thwip-cli 1.1.3__tar.gz → 1.3.0__tar.gz

{thwip_cli-1.1.3 → thwip_cli-1.3.0}/.github/workflows/publish.yml

@@ -49,7 +49,7 @@ jobs:
         run: uv build
 
       - name: Verify package metadata
-        run: uvx twine check dist/*
+        run: uvx twine check dist/*.whl dist/*.tar.gz
 
       - name: Publish to PyPI
         uses: pypa/gh-action-pypi-publish@release/v1
@@ -60,5 +60,7 @@ jobs:
         uses: softprops/action-gh-release@v2
         if: startsWith(github.ref, 'refs/tags/')
         with:
-          files: dist/*
+          files: |
+            dist/*.whl
+            dist/*.tar.gz
           generate_release_notes: true

{thwip_cli-1.1.3 → thwip_cli-1.3.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: thwip-cli
-Version: 1.1.3
+Version: 1.3.0
 Summary: Universal coding agent multiplexer: detect, switch, and route between AI coding agents seamlessly
 Project-URL: Homepage, https://github.com/tanmayhutt/thwip-cli
 Project-URL: Repository, https://github.com/tanmayhutt/thwip-cli
@@ -46,7 +46,8 @@ Description-Content-Type: text/markdown
 ## Features
 
 - **Auto-Detection**: Discovers installed AI coding agents (Claude Code, Antigravity, Gemini CLI, OpenAI/Codex, Aider, Copilot, Cursor, Windsurf, Cline, Ollama) and configured credentials.
-- **Context Portability**: Switch between Anthropic Claude, Google Gemini, OpenAI Codex, or local Ollama mid-project. Conversation history and working files transfer directly.
+- **Context Portability**: Switch providers with stored conversational text. Working files remain in the selected local project; they are not automatically uploaded.
+- **Handoff Preview**: Inspect text continuity, omitted state, capability changes, approximate context pressure, and a text fingerprint before switching. Runs locally without model calls.
 - **Dynamic UI**: Terminal interface adapts its status bar, capabilities, and theme based on the active provider.
 - **Capability Disclaimers**: Highlights when an agent lacks specific capabilities such as file editing or code execution.
 - **Rate Limit Failover**: Detects HTTP 429 errors or quota exhaustion and prompts instant switching to ready fallback models.
@@ -89,6 +90,8 @@ thwip
 | Command | Action |
 |:---|:---|
 | `/switch [agent] [model]` | Switch active agent or model mid-conversation |
+| `/handoff [agent] [model]` | Preview a target locally without switching or sending data |
+| `/native codex` | Save and leave Thwip for the installed Codex CLI using its own authentication |
 | `/agents` | Show all detected coding agents, company status, and capabilities |
 | `/models [agent]` | List available models for current or target agent |
 | `/key [provider]` | Enter an API key securely without placing it in prompt history |
@@ -103,14 +106,61 @@ thwip
 | `/cost` | Show estimated session and cumulative cost |
 | `/project [path]` | View or change project working directory |
 | `Ctrl + S` | Quick switch agent prompt |
-
-Short aliases are available for frequent commands: `/a`, `/m`, `/s`, `/k`, `/g`, and `/t`.
 | `Ctrl + T` | Show agent status |
 | `Ctrl + H` | View history |
 | `/quit` | Exit thwip |
 
+Short aliases are available for frequent commands: `/a`, `/m`, `/s`, `/k`, `/g`, and `/t`.
+
+### Existing Codex login
+
+Use `/native codex` to open the installed native CLI without configuring a Thwip API
+key. After confirmation, Thwip saves its session and replaces itself with Codex in
+the selected project, using a read-only sandbox and on-request approvals. Codex
+owns authentication, model selection, permissions, and usage limits. An expired
+or missing login must be resolved in Codex itself.
+
+This is a launcher, not an in-REPL provider adapter. Conversation history is not
+transferred, native activity is not included in Thwip usage totals, and exiting
+Codex does not automatically restart Thwip. Restart `thwip` and use `/session load`
+to resume the saved Thwip conversation. Claude and Gemini native launchers are not
+implemented. Direct API mode remains unchanged.
+
 ---
 
