Workshop worksheet
Keep enough evidence to explain your result without copying your entire history.
A small record for each ticket
Copy this once per ticket into your own notes, then use the exercise records below in workshop order:
| Field | Your observation |
|---|---|
| Ticket and acceptance criteria | |
| App session, source directory and handoff branch | |
| Mode, model and effort | |
| Repository/file-access and MCP approvals, denials or unresolved permissions | |
| Skill loaded and guidance used | |
| Generated test assertions reviewed before implementation | |
| Initial failing tests and assertion messages | |
| Reviewed production/test/UI changes | |
| Test command and results (tests, failures, errors, skipped) | |
| Commit and preview outcome | |
| PR link, actual Merged status, verified main commit and merged-revision checks | |
| Worked alone, with a partner, or from a demonstration/reference? | |
| Unfinished work or results not yet checked |
Do not include API tokens, passwords, cookies or unrelated personal work. If you use a reference solution or watch a demonstration, note which steps you also ran yourself.
Three task handoffs
| Session | Reviewed PR | Verified main commit and checks |
|---|---|---|
| Cart: quantity bug | ||
| Cart: item removal, including Canvas | ||
| Cart: query optimization |
Merge and verify each PR before creating the next session from updated main.
Name each session only after sending its first prompt.
Canvas: one observed improvement
| Field | Your observation |
|---|---|
| Active session/directory and server owner | |
| Browser canvas URL and selected element | |
| Observed UI issue and accepted scope | |
| impeccable guidance actually used, or unavailable | |
| Before/after keyboard and narrow-screen observations | |
| Regression results and reviewed feature handoff | |
| Native picker used, browser fallback or demonstration |
MCP before and after
| Field | Before | After |
|---|---|---|
| Reviewed source directory/branch | ||
Workload: eight distinct products, GET /api/cart |
||
| Request ID, method/path and time | ||
| Tool names and actual arguments | ||
| Observed SELECT count | ||
| Product fields, count and total preserved? | ||
| Source: MCP call in the App, saved reference or demonstration |
Wait for DevTools to load the fix, then capture a fresh request profile. Reusing the baseline profile does not show whether the optimization worked.
Also record query-contract test results for 8 and 16 distinct products:
exactly 1 SELECT per completed HTTP GET /api/cart, excluding setup queries
and preserving response values. The eight-product MCP capture is separate evidence.
Parallel handoff
Keep all three task PR links and verified merge results. The MCP handoff PR
establishes the published origin/main commit used to start both workers.
Then keep each worker’s handoff: its branch/commit, changed files, executed checks
and limitations. Both must start from that same updated main commit; let
Copilot capture the commit details. If the MCP handoff used local recovery, record
the accepted local starting branch/commit and main not updated: local recovery used.
Only record a preview or runtime result if that worker ran the
matching application. Mark integration checks pending until you test the
merged result.
If a boundary case exposed a bug, record the failing regression, approved fix and passing rerun in that session. For overlapping files, record any conflict resolutions and checks that both contributions still work; no conflict is also a valid outcome.
Agent Merge and integration
| Field | Your observation |
|---|---|
Own repository and both worker PRs targeting main |
|
| UI PR link, linked in Cart: empty-state UX, head and CI results | |
| UI PR merge authorization and actual status | |
| UI merged-main revision and its passing tests CI job | |
| Boundary-test PR link, linked in Cart: boundary tests, head/updated commit and fresh CI results | |
| Boundary-test PR merge authorization and actual status | |
| Agent Merge corrections or conflicts reviewed | |
| Merged main commit checked in Cart: boundary tests before checkout, both contributions and full-suite results | |
| Combined preview from the verified directory: quantities, removal, empty-cart link, keyboard and narrow screen | |
| Recovery used, open PRs and unfinished work |
Keep the PRs sequential. A running agent, local test pass or enabled merge permission is not confirmation that GitHub merged the PR. If using local recovery, record Agent Merge not completed: local recovery used.
