# OpsDesk course workbook Use synthetic inputs. Keep raw receipts, recordings and credentials private. A task checkbox is self-reported. Provider acceptance needs actual independent readback. ## Part 1: From idea to a working app A working OpsDesk app: create/search/filter/select tickets, run one real AI investigation, show honest loading/errors, and restart it from written instructions. - [ ] Define the useful action Expected: Explain the difference between UI, model and agent in your own words. My result / evidence reference: ____________________ - [ ] Build the visible inbox Expected: All four interactions work and the AI state is disconnected. My result / evidence reference: ____________________ - [ ] Connect one agent action Expected: The selected ticket produces an actual model/tool answer in the app. A provider failure is a blocker to fix, not first-hour completion. My result / evidence reference: ____________________ - [ ] Verify the complete user journey Expected: The real app completes ticket creation → selection → AI answer. My result / evidence reference: ____________________ - [ ] Test loading and failure states Expected: No spinner stays stuck and no canned answer masks a failed provider. My result / evidence reference: ____________________ - [ ] Save the working baseline Expected: Another learner can restart this scoped app and reproduce its AI action with their own model. My result / evidence reference: ____________________ ## Part 2: Upgrade your app’s agent engine Your own agent core, a reviewed tool registry and passing behavioral contracts. - [ ] Open a fresh learner workspace Expected: You can identify the model, tool and observer interfaces. My result / evidence reference: ____________________ - [ ] Ask Codex for the smallest loop Expected: The core returns an answer or an explicit terminal failure; it cannot run shell commands. My result / evidence reference: ____________________ - [ ] Make the tools deliberately small Expected: Valid calls reach one handler; invalid calls do not execute a handler. My result / evidence reference: ____________________ - [ ] Walk through an observation Expected: The fixed contracts pass and the original app still completes a real investigation through the new engine. My result / evidence reference: ____________________ ## Part 3: Connect models. Know the cost. One real model run, a capability record and a cost estimate with explicit assumptions. - [ ] Bind one actual model Expected: Configuration names one explicit model, allowed host and output bound. My result / evidence reference: ____________________ - [ ] Prove one native tool round-trip Expected: The second task has an actual native tool call and observation; errors remain visible. My result / evidence reference: ____________________ - [ ] Compare behavior fairly Expected: Your comparison labels the exact task, returned model and known limitations. My result / evidence reference: ____________________ - [ ] Calculate only what is known Expected: A reviewer can reproduce the estimate; local GPU costs are not called zero. My result / evidence reference: ____________________ ## Part 4: Follow every action A trace map, one exact Langfuse readback and a documented app/log correlation path. - [ ] Name the boundaries Expected: The map explains who records, exports and independently queries each signal. My result / evidence reference: ____________________ - [ ] Enable metadata-only OTel Expected: The local receipt has actual span IDs and an exporter result. My result / evidence reference: ____________________ - [ ] Verify the exact trace Expected: The expected session, generation/tool observations and exact trace are found, or the missing checkpoint is named. My result / evidence reference: ____________________ - [ ] Join application and log evidence Expected: Live acceptance requires independent readback from each configured destination. Without OCI, submit a clearly labelled correlation design. My result / evidence reference: ____________________ ## Part 5: Make the safe path the easy path A small threat model, negative-test evidence and a mitigation/retest record. - [ ] Mark the trust boundaries Expected: Each control has a concrete code location and expected failure outcome. My result / evidence reference: ____________________ - [ ] Prove the change stop Expected: The model factory is not constructed for a change intent. My result / evidence reference: ____________________ - [ ] Treat hostile instructions as data Expected: No unregistered handler executes. Preserve the real result even if the model follows a malicious instruction in text. My result / evidence reference: ____________________ - [ ] Break it and explain it Expected: Each case has the expected terminal outcome and no cross-session data exposure. My result / evidence reference: ____________________ ## Part 6: Give your agent a cloud home A release packet, protocol adapter design, rollback plan and explicit acceptance matrix. - [ ] Choose the runtime and protocol Expected: The design names the actual image entrypoint and supported request/response protocol. My result / evidence reference: ____________________ - [ ] Prepare a reviewable deployment Expected: A reviewer can identify the account, region, compartment, resource changes and cleanup plan without exposing private values. My result / evidence reference: ____________________ - [ ] Rehearse acceptance before rollout Expected: Acceptance records distinguish locally tested, configured and provider verified. Without a cloud target, submit the unexecuted runbook. My result / evidence reference: ____________________ - [ ] Hand it to the next operator Expected: A second operator can explain how to start, verify and stop the system without borrowing presenter credentials. My result / evidence reference: ____________________