Project Review

In The Project Brief, you set the scope and goals for your build. In the build lessons through Prove Professional Value, you created a working AI employee for your profession. Now you evaluate what you actually produced.

This assessment is not a test — it is an honest inventory. The checklist below tells you exactly what a completed build looks like at each tier. Check what you finished, note what you did not, and decide what comes next.

Submission Checklist

Review your nanoclaw-employee/ repository against the criteria for your tier. Check each item you completed.

Bronze Tier

• groups/main/CLAUDE.md with profession-specific identity — not a generic assistant, but an employee that knows your domain vocabulary, common tasks, and professional standards (Give Your Employee an Identity)
• Custom SKILL.md in .claude/skills/ with 5 or more domain decision rules — rules that encode how a professional in your field makes judgment calls, not just task instructions (Teach Your Employee a Skill)
• Working channel or MCP connection documented — your employee can communicate through at least one external channel beyond the terminal (Connect Your Employee to the World)
• conversation-log.md with 3 to 5 real professional tasks — actual work you delegated, not toy examples, with the employee's responses and your assessment of quality (Bronze Capstone: First Real Day)
• evaluation.md with completed rubric and honest reflection — did the employee perform at the level of a junior colleague, an intern, or not yet useful? (Bronze Capstone: First Real Day)

Silver Tier (All Bronze items, plus)

• scheduler-config.md with scheduled task design — what runs, when, and why that cadence matters for your profession (Make Your Employee Proactive)
• hitl-boundaries.md with 4 or more categorized actions — each with domain-specific reasoning for why it is auto-approve, needs-approval, or never-automate (Teach Your Employee Boundaries)
• memory-config.md with action log and knowledge store schemas — working SQLite tables that log autonomous actions and persist corrections from real interactions (Give Your Employee a Memory)
• domain-report-sample.md — an actual report your employee generated autonomously, pulling from 2 or more data sources, containing at least one proactive recommendation (Prove Professional Value)

Gold Tier (All Silver items, plus)

• 3 groups configured with distinct CLAUDE.md files — main (admin), a professional work group, and a client-facing or external group, each with different identity and permissions (Prove Professional Value)
• Isolation test results documented — evidence that the non-admin group cannot perform admin actions or access restricted data (Prove Professional Value)
• System architecture diagram — showing the three groups, their data access boundaries, and communication flows between them (Prove Professional Value)
• Memory isolation demonstrated — each group has independent action logs and knowledge stores; non-admin groups cannot access admin memory (Give Your Employee a Memory + Prove Professional Value)

Reflection

Pull up your Layer 3 design from the "NanoClaw Hands-On Setup" lesson in "Meet Your First AI Employee - OpenClaw". That design listed the skills you planned to build and the MCP servers you planned to connect. Compare it with what you actually built.

Consider these questions:

How much of your Layer 3 design did you implement? Count the skills you designed versus the skills you built. Count the MCP connections you planned versus the ones that work. The gap between plan and execution is normal — the question is whether you understand why the gap exists.

What would take your employee from its current tier to the next? If you completed Bronze, what specific work would get you to Silver? If you completed Silver, what is the hardest part of Gold? Be concrete — "add scheduling" is vague, "build a weekly cash flow report skill that reads bank CSV and invoice spreadsheet" is actionable.

How did your employee's behavior change after corrections? Review your knowledge store entries. Did the employee behave differently the next time a relevant task came up? For Gold students: does your memory isolation actually hold, or can groups leak knowledge to each other?

What surprised you about encoding your professional expertise into an AI system? Most people discover that the hardest part is not the technology — it is articulating the judgment calls they make unconsciously every day. What did you know that you did not know you knew?