+## Auditable handoffs
+
+```text
+/handoff
+/handoff google
+/handoff openai gpt-5.6-terra
+```
+
+`/handoff` previews the current target; specifying a provider uses its default model
+unless you supply a model ID. Targets can be inspected without credentials or an
+installed provider. `/switch` shows the same report before changing the active agent.
+
+The report includes:
+
+- Exact counts of transferred user/assistant text messages and excluded stored records.
+- A count of observed transient tool results that are not in portable history. New
+  sessions track from creation; older saved sessions explicitly report partial coverage.
+- Capability gains and losses using the local model catalog.
+- Approximate request size including tool schemas, with up to 4,096 tokens reserved
+  for an answer in the advisory calculation. This does not change generation settings.
+- SHA-256 of canonical system-prompt and conversational-text JSON. It stays the same
+  across targets when that text is unchanged. Tool schemas and attribution metadata
+  are intentionally outside this text fingerprint.
+
+The preview makes no model requests, saves no transcript exports, executes no tools,
+and does not trim or summarize history. Token sizing uses UTF-8 bytes divided by four
+plus per-message overhead, not a provider tokenizer. Catalog limits can be stale;
+an apparently fitting request can still fail. Warnings are advisory, not switch gates.
+Hidden reasoning and provider-native state do not transfer. The digest proves neither
+delivery nor semantic understanding, and is not a signature or privacy guarantee.
+
+See [research and prior art](docs/handoff-research.md) for the differentiation rationale.
+
 ## Supported Companies and Agents
 
 | Company | Agent | Capabilities |
@@ -119,7 +169,7 @@ Short aliases are available for frequent commands: `/a`, `/m`, `/s`, `/k`, `/g`,
 | Google | Gemini API (3.1 Pro Preview, 3.7 Flash, 3.5 Flash-Lite) | Chat, File Edit, Code Run, Terminal, Git |
 | OpenAI | OpenAI API (GPT-5.6 Sol, Terra, Luna) | Chat, File Edit, Code Run, Terminal, Git |
 | DeepSeek | DeepSeek V3 / R1 Reasoner | Chat, File Edit, Code Run, Reasoning |
-| Groq | Llama 3.3 70B, Mixtral | Chat, File Edit, Code Run |
+| Groq | GPT-OSS 120B (default); Llama 3.3 for eligible enterprise accounts only | Chat, File Edit, Code Run |
 | Ollama | Local Models (Llama 3.3, Qwen Coder, DeepSeek R1) | Chat, File Edit, Code Run (Local, Offline) |
 | OpenRouter | Multi-Company Models | Gateway Routing |
 

{thwip_cli-1.1.3 → thwip_cli-1.3.0}/README.md

@@ -9,7 +9,8 @@
 ## Features
 
 - **Auto-Detection**: Discovers installed AI coding agents (Claude Code, Antigravity, Gemini CLI, OpenAI/Codex, Aider, Copilot, Cursor, Windsurf, Cline, Ollama) and configured credentials.
-- **Context Portability**: Switch between Anthropic Claude, Google Gemini, OpenAI Codex, or local Ollama mid-project. Conversation history and working files transfer directly.
+- **Context Portability**: Switch providers with stored conversational text. Working files remain in the selected local project; they are not automatically uploaded.
+- **Handoff Preview**: Inspect text continuity, omitted state, capability changes, approximate context pressure, and a text fingerprint before switching. Runs locally without model calls.
 - **Dynamic UI**: Terminal interface adapts its status bar, capabilities, and theme based on the active provider.
 - **Capability Disclaimers**: Highlights when an agent lacks specific capabilities such as file editing or code execution.
 - **Rate Limit Failover**: Detects HTTP 429 errors or quota exhaustion and prompts instant switching to ready fallback models.
@@ -52,6 +53,8 @@ thwip
 | Command | Action |
 |:---|:---|
 | `/switch [agent] [model]` | Switch active agent or model mid-conversation |
+| `/handoff [agent] [model]` | Preview a target locally without switching or sending data |
+| `/native codex` | Save and leave Thwip for the installed Codex CLI using its own authentication |
 | `/agents` | Show all detected coding agents, company status, and capabilities |
 | `/models [agent]` | List available models for current or target agent |
 | `/key [provider]` | Enter an API key securely without placing it in prompt history |
@@ -66,14 +69,61 @@ thwip
 | `/cost` | Show estimated session and cumulative cost |
 | `/project [path]` | View or change project working directory |
 | `Ctrl + S` | Quick switch agent prompt |
-
-Short aliases are available for frequent commands: `/a`, `/m`, `/s`, `/k`, `/g`, and `/t`.
 | `Ctrl + T` | Show agent status |
 | `Ctrl + H` | View history |
 | `/quit` | Exit thwip |
 
