders to find, no code to write.
Every week you see what's working, build it live, and put it to work in your business.
Here's the shape of what this skill hands you when it runs. Copy it, use it, keep moving.
You don't need to hold any of this in your head. That's the skill's job. If you're curious what it's doing for you, here's the idea underneath it.
The most dangerous moment with any automation is when it works once and you assume it works always. On-ramp closes that gap. It runs your build for 7 days, makes the build log its own performance after every run, and hands you one report with a clear signal. You trust the build because it earned it across a full week of real runs.
A fixed window that turns one lucky run into seven days of evidence. Long enough to catch the quiet failures, short enough to keep momentum.
The build records status, items found and processed, errors, output accuracy, confidence, and a security check after every single run. The evidence writes itself.
On Day 7 you get one message with a green, yellow, or red signal, the checklist, and the numbers. You review it in a minute and reply with one word.
Credential storage and minimum-permission checks are part of the on-ramp from the first run, so a build never scales carrying a security gap.
The agent does the evaluation work. You make the call. That division of labor is what makes this repeatable across every build you ship.
The 7-day on-ramp is a fixed testing window you run right after a new agent, cron job, or automation goes live. It exists because a build that works once has not proven it works reliably. Over seven days the build logs its own performance after every run, so you get real evidence across a full week before you give it more responsibility.
The Accountability Log is where the agent records its own performance after every run. It captures run status, items found, items processed, items skipped, errors, output accuracy, a self-assessed confidence score from 1 to 5, the most valuable item from the run, and a security check. One log supports every agent you on-ramp, so the evidence is consistent across builds.
On Day 7 the agent sends one graduation report with a green, yellow, or red signal, a checklist covering reliability and security, and the numbers from the week. You review it in about a minute and reply with one word. Graduate moves the build to the next phase, Run another week gives it seven more days, and Fix tells the agent what to adjust.