Complete example programs

Start with the worked tasks, not with copying the entire directory into an application. These programs demonstrate the decisions in the skill through concrete code and regression tests. They have no runtime network calls or credentials; search and log events are simulated locally.

Try one

From this directory, with a Go toolchain compatible with go.mod:

go mod tidy
go test -race ./...
go vet ./...
go run ./cmd/picker

Run go run ./cmd/search or go run ./cmd/tail separately for the other demos. Resolve dependencies once; go.sum is intentionally not fabricated. Transitive dependency checksums were unavailable in the authoring environment.

Picker. An inline selector. Up/Down moves; / enters search. Escape/Enter from search returns to the list, keeping the query. Enter opens confirmation; Enter confirms and emits the captured ID on stdout. Escape from confirmation returns to the list. q quits only in browse mode. Ctrl+C cancels anywhere. --list and --id prod-eu never open a terminal. Interactive mode permits stdout redirection but requires terminal stdin/stderr; it does not read candidate data from stdin.

Search. An inline text input with a 150 ms cancellable debounce and injected backend. slow takes longer, error returns a recoverable failure, and queue produces a match. Ctrl+R retries with fresh request identity; Escape/Ctrl+C exits. The whole request lifetime, including debounce, has a three-second timeout. Results clear on edit rather than being displayed as stale. A result list larger than the visible area would need navigation/paging before adopting this demo for such a dataset.

Tail. A full-screen log monitor with a bounded 16-record transport queue and a 32-record display ring. Up/Down or j/k scrolls; End follows latest; q/Ctrl+C quits. A finite synthetic stream allows testing closure. Retention eviction is disclosed. It is not a persistent logger or a multi-line transcript renderer.

Where the behavior lives

Tests and what has actually run

The exact core files used by these demos have 26 executed tests across four groups. This includes genuine context/timer/channel cancellation tests, not a mock of Bubble Tea. The source for all Go files was parsed/formatted by gofmt. Full module compilation was not possible here: Go 1.23.2 is installed, the module requests a newer toolchain, and network dependency resolution is blocked. Additional Charm model/render tests are supplied but remain unexecuted.

From the skill root:

python3 scripts/test_offline_examples.py

That script compiles the named dependency-free files directly with module mode disabled. It deliberately does not compile model.go, render.go, or the CLI entry points. This is not a replacement for go test -race ./... after dependency resolution. See verification and the raw core test output.

No terminal/PTY session, Unicode display inspection, agent evaluation, or target-project before/after comparison was executed during authoring. Do those before claiming an adapted UI is polished or behavior-equivalent.