+Short aliases are available for frequent commands: `/a`, `/m`, `/s`, `/k`, `/g`, and `/t`.
+
+### Existing Codex login
+
+Use `/native codex` to open the installed native CLI without configuring a Thwip API
+key. After confirmation, Thwip saves its session and replaces itself with Codex in
+the selected project, using a read-only sandbox and on-request approvals. Codex
+owns authentication, model selection, permissions, and usage limits. An expired
+or missing login must be resolved in Codex itself.
+
+This is a launcher, not an in-REPL provider adapter. Conversation history is not
+transferred, native activity is not included in Thwip usage totals, and exiting
+Codex does not automatically restart Thwip. Restart `thwip` and use `/session load`
+to resume the saved Thwip conversation. Claude and Gemini native launchers are not
+implemented. Direct API mode remains unchanged.
+
 ---
 
+## Auditable handoffs
+
+```text
+/handoff
+/handoff google
+/handoff openai gpt-5.6-terra
+```
+
+`/handoff` previews the current target; specifying a provider uses its default model
+unless you supply a model ID. Targets can be inspected without credentials or an
+installed provider. `/switch` shows the same report before changing the active agent.
+
+The report includes:
+
+- Exact counts of transferred user/assistant text messages and excluded stored records.
+- A count of observed transient tool results that are not in portable history. New
+  sessions track from creation; older saved sessions explicitly report partial coverage.
+- Capability gains and losses using the local model catalog.
+- Approximate request size including tool schemas, with up to 4,096 tokens reserved
+  for an answer in the advisory calculation. This does not change generation settings.
+- SHA-256 of canonical system-prompt and conversational-text JSON. It stays the same
+  across targets when that text is unchanged. Tool schemas and attribution metadata
+  are intentionally outside this text fingerprint.
+
+The preview makes no model requests, saves no transcript exports, executes no tools,
+and does not trim or summarize history. Token sizing uses UTF-8 bytes divided by four
+plus per-message overhead, not a provider tokenizer. Catalog limits can be stale;
+an apparently fitting request can still fail. Warnings are advisory, not switch gates.
+Hidden reasoning and provider-native state do not transfer. The digest proves neither
+delivery nor semantic understanding, and is not a signature or privacy guarantee.
+
+See [research and prior art](docs/handoff-research.md) for the differentiation rationale.
+
 ## Supported Companies and Agents
 
 | Company | Agent | Capabilities |
@@ -82,7 +132,7 @@ Short aliases are available for frequent commands: `/a`, `/m`, `/s`, `/k`, `/g`,
 | Google | Gemini API (3.1 Pro Preview, 3.7 Flash, 3.5 Flash-Lite) | Chat, File Edit, Code Run, Terminal, Git |
 | OpenAI | OpenAI API (GPT-5.6 Sol, Terra, Luna) | Chat, File Edit, Code Run, Terminal, Git |
 | DeepSeek | DeepSeek V3 / R1 Reasoner | Chat, File Edit, Code Run, Reasoning |
-| Groq | Llama 3.3 70B, Mixtral | Chat, File Edit, Code Run |
+| Groq | GPT-OSS 120B (default); Llama 3.3 for eligible enterprise accounts only | Chat, File Edit, Code Run |
 | Ollama | Local Models (Llama 3.3, Qwen Coder, DeepSeek R1) | Chat, File Edit, Code Run (Local, Offline) |
 | OpenRouter | Multi-Company Models | Gateway Routing |
 

