CAMEL-23978: Add AI panel slash commands to Camel TUI

[ammachado]

Description

Adds local slash commands to the Camel TUI AI panel so operators can control the panel and invoke Camel CLI actions without sending prompts to the configured AI provider.

Highlights

• Built-in panel commands: /help, /provider, /model, /clear, /close, /quit (with /exit, /q, /x aliases).
• CLI bridge commands: /run, /infra, and /send run through a TUI-safe executor with output capture and cancellation.
• Provider switching and persisted model selection, with live command hints and TAB completion.
• /run and /infra run launch as tracked background processes; /infra list remains in-process.
• Fixed conversation auto-scrolling by using the exact rendered Markdown height.
• User manual documents the supported slash commands.

Review follow-up

• Identifies an Ollama endpoint by its root GET response (Ollama is running), so detection no longer relies on port 11434 and does not classify arbitrary reachable URLs as Ollama.
• Serializes all in-process CLI execution used by slash commands and the AI cli_exec tool to protect the shared Camel JBang/picocli instance.
• Keeps the AI thinking state independent from background slash-command completion, and makes agent cancellation non-blocking for the TUI event thread.
• Blocks /provider and /model while either an AI request or background slash CLI command is active.

Tracking

https://issues.apache.org/jira/browse/CAMEL-23978

Verification

• mvn -Dtest=LlmClientListModelsTest test in dsl/camel-jbang/camel-jbang-core
• mvn -Dtest=AskCliToolsTest,LlmClientListModelsTest test in dsl/camel-jbang/camel-jbang-core
• mvn -Dtest=AiPanelTest test in dsl/camel-jbang/camel-jbang-plugin-tui
• mvn -Dtest=AiCliCommandExecutorTest,AiPanelTest test in dsl/camel-jbang/camel-jbang-plugin-tui

AI-generated by Codex on behalf of ammachado. Commits include Co-authored-by: Codex <noreply@openai.com>.

[github-actions]

🌟 Thank you for your contribution to the Apache Camel project! 🌟
🤖 CI automation will test this PR automatically.

🐫 Apache Camel Committers, please review the following items:

• First-time contributors require MANUAL approval for the GitHub Actions to run
• You can use the command /component-test (camel-)component-name1 (camel-)component-name2.. to request a test from the test bot although they are normally detected and executed by CI.
• You can label PRs using skip-tests and test-dependents to fine-tune the checks executed by this PR.
• Build and test logs are available in the summary page. Only Apache Camel committers have access to the summary.

⚠️ Be careful when sharing logs. Review their contents before sharing them publicly.

[github-actions]

🧪 CI tested the following changed modules:

• docs
• dsl/camel-jbang/camel-jbang-core
• dsl/camel-jbang/camel-jbang-plugin-tui

---

🔬 Scalpel shadow comparison — Scalpel: 8 tested, 8 compile-only — current: 6 all tested

Maveniverse Scalpel detected 16 affected modules (current approach: 6).

⚠️ Modules only in Scalpel (10)

• camel-jbang-core
• camel-jbang-it
• camel-jbang-main
• camel-jbang-plugin-edit
• camel-jbang-plugin-generate
• camel-jbang-plugin-kubernetes
• camel-jbang-plugin-test
• camel-launcher
• coverage
• docs

Skip-tests mode would test 8 modules (3 direct + 5 downstream), skip tests for 8 (generated code, meta-modules)

Modules Scalpel would test (8)

• camel-jbang-core
• camel-jbang-mcp
• camel-jbang-plugin-mcp
• camel-jbang-plugin-route-parser
• camel-jbang-plugin-tui
• camel-jbang-plugin-validate
• camel-launcher-container
• docs

Modules with tests skipped (8)

• camel-jbang-it
• camel-jbang-main
• camel-jbang-plugin-edit
• camel-jbang-plugin-generate
• camel-jbang-plugin-kubernetes
• camel-jbang-plugin-test
• camel-launcher
• coverage

