Capstone Project Selection Criteria
Evaluate your top 3 automation candidates for the Day 3 capstone.
Rate each criterion: Yes (fully meets), Partial (meets with workarounds), or No (does not meet).
| Criterion | Candidate 1: ___ | Candidate 2: ___ | Candidate 3: ___ |
|---|---|---|---|
| Decomposable into discrete steps? | Yes / Partial / No | Yes / Partial / No | Yes / Partial / No |
| Tools accessible via CLI/API? | Yes / Partial / No | Yes / Partial / No | Yes / Partial / No |
| Clear success/failure criteria? | Yes / Partial / No | Yes / Partial / No | Yes / Partial / No |
| Safe with approval gates? | Yes / Partial / No | Yes / Partial / No | Yes / Partial / No |
| Testable with mock data? | Yes / Partial / No | Yes / Partial / No | Yes / Partial / No |
Scoring: 3 pts for Yes, 1 pt for Partial, 0 pts for No. Max = 15.
| Candidate 1 | Candidate 2 | Candidate 3 | |
|---|---|---|---|
| Total (out of 15) |
Selected Capstone Task
Task name: _______________
Score on selection criteria: ___ / 15
Why this task: (3 sentences — what the task is, why it scores well, and what the agent would deliver)
Agent Design (Draft)
Tools the agent would need:
- Tool 1: _______ (e.g.,
aws cloudwatch describe-alarms) - Tool 2: _______ (e.g.,
git log) - Tool 3: _______ (e.g., Jira API - create issue)
Domain context the agent would need (future SKILL.md):
- Runbook: _______
- Infrastructure topology: _______
- Decision criteria: _______
Governance — what requires human approval:
Mock data needed for lab testing:
Criteria Reference
| Criterion | Good Example | Bad Example |
|---|---|---|
| Decomposable | "Triage alert: read alarm → check deployments → follow runbook → create ticket" | "Use good judgment about the infrastructure" |
| Tool accessible | AWS CLI, kubectl, git, REST API | Data locked in GUI-only tool with no API |
| Clear success/failure | "Alert is resolved / no new SEV-1 created" | "Infrastructure is healthy" |
| Safe with approval gates | "Agent drafts rollback command, engineer approves" | "Agent has unrestricted production access" |
| Testable with mock data | CloudWatch JSON fixtures, mock kubectl output | Requires live prod queries to function at all |