thwip_cli-1.3.0/docs/handoff-research.md

@@ -0,0 +1,50 @@
+# Handoff preview: research and scope
+
+Research date: 2026-09-04. This is a bounded public-source comparison, not a patent
+search or proof of worldwide novelty. Search indexes miss unpublished work, private
+products, and features documented under different names. No "world first" claim is made.
+
+## What already exists
+
+| Primary source | Existing approach | Implication for Thwip |
+| --- | --- | --- |
+| [Aider chat modes](https://aider.chat/docs/usage/modes.html) | Architect and editor model pairing | Multi-model collaboration alone is not novel. |
+| [Aider model warnings](https://aider.chat/docs/llms/warnings.html) and [token limits](https://aider.chat/docs/troubleshooting/token-limits.html) | Metadata warnings, token accounting, and overflow guidance | Model-fit diagnostics alone are not novel. |
+| [OpenCode compaction](https://opencode.ai/v2/docs/compaction) | Structured checkpoints plus recent context; durable messages retained separately | Summaries and checkpoints alone are not novel. |
+| [OpenRouter fallbacks](https://openrouter.ai/docs/guides/routing/model-fallbacks) | Ordered fallback models on request failures | Routing and failover alone are not novel. |
+| [OpenRouter message transforms](https://openrouter.ai/docs/guides/features/message-transforms) | Context compression by removing or truncating messages | Automatic fitting can trade away recall. |
+| [Agent Handoff](https://github.com/AniruddhaHumane/handoff) | Portable file-backed resume briefs across agents | Cross-agent memory alone is not novel. |
+| [rosehgal/handoff](https://github.com/rosehgal/handoff) | Append-only action log and rendered handoff document | Tool-action tracking and durable handoffs already exist. |
+
+Searches included combinations of "coding agent", "model switch", "handoff",
+"context loss report", "preflight", "capability loss", "fingerprint", and "receipt",
+as well as product-specific documentation queries. Sources were inspected through
+web search retrieval. No restricted pages or private data were scraped.
+
+## Chosen differentiation
+
+Thwip already owns the provider switch boundary. Put a local continuity report at
+that boundary: exact text-message coverage, explicit state omissions, capability
+differences, advisory context pressure, and a deterministic text fingerprint together.
+The examined sources establish prior art for the components; they did not establish
+the same integrated report. That is an opportunity hypothesis, not proof of uniqueness.
+
+## Implementation boundaries
+
+- `/handoff [provider] [model]` is a non-mutating dry run. `/switch` displays it too.
+- No new dependencies, model calls, summarization costs, or external data storage.
+- Only a result count is added to saved sessions, not raw tool inputs or outputs.
+- Legacy sessions explicitly show incomplete tool-result tracking.
+- SHA-256 covers canonical system prompt and portable text, not provider wire formats,
+  delivery, hidden reasoning, tool schemas, project file contents, or model comprehension.
+- Context estimates are heuristic and warnings advisory. Provider-native tool
+  continuation correctness remains a separate issue, not solved by this report.
+- Production deployment and other defects from the previous audit remain separate work.
+
+## Validation
+
+Offline tests cover stable/change-sensitive fingerprints, exclusions, capability
+gains/losses, context thresholds and unknown limits, unknown targets, private session
+persistence, legacy migration, clear commands, dry-run non-mutation, safe terminal
+rendering, switch integration, and command completion. Live generation is not needed
+to validate this diagnostic, and no claim about live provider success follows from it.

