Clarify TUI action names and keys

Author: Chris Jean

README.md

@@ -73,7 +73,7 @@ cd /path/to/your/project
 crucible
 ```
 
-The TUI loads the selected run archive directly from `.crucible/runs/`, shows run status, leaderboard rows, selected candidate details, and forms for auto run setup, discovery, evaluator generation, competitor generation, adoption, promotion, evaluation, next-round preparation, evolution, reports, index rebuilds, archive queries, and candidate inspection. When a non-baseline candidate has passed, the dashboard selects it by default; otherwise it selects the first non-baseline candidate before falling back to the baseline. The candidate table is the picker for candidate-specific actions, so inspecting the selected candidate runs directly from the table instead of asking for an ID. The new-run form defaults to `run --auto --generate --evaluate`, so typing the optimization request and pressing Enter is enough to create the run, generate and validate an evaluator, record baseline metrics, generate competitors, evaluate them, and return to a leaderboard. The forms execute the same command paths as the shell CLI, show a live spinner with elapsed time while actions run, allow cancellation requests for long-running actions, and return to the dashboard when the action finishes. Press `h` to open the in-session command history, which records the selected options before each generated command plus completion status and captured output byte counts. Canceled evaluation work records an event and marks affected unevaluated candidates as `canceled` without overwriting completed `passed` or `failed` results. Canceled generation also records valid unadopted candidate artifacts as `canceled` and records partial candidate directories in the cancellation event.
+The TUI loads the selected run archive directly from `.crucible/runs/`, shows run status, leaderboard rows, selected candidate details, and uses goal-oriented actions for the normal workflow: `[S] Start tournament`, `[R] Review selected candidate`, `[A] Apply selected candidate`, `[C] Continue tournament`, `[T] Test & score candidates`, `[G] Create more candidates`, `[E] Export report`, `[H] Command history`, and `[?] Advanced`. The selected leaderboard row drives candidate-specific actions, so reviewing or applying a candidate uses the highlighted row instead of asking for an ID. The new-run form defaults to `run --auto --generate --evaluate`, so typing the improvement request and pressing Enter is enough to create the run, generate and validate a test harness, record baseline metrics, create candidates, test and score them, and return to a leaderboard. Lower-frequency actions such as Discovery Only, Create Test Harness, Import Generated Candidates, Create Next Round, Refresh Archive Index, and Browse Archive live in Advanced. The forms execute the same command paths as the shell CLI, show a live spinner with elapsed time while actions run, allow cancellation requests for long-running actions, and return to the dashboard when the action finishes. Press `h` to open the in-session command history, which records the selected options before each generated command plus completion status and captured output byte counts. Canceled evaluation work records an event and marks affected unevaluated candidates as `canceled` without overwriting completed `passed` or `failed` results. Canceled generation also records valid unadopted candidate artifacts as `canceled` and records partial candidate directories in the cancellation event.
 
 `crucible tui` remains an explicit alias for launching the same dashboard with options:
 
@@ -718,6 +718,7 @@ Current priorities:
 - [x] Make prompt-mode candidate inspection use a numbered selector with non-baseline defaults
 - [x] Make bare `crucible` launch the TUI and move the prompt workflow to `crucible prompt`
 - [x] Add a TUI command history screen and return to the dashboard after actions instead of showing raw output
+- [x] Rename TUI actions around user goals and move lower-frequency tools behind Advanced
 
 Deferred roadmap:
 

docs/architecture.md

@@ -36,7 +36,7 @@ crucible run "make this feature faster"
 crucible run --task-file crucible-task.md
 ```
 