ℹ️ Shadow mode — Scalpel observes but does not affect test execution. Learn more

⚠️ Some tests are disabled on GitHub Actions (@DisabledIfSystemProperty(named = "ci.env.name")) and require manual verification:

• dsl/camel-jbang/camel-jbang-core: 1 test(s) disabled on GitHub Actions

💡 Manual integration tests recommended:

You modified dsl/camel-jbang/camel-jbang-core. The related integration tests in dsl/camel-jbang/camel-jbang-it are excluded from CI. Consider running them manually:

mvn verify -f dsl/camel-jbang/camel-jbang-it -Djbang-it-test

All tested modules (16 modules)

• Camel :: Coverage
• Camel :: Docs
• Camel :: JBang :: Core
• Camel :: JBang :: Integration tests
• Camel :: JBang :: MCP
• Camel :: JBang :: Main
• Camel :: JBang :: Plugin :: Edit
• Camel :: JBang :: Plugin :: Generate
• Camel :: JBang :: Plugin :: Kubernetes
• Camel :: JBang :: Plugin :: MCP
• Camel :: JBang :: Plugin :: Route Parser
• Camel :: JBang :: Plugin :: TUI
• Camel :: JBang :: Plugin :: Testing
• Camel :: JBang :: Plugin :: Validate
• Camel :: Launcher
• Camel :: Launcher :: Container

---

⚙️ View full build and test results

[davsclaus]

Review: CAMEL-23978 — AI panel slash commands

Well-structured feature PR. Clean separation of concerns (AiSlashCommandRegistry, AiSlashCommandContext, AiCliCommandExecutor), thorough test coverage (concurrency edge cases, parsing, error paths, rendering), and good documentation update.

Positive highlights:

• Migrating ConversationEntry.role from String to the new AiRole enum catches mismatches at compile time.
• The settings persistence integrates naturally with the existing TuiSettings/SettingsPopup pattern.
• Architecture is independently testable per component.

Findings (inline comments below):

1. thinkingVerb field data race — easy fix with volatile.
2. tryExplicitUrl() behavioral change — worth noting.
3. Minor scope: Ask.java formatter change mixed in.

This review evaluates the PR against the project's contribution rules and conventions. It does not replace specialized review tools (CodeRabbit, SonarCloud) or static analysis.

This review was generated by an AI agent and may contain inaccuracies. Please verify all suggestions before applying.

Claude Code on behalf of davsclaus

[gnodet]

Good feature addition — clean architecture, thorough test coverage, and docs updated.

One finding:

AiSlashCommandRegistryTest.java — the NoopSlashContext inner class uses java.util.concurrent.CompletableFuture as a fully qualified class name inline (2 occurrences) instead of importing it. This violates the project coding convention (no FQCNs — always use imports).

Reviewed with Claude Code on behalf of gnodet. This review was generated by an AI agent and may contain inaccuracies; please verify all suggestions before applying.

[ammachado]

@gnodet Fixed in ae38606 — AiSlashCommandRegistryTest now imports java.util.concurrent.CompletableFuture and uses the simple name in NoopSlashContext (both occurrences), per the no-FQCN coding convention.

Claude Code on behalf of ammachado

[gnodet]

Claude Code review on behalf of @gnodet

Re-review after 2 new commits. Previous FQCN feedback addressed ✅. One new issue found.