thwip_cli-1.3.0/docs/verification.md

@@ -0,0 +1,28 @@
+# Verification status
+
+Verified locally on 2026-09-07 for v1.3.0. The Python suite passes 186 tests.
+Exhaustive behavior across every provider and
+configuration has not been established.
+
+| Area | Evidence | Remaining limitation |
+| --- | --- | --- |
+| Commands and aliases | Offline command smoke tests, invalid input cases | Most smoke tests check exceptions, not all rendered content |
+| Sessions | Save/load, separate fresh conversations, malformed metadata, permissions, project rebinding | Concurrent writes to the same explicitly named session are not coordinated |
+| Provider switching and handoff | All seven provider catalogs, portable history, bounded failover | Live account quotas and model availability unverified |
+| Tool execution | Real temporary-file operations, path containment, process timeout/cancellation, invalid arguments | Shell and code tools retain local user privileges; output capture memory is unbounded |
+| Native Codex launcher | Save-before-launch, consent, flags, missing binary, terminal checks, failures | Mocked process replacement; live native interaction unverified; no conversation transfer |
+| Provider responses | Mocked native tool continuations; DeepSeek/Groq/OpenRouter streaming, usage-only chunks, 429/500 errors and serialized tool arguments | Other streaming and error branches still have coverage gaps |
+| Display/config/auth | Configuration validation, display settings, credential boundaries, short-key masking | No full terminal/platform matrix |
+| Usage | Atomic writes, malformed records, valid totals | Unknown catalog pricing may appear as zero estimated cost |
+| Website | Production build and deterministic demo completion/replay test | Real browser, layout, clipboard, keyboard and accessibility checks remain incomplete |
+| Dependencies | npm audit: zero advisories; Python installed-dependency audit: none found | Python audit skipped legacy local `thwip 1.0.0` metadata |
+| Packaging | Wheel/source build and metadata checks | Publication is separately verified through the release workflow |
+
+The Python suite measured 69% statement coverage with 160 passing tests before
+the final credential-masking tests were added. Coverage is diagnostic evidence,
+not proof that every feature works. The website test uses a minimal DOM stand-in
+and controlled timers, not a browser.
+
+This pass fixed default-session save collisions, malformed session/message
+metadata acceptance, tool-argument display crashes, Rich markup interpretation
+in action output, and short-key masking leakage.

{thwip_cli-1.1.3 → thwip_cli-1.3.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "thwip-cli"
-version = "1.1.3"
+version = "1.3.0"
 description = "Universal coding agent multiplexer: detect, switch, and route between AI coding agents seamlessly"
 readme = "README.md"
 license = "MIT"