-The bare `crucible` command opens the Bubble Tea dashboard for reviewing the latest or selected run status, leaderboard rows, selected candidate details, and forms for auto run setup, discovery, evaluator generation, competitor generation, adoption, promotion, evaluation, next-round preparation, evolution, reports, index rebuilds, archive queries, and candidate inspection. `crucible tui` opens the same dashboard explicitly and supports TUI-specific options such as selecting a run. TUI new-run setup defaults to `run --auto --generate --evaluate`, so entering only the optimization request starts discovery, evaluator generation, baseline validation, competitor generation, candidate evaluation, and dashboard refresh. TUI actions return to the dashboard after completion, and the `h` key opens an in-session command history that records the selected form options before each generated command, along with status, duration, captured output byte counts, and cancellation artifacts. The older line-oriented guided workflow remains available as `crucible prompt`; it confirms the project directory, initializes `.crucible/` when needed, creates a discovery plan for new optimization requests, optionally runs discovery through a configured agent, records clarifying answers, creates runs, and offers common follow-up actions for existing run data. Prompt-mode run creation can opt into advanced external/round/exploration options, run actions auto-select the only available run and prompt for a run selector only when multiple runs exist, standalone discovery can run locally or through a discovery-capable agent, evaluators can be generated for existing runs, generated candidates can be adopted, passed candidates can be promoted back to the host project, inspect can select candidates from a numbered leaderboard-backed list, next rounds can be prepared, archive queries can be run, reports can choose output/JSON options, index rebuilds can scope to one run or print JSON, generate can review or edit interface docs, evaluator scaffolds, generation prompts, and discovery handoffs through `$VISUAL` or `$EDITOR`, generate can opt into advanced model/output/dry-run and Codex profile/sandbox/approval options, evaluate can opt into candidate/resource/sandbox options, evaluator/generate/evolve actions can select a generation-capable agent, and agent settings can update the project `default_agent`. `crucible provider template` prints non-mutating config fragments for command providers such as Claude so users can inspect provider setup before editing `.crucible/config.json`. When a structured agent discovery handoff is available, the wizard can use the recommended source path and copies the handoff into the run's `docs/` directory. The project receives a `.crucible/` directory automatically when the first run or discovery plan is created. This keeps optimization artifacts close to the code being evaluated without requiring the host project to adopt Code Crucible as a dependency. `crucible init` remains available for explicit preflight setup or a custom project name.
+The bare `crucible` command opens the Bubble Tea dashboard for reviewing the latest or selected run status, leaderboard rows, selected candidate details, and goal-oriented forms for starting tournaments, reviewing candidates, applying candidates, continuing tournaments, testing and scoring candidates, creating more candidates, exporting reports, and reviewing command history. `crucible tui` opens the same dashboard explicitly and supports TUI-specific options such as selecting a run. TUI new-run setup defaults to `run --auto --generate --evaluate`, so entering only the improvement request starts discovery, test-harness generation, baseline validation, candidate creation, candidate testing/scoring, and dashboard refresh. Lower-frequency tools such as Discovery Only, Create Test Harness, Import Generated Candidates, Create Next Round, Refresh Archive Index, and Browse Archive live behind the TUI Advanced menu. TUI actions return to the dashboard after completion, and the `h` key opens an in-session command history that records the selected form options before each generated command, along with status, duration, captured output byte counts, and cancellation artifacts. The older line-oriented guided workflow remains available as `crucible prompt`; it confirms the project directory, initializes `.crucible/` when needed, creates a discovery plan for new optimization requests, optionally runs discovery through a configured agent, records clarifying answers, creates runs, and offers common follow-up actions for existing run data. Prompt-mode run creation can opt into advanced external/round/exploration options, run actions auto-select the only available run and prompt for a run selector only when multiple runs exist, standalone discovery can run locally or through a discovery-capable agent, evaluators can be generated for existing runs, generated candidates can be adopted, passed candidates can be promoted back to the host project, inspect can select candidates from a numbered leaderboard-backed list, next rounds can be prepared, archive queries can be run, reports can choose output/JSON options, index rebuilds can scope to one run or print JSON, generate can review or edit interface docs, evaluator scaffolds, generation prompts, and discovery handoffs through `$VISUAL` or `$EDITOR`, generate can opt into advanced model/output/dry-run and Codex profile/sandbox/approval options, evaluate can opt into candidate/resource/sandbox options, evaluator/generate/evolve actions can select a generation-capable agent, and agent settings can update the project `default_agent`. `crucible provider template` prints non-mutating config fragments for command providers such as Claude so users can inspect provider setup before editing `.crucible/config.json`. When a structured agent discovery handoff is available, the wizard can use the recommended source path and copies the handoff into the run's `docs/` directory. The project receives a `.crucible/` directory automatically when the first run or discovery plan is created. This keeps optimization artifacts close to the code being evaluated without requiring the host project to adopt Code Crucible as a dependency. `crucible init` remains available for explicit preflight setup or a custom project name.
 