Rubber duck: critique and decision
| Field | Your observation |
|---|---|
| Checked result and comparison starting point | |
| Actual critic invocation/model if exposed, or unavailable | |
| Finding, code/evidence and impact | |
| Accepted, rejected or deferred, with reason | |
| Follow-up branch/PR, merge status and verified main result if corrected |
Five subagents: mock-checkout implementation
| Field | Your observation |
|---|---|
| Shared goal, reviewed base and checkout feature branch | |
| Cart: mock checkout session, new worktree directory and verified starting commit | |
| Approved HTTP contract, test setup/probes and exclusive file/fixture ownership | |
| Common starting commit verified in all five child worktrees | |
| Five actual assignments/contributions; concurrency observed | |
| Load check, five-way concurrency approval and observed execution overlap or limitation | |
| Reviewer’s findings, reviewed PR head at merge authorization and fix/defer decisions | |
| Later Agent Merge repair commits, checks and review coverage or limitation | |
| Questions/answers exchanged, directly or through the orchestrator | |
| Implemented API, order snapshot, atomic clearing and duplicate handling | |
| Conflicting proposals resolved or decisions still open | |
| Full-suite counts, including baseline, generated cart/removal/query, boundary, CheckoutContractTest and CheckoutResilienceTest | |
| Rollback, snapshot and concurrent/later-retry evidence | |
| Browser canvas receipt, cleared cart, keyboard and narrow-screen observations | |
| Final diff accepted, checkout publication approved, Agent Merge repairs reviewed and merge separately authorized | |
Checkout PR URL, repository, feature head and base main checked |
|
| Actual merge status/commit and checks for that merged revision | |
| Verified main commit, local full-suite results and running directory/URL | |
| Place order and receipt verified with the app serving the verified merged-main code | |
| Delegation or messaging unavailable, or demonstration |
Record four writers’ local handoffs and the read-only review:
| Child session | Session link, worktree and branch | Integrated handoff commit or reviewed commit/findings |
|---|---|---|
| Checkout: backend | ||
| Checkout: UI | ||
| Checkout: HTTP tests | ||
| Checkout: resilience tests | ||
| Checkout: review | No code commit; record findings |
Mark checkout merged and verified on main, unfinished, or demonstrated
according to the evidence. Use verified locally; delivery blocked or
merged; post-merge verification blocked when appropriate.
Record actual contributions. Write none exchanged if there were no questions/answers, and
unknown for concurrency you did not observe. Local checkout code is not on
main until its checkout PR actually merges. Keep the approved plan and agent
activity available for Chronicle.
Chronicle: one better habit
| Field | Your observation |
|---|---|
| Privacy choice and scope controls checked before running | |
| Cart: Chronicle analysis session, excluded from source sessions | |
| Current repository, reported seven-day range/timezone and actual source sessions | |
| Returned scope and coverage | |
| Reported input/output/cache/usage values with units, or unavailable | |
| Observed turnaround/rework and quality evidence, or unavailable | |
| Token/time/quality trade-off and supporting session/example | |
| One prompt/workflow change | |
| Suggested repository instruction (proposal only, not applied) | |
| Expected token/time/quality benefit (prediction until measured) | |
| Acceptance and review checks retained | |
| Result observed, or improvement not yet checked |
There is no required savings percentage. Different tasks are not controlled trials. Do not convert a usage multiplier or byte count into dollars/tokens. Keep your full Chronicle output private; share only the relevant, reviewed excerpt.
Final review and handoff
Return to Cart: mock checkout for the read-only final review. Record the verified result separately from unfinished local work.
| Field | Your observation |
|---|---|
| Read-only review session/worktree and verified result commit | |
| Comparison starting point and result commit, reported by Copilot | |
| Local edits/untracked files, excluded from the verified result | |
| Final full-suite counts and per-class reports, including both checkout suites | |
| Findings/decisions and separately approved correction PRs | |
| Actual PR/CI status, verified revision and running directory/URL | |
| Completed, demonstrated, unfinished or blocked status, with reason | |
| Handoff: what changed, checked, still unknown and one reusable habit |
Keep the handoff concise; do not attach private Chronicle output or unrelated history.