thwip_cli-1.3.0/tests/test_audit_regressions.py

@@ -0,0 +1,164 @@
+import json
+from io import StringIO
+from types import SimpleNamespace
+
+import pytest
+from rich.console import Console
+
+from thwip.agents import AgentRegistry
+from thwip.agents.base import AgentDone, Capability, LimitHit, LimitStatus, TextDelta
+from thwip.cli import ThwipCLI
+from thwip.config import ThwipConfig, get_usage_path
+from thwip.limits import UsageTracker
+from thwip.session import Session
+from thwip.tools import ToolManager
+
+
+@pytest.fixture
+def cli(tmp_path, monkeypatch):
+    monkeypatch.setenv('THWIP_CONFIG_DIR', str(tmp_path / 'config'))
+    cli = ThwipCLI.__new__(ThwipCLI)
+    cli.config = ThwipConfig(project=str(tmp_path), auto_save=False)
+    cli.registry = AgentRegistry(cli.config)
+    for a in cli.registry.list_agents():
+        monkeypatch.setattr(a, 'is_installed', lambda: True)
+        monkeypatch.setattr(a, 'is_configured', lambda: False)
+        if a.name == 'ollama':
+            a._cached_models = a.get_handoff_models()
+    cli.current_agent = cli.registry.get_agent('claude')
+    cli.session = Session(project_path=str(tmp_path), current_agent='claude', current_model=cli.current_agent.get_default_model())
+    cli.usage_tracker = UsageTracker()
+    cli.tool_manager = ToolManager(str(tmp_path))
+    cli.detector = SimpleNamespace(scan_all=list)
+    monkeypatch.setattr('builtins.input', lambda *args: '')
+    monkeypatch.setattr('getpass.getpass', lambda *args: '')
+    console = Console(file=StringIO(), width=100, color_system=None)
+    monkeypatch.setattr('thwip.cli.console', console)
+    monkeypatch.setattr('thwip.theme.console', console)
+    return cli
+
+
+@pytest.mark.parametrize('command', [
+    '/help', '/h', '/about', '/guide', '/g', '/info', '/agents', '/a', '/list',
+    '/models', '/m', '/models flagship', '/models balanced', '/models fast', '/models google',
+    '/tools', '/t', '/status', '/limits', '/detect', '/history', '/clear', '/reset', '/cost',
+    '/project', '/session save audit', '/session load missing', '/session list', '/session clear',
+    '/key', '/key google', '/key invalid', '/key openai REJECTED_TEST_KEY',
+    '/handoff', '/handoff invalid', '/handoff google invalid', '/switch', '/switch invalid',
+    '/switch google invalid', '/unknown', '/q', '/exit', '/quit',
+])
+@pytest.mark.asyncio
+async def test_command_no_exception(cli, command):
+    await cli.handle_command(command)
+
+
+@pytest.mark.parametrize('provider', ['claude','google','openai','deepseek','groq','ollama','openrouter'])
+@pytest.mark.asyncio
+async def test_all_catalogued_handoffs(cli, provider):
+    before = list(cli.session.messages)
+    for model in cli.registry.get_agent(provider).get_handoff_models():
+        await cli.handle_command(f'/handoff {provider} {model.id}')
+    assert cli.session.messages == before
+
+
+@pytest.mark.asyncio
+async def test_project_path_with_spaces(cli, tmp_path):
+    project = tmp_path / 'project with spaces'
+    project.mkdir()
+    await cli.handle_command(f'/project {project}')
+    assert cli.session.project_path == str(project)
+
+
+def test_usage_file_private(cli):
+    cli.usage_tracker.record_usage('openai', 'gpt-5.6-terra', 10, 5)
+    assert get_usage_path().stat().st_mode & 0o777 == 0o600
+
+
+@pytest.mark.parametrize('tool,args', [
+    ('read_file', {'file_path': 12}),
+    ('list_files', {'sub_dir': ['bad']}),
+    ('run_python', {'code': None}),
+])
+def test_invalid_tool_args_return_error_not_exception(cli, tool, args):
+    result = cli.tool_manager.execute_tool(tool, args)
+    assert isinstance(result, str)
+
+
+@pytest.mark.asyncio
+async def test_history_displays_literal_markup(cli):