Feedback addressed

• ✅ CompletableFuture is now properly imported in AiSlashCommandRegistryTest
• ✅ thinkingVerb marked volatile for thread safety between UI and callback threads
• ✅ Unrelated printErr line wrap reverted

New issue: duplicate placeholder block in AiPanel.renderInput()

The placeholder rendering code is duplicated — the same block appears twice consecutively:

if (cursorPos == text.length()) {
    Optional<String> placeholder = slashCommands.placeholderFor(text);
    if (placeholder.isPresent()) {
        spans.add(Span.styled(" " + placeholder.get(), Style.EMPTY.dim()));
    }
}
if (cursorPos == text.length()) {
    Optional<String> placeholder = slashCommands.placeholderFor(text);
    if (placeholder.isPresent()) {
        spans.add(Span.styled(" " + placeholder.get(), Style.EMPTY.dim()));
    }
}

This will render the placeholder hint text twice in the input line (e.g., /run would show <files...> [--dev] [--port=8080] [...] appended twice). One of these blocks should be removed.

Slash command placeholder feature

The placeholderFor() implementation and test coverage look good — clean design with proper Optional handling.

[ammachado]

@gnodet Fixed in eebbb7b — removed the duplicated placeholder block in AiPanel.renderInput(); the hint now renders once. I also strengthened commandPlaceholderRendersAfterTrailingSpace to assert the placeholder appears exactly once (the previous contains() assertion passed even with the duplicate). Verified the strengthened test fails on the duplicated code and passes after the fix.

Claude Code on behalf of ammachado

[gnodet]

Claude Code review on behalf of @gnodet

Duplicate placeholder fix verified ✅

• Duplicate placeholderFor block removed from renderInput()
• Test commandPlaceholderRendersAfterTrailingSpace now asserts the placeholder renders exactly once (via countOccurrences), guarding against regression

No remaining concerns from my reviews. LGTM.

[davsclaus]

LGTM — well-executed feature addition.

Clean architecture with proper separation (AiSlashCommandRegistry, AiSlashCommandContext, AiCliCommandExecutor, AiProviderSelector), introduction of AiRole enum replacing error-prone string role identifiers, and comprehensive test coverage across all new classes.

All previous review findings from @davsclaus and @gnodet have been addressed in follow-up commits. No unresolved conversations remain.

Observations (non-blocking):

1. Branch naming: feature/ai-slash-commands — project convention is feature/<ISSUE_ID>-<short-slug> (i.e. feature/CAMEL-23978-ai-slash-commands). Minor, for future reference.
2. tryExplicitUrl() side-effect: Now probes /api/tags and /v1/models before bare-URL fallback, setting apiType as a side-effect. Custom proxy/gateway URLs that don't serve those sub-paths get two extra probe round-trips. Documented in commit message — worth monitoring if users report slow startup with custom AI URLs.
3. extractStringList() unchecked cast in LlmClient.listModels(): (JsonArray) response.get(arrayField) is fail-safe (outer caller handles null returns) but silently swallows unexpected API response shapes.

Ready to merge (squash).

Claude Code on behalf of davsclaus

This review was generated by an AI agent and may contain inaccuracies. Please verify all suggestions before applying.

[ammachado]

PR is back to draft since I've found a few issues testing.

[davsclaus]

okay was about to, let us know when its ready next time

[gnodet]

Claude Code review on behalf of @gnodet

Re-review: 3 new commits since last review

Commit ac20f92c — Launch /run and /infra as tracked background processes: Well-designed. /run and /infra run now go through LaunchManager (same path as F2 Actions), so they appear as tracked background processes instead of blocking the panel in-process. Key strengths:

• LaunchSpec record with List.copyOf() for immutability
• Infra auto-start before example launches (with findMissingInfraServices check)
• launchDetachedQuietly() gracefully reports launch failures to the conversation
• /infra list correctly stays in-process while /infra run detaches — isInfraRun() check is clean
• Comprehensive test coverage: RecordingLaunchContext, LaunchSpec.forRun() / forInfra() tests

Minor: exampleCatalog volatile field has a benign data race (two threads may load the catalog simultaneously) since there's no synchronization around the read-then-write. Harmless because ExampleHelper.loadCatalog() is idempotent and reference assignment is atomic, but worth noting.