 ## Discovery Archive
 

docs/tui.md

@@ -19,8 +19,8 @@ Code Crucible uses Bubble Tea v1.2.x and Bubbles v0.20.x because that pair suppo
 
 - Launch the TUI from bare `crucible`, keep `crucible tui` as an explicit alias, and preserve the line-oriented prompt workflow as `crucible prompt`.
 - Use the existing `internal/cli.WorkflowController` from prompt mode and the TUI so both interfaces share command construction and execution.
-- The first implementation includes a run dashboard: latest or selected run status, leaderboard rows, selected candidate detail, and common next actions. The dashboard defaults to a passed non-baseline candidate when one exists, and the candidate table is the picker for candidate-specific actions such as inspect. Actions that have TUI forms are shown as key-first prompts; workflows that are not yet implemented in the TUI use short CLI hints instead of long absolute commands.
-- Forms cover low-prompt auto run setup, discovery, evaluator generation, generation, adoption, promotion, evaluation, next-round preparation, evolution, reports, index rebuilds, archive queries, and inspection. The new-run form defaults to `crucible run --auto --generate --evaluate`, so the operator can type only the optimization request and press Enter to create the run, generate and validate an evaluator, record baseline metrics, generate competitors, evaluate them, and refresh the dashboard. Forms use Bubbles text inputs, execute through the shared CLI controller, show a live spinner with elapsed time while actions run, allow cancellation requests for long-running actions, and return to the dashboard after completion instead of forcing a raw output screen.
+- The first implementation includes a run dashboard: latest or selected run status, leaderboard rows, selected candidate detail, and common next actions. The dashboard defaults to a passed non-baseline candidate when one exists, and the candidate table is the picker for candidate-specific actions such as review and apply. Top-level actions use mnemonic goal labels: `[S] Start tournament`, `[R] Review selected candidate`, `[A] Apply selected candidate`, `[C] Continue tournament`, `[T] Test & score candidates`, `[G] Create more candidates`, `[E] Export report`, `[H] Command history`, and `[?] Advanced`.
+- Forms cover low-prompt tournament setup, discovery, test-harness generation, candidate creation, generated-candidate import, selected-candidate application, candidate testing/scoring, next-round preparation, tournament continuation, reports, index rebuilds, archive browsing, and candidate review. The new-run form defaults to `crucible run --auto --generate --evaluate`, so the operator can type only the improvement request and press Enter to create the run, generate and validate a test harness, record baseline metrics, create candidates, test and score them, and refresh the dashboard. Lower-frequency actions such as Discovery Only, Create Test Harness, Import Generated Candidates, Create Next Round, Refresh Archive Index, and Browse Archive live behind the Advanced menu. Forms use Bubbles text inputs, execute through the shared CLI controller, show a live spinner with elapsed time while actions run, allow cancellation requests for long-running actions, and return to the dashboard after completion instead of forcing a raw output screen.
 - Press `h` from the dashboard to open the in-session command history. Each entry lists the selected form options before the generated CLI command, followed by completion status, duration, captured output byte counts, and cancellation artifact pointers when applicable.
 - Cancellation is conservative: Code Crucible records a cancellation event and marks affected unevaluated evaluation candidates as `canceled`, but does not overwrite completed `passed` or `failed` results. Canceled generation records valid unadopted candidate artifacts as `canceled` and records partial candidate directories in the same event stream.
 - Candidate selection uses the Bubbles table widget, and candidate details use a scrollable viewport.