+    cli.session.add_user_message('Show [/not-a-tag] literally')
+    await cli.handle_command('/history')
+
+
+@pytest.mark.parametrize('value', ['wrong', None, -1, True])
+def test_corrupted_handoff_counter_rejected_or_sanitized(cli, value):
+    path = cli.session.save('damaged')
+    data = json.loads(path.read_text())
+    data['observed_tool_results'] = value
+    path.write_text(json.dumps(data))
+    loaded = Session.load('damaged')
+    assert loaded is None or (type(loaded.observed_tool_results) is int and loaded.observed_tool_results >= 0)
+
+
+def test_invalid_config_type_rejected_or_defaulted(cli):
+    cli.config._apply_toml({'defaults': {'project': 123, 'confirm_tools': 'false'}})
+    assert isinstance(cli.config.project, str)
+    assert type(cli.config.confirm_tools) is bool
+
+
+@pytest.mark.asyncio
+async def test_session_roundtrip_rebinds_tools(cli, tmp_path):
+    target = tmp_path / 'other'
+    target.mkdir()
+    saved = Session(project_path=str(target), current_agent='google', current_model='missing')
+    saved.save('loadable')
+    await cli.handle_command('/session load loadable')
+    assert cli.tool_manager.file_editor.project_path == target
+    assert cli.current_agent.name == 'google'
+    assert cli.session.current_model == cli.current_agent.get_default_model()
+
+
+def test_all_tools_and_safety(cli, tmp_path):
+    tm = cli.tool_manager
+    assert 'Successfully' in tm.execute_tool('write_file', {'file_path':'a.txt','content':'one'})
+    assert tm.execute_tool('read_file', {'file_path':'a.txt'}) == 'one'
+    assert 'Successfully' in tm.execute_tool('edit_file', {'file_path':'a.txt','old_str':'one','new_str':'two'})
+    assert 'a.txt' in tm.execute_tool('list_files', {})
+    assert '42' in tm.execute_tool('run_python', {'code':'print(6 * 7)'})
+    assert 'Exit code: 0' in tm.execute_tool('run_command', {'command':'printf audit-ok'})
+    assert isinstance(tm.execute_tool('git_status', {}), str)
+    assert isinstance(tm.execute_tool('git_diff', {}), str)
+    assert 'outside' in tm.execute_tool('read_file', {'file_path':'../outside'})
+    outside = tmp_path.parent / 'audit-outside-file'
+    outside.write_text('not accessible')
+    (tmp_path / 'link').symlink_to(outside)
+    assert 'outside' in tm.execute_tool('read_file', {'file_path':'link'})
+    assert 'Error' in tm.execute_tool('unknown', {})
+    assert {t['name'] for t in tm.get_anthropic_tools()} == {t['function']['name'] for t in tm.get_openai_tools()}
+
+
+@pytest.mark.asyncio
+async def test_failover_does_not_retry_already_failed_provider(cli, monkeypatch):
+    agents = [cli.registry.get_agent('claude'), cli.registry.get_agent('google')]
+    attempts = []
+    for agent in agents:
+        monkeypatch.setattr(agent, 'is_configured', lambda: True)
+        monkeypatch.setattr(agent, 'get_capabilities_for_model', lambda model: {Capability.CHAT})
+
+        def make_chat(name):
+            async def chat(**kwargs):
+                attempts.append(name)
+                # Stop safely after four simulated rate limits; no actual network calls.
+                if len(attempts) <= 4:
+                    yield LimitHit(error_type=LimitStatus.RATE_LIMITED, message='test quota')
+                else:
+                    yield TextDelta(content='stop test')
+                    yield AgentDone()
+            return chat
+
+        monkeypatch.setattr(agent, 'chat', make_chat(agent.name))
+    monkeypatch.setattr(cli.registry, 'get_ready_agents', lambda: agents)
+    cli.config.fallback.chain = ['claude', 'google']
+    cli.config.limits.auto_switch = True
+    await cli.process_user_message('hello')
+    assert len(attempts) <= len(agents), f'Repeated failed providers: {attempts}'