Commit 822be6d5 — Persist /model selection: Clean fix. persistModelSelection() correctly updates both TuiSettings (disk) and sessionProviderChoice (memory), so neither a restart nor a panel re-init reverts the model. Test uses @TempDir with proper finally cleanup of CommandLineHelper.useHomeDir.

Commit 427d2b3c — TAB completion and clipped output fix: Two independent improvements:

1. TAB completion: Nice state machine — longest common prefix first, then cycle, Shift+TAB for backward, any edit resets. The completionSnapshot comparison for detecting edits is clean and avoids explicit reset logic in every key handler.
2. Clipped output fix: Replacing character-count estimation with MarkdownView.computeHeight() is the right fix — the old approach under-counted lines for markdown block elements (headings, lists, fenced code, blockquotes) that render with surrounding blank rows. Re-measuring at the narrower width after adding the scrollbar column is a good attention to detail.

One minor nit: The longResponseAutoScrollsToShowLastLine test uses Thread.onSpinWait() in a busy-wait loop. While onSpinWait() isn't Thread.sleep(), Awaitility would be more idiomatic:

await().atMost(5, TimeUnit.SECONDS).until(() -> !panel.isAgentThreadRunningForTesting());

Approved — all three commits are well-implemented with thorough tests.

[gnodet]

Claude Code review on behalf of @gnodet

Re-review: New commit ba40132d — Harden AI model selection

Three clean changes:

1. LlmClient.extractStringList(): Replaced unchecked cast (JsonArray) response.get(arrayField) with instanceof pattern matching. Prevents ClassCastException when a provider returns an unexpected shape (e.g., {"models":{}} instead of {"models":[]}). Good defensive fix.

2. New test returnsEmptyListWhenModelListHasUnexpectedShape(): Covers the exact case — object instead of array. ✓

3. AiPanel.initClient(): Removed hardcoded .withModel("llama3.2"). Correct — a hardcoded Ollama-specific default doesn't belong in a multi-provider context. The model now comes from sessionProviderChoice or the provider's own default.

LGTM — still approved.

[davsclaus]

Nice work on this PR — the slash command architecture (registry, executor, context) is clean, the TAB completion is a thoughtful UX touch, and the test coverage across 7 new test files is excellent. A few observations below.

This review does not replace specialized tools such as CodeRabbit, Sourcery, or SonarCloud for deeper static analysis.

This review was generated by an AI agent and may contain inaccuracies. Please verify all suggestions before applying.

Claude Code on behalf of davsclaus

[gnodet]

Claude Code review on behalf of @gnodet

Re-review: New commit d97d48fd — Address latest AI panel review feedback

Four well-motivated changes:

1. Non-blocking cancellation (stopAgentThread()): Replaced the blocking t.join(30_000) on the UI thread with a daemon watcher thread. This prevents the TUI from freezing for up to 30s when an HTTP client ignores interruption. The agent thread's own finally block clears the thinking state, so removing the redundant thinking.set(false) from stopAgentThread() is correct.

2. CLI/LLM state separation: Removed thinking.set(false) from handleCliCompletion(). Since CLI commands (like /send) no longer put the panel into thinking state (previous commit), clearing it on completion was wrong — it would accidentally kill an active LLM request's thinking indicator. The new test cliCompletionDoesNotClearLlmThinkingState is the exact regression test for this.

3. Awaitility migration: All Thread.onSpinWait() busy-wait loops replaced with await().atMost(...) — this directly addresses the review feedback from the TAB completion commit. Awaitility dependency added to pom.xml. Five test methods updated consistently. ✓

4. Ollama detection rework: Replaced the fragile inferApiTypeFromUrlHost() URL-pattern heuristic with isOllamaRoot(), which actually checks the server's response body for "Ollama is running". Much more reliable — the old approach guessed from URL substrings like "11434" or "ollama". New test with a real HTTP server covers non-default port detection.

LGTM — all changes are well-tested. Still approved.