{thwip_cli-1.1.3 → thwip_cli-1.3.0}/tests/test_cli.py

@@ -67,6 +67,7 @@ async def test_tool_results_are_returned_to_agent(tmp_path):
 
     assert len(cli.current_agent.calls) == 2
     assert cli.session.messages[-1].content == "Used the tool result."
+    assert cli.session.observed_tool_results == 1
 
 
 class UnconfiguredAgent(ToolCallingAgent):
@@ -74,6 +75,30 @@ class UnconfiguredAgent(ToolCallingAgent):
         return False
 
 
+@pytest.mark.asyncio
+@pytest.mark.parametrize("arguments", [None, [], "bad", {"file_path": "[broken]"}])
+async def test_invalid_tool_arguments_return_errors_without_crashing(tmp_path, arguments):
+    class InvalidToolAgent(ToolCallingAgent):
+        async def chat(self, messages, **kwargs):
+            self.calls.append(messages)
+            if len(self.calls) == 1:
+                yield ToolUseStart(tool_id="bad", tool_name="read_file", args=arguments)
+                yield AgentDone()
+            else:
+                assert "Error" in messages[-1]["content"]
+                yield TextDelta(content="Recovered from invalid tool arguments.")
+                yield AgentDone()
+
+    cli = ThwipCLI.__new__(ThwipCLI)
+    cli.config = SimpleNamespace(stream=True, confirm_tools=True)
+    cli.session = Session(current_agent="fake", current_model="fake-model")
+    cli.current_agent = InvalidToolAgent()
+    cli.tool_manager = ToolManager(tmp_path)
+    cli.usage_tracker = FakeUsageTracker()
+    await cli.process_user_message("Read")
+    assert cli.session.messages[-1].content == "Recovered from invalid tool arguments."
+
+
 @pytest.mark.asyncio
 async def test_unconfigured_agent_shows_setup_guidance(tmp_path):
     cli = ThwipCLI.__new__(ThwipCLI)

thwip_cli-1.3.0/tests/test_compatible_streaming.py

@@ -0,0 +1,75 @@
+"""Exercise streaming completion and provider error boundaries without network calls."""
+
+import json
+from types import SimpleNamespace
+
+import httpx
+import openai
+import pytest
+
+from thwip.agents.base import AgentDone, LimitHit, LimitStatus, TextDelta
+from thwip.agents.deepseek_agent import DeepSeekAgent
+from thwip.agents.groq_agent import GroqAgent
+from thwip.agents.openrouter_agent import OpenRouterAgent
+
+
+@pytest.mark.parametrize("agent_type", [DeepSeekAgent, GroqAgent, OpenRouterAgent])
+@pytest.mark.asyncio
+async def test_tool_continuation_serializes_arguments_without_mutating_history(agent_type):
+    original = {"role": "assistant", "content": "", "tool_calls": [{
+        "id": "call1", "type": "function", "function": {
+            "name": "read_file", "arguments": {"file_path": "note.txt"},
+        },
+    }], "_native_state": {"other": "private"}}
+
+    async def create(**kwargs):
+        sent = kwargs["messages"][0]
+        assert "_native_state" not in sent
+        assert json.loads(sent["tool_calls"][0]["function"]["arguments"]) == {"file_path": "note.txt"}
+        return SimpleNamespace(choices=[SimpleNamespace(message=SimpleNamespace(content="done", tool_calls=[]))], usage=None)
+
+    agent = agent_type(api_key="test")
+    agent._client = SimpleNamespace(chat=SimpleNamespace(completions=SimpleNamespace(create=create)))
+    events = [event async for event in agent.chat([original], stream=False)]
+    assert any(isinstance(event, AgentDone) for event in events)
+    assert isinstance(original["tool_calls"][0]["function"]["arguments"], dict)
+    assert "_native_state" in original
+
+
+@pytest.mark.parametrize("agent_type", [DeepSeekAgent, GroqAgent, OpenRouterAgent])
+@pytest.mark.asyncio
+async def test_stream_text_and_usage_only_final_chunk(agent_type):
+    async def chunks():
+        yield SimpleNamespace(choices=[SimpleNamespace(delta=SimpleNamespace(
+            content="hello", tool_calls=None))], usage=None)
+        yield SimpleNamespace(choices=[], usage=SimpleNamespace(prompt_tokens=9, completion_tokens=3))
+
+    async def create(**kwargs):
+        assert kwargs["stream"] is True
+        return chunks()
+
+    agent = agent_type(api_key="test")
+    agent._client = SimpleNamespace(chat=SimpleNamespace(completions=SimpleNamespace(create=create)))
+    events = [event async for event in agent.chat([{"role": "user", "content": "hi"}])]
+    assert isinstance(events[0], TextDelta) and events[0].content == "hello"
+    assert isinstance(events[-1], AgentDone)
+    assert events[-1].usage.input_tokens == 9
+    assert events[-1].usage.output_tokens == 3
+    assert agent.check_limits() == LimitStatus.OK
+
+
+@pytest.mark.parametrize("agent_type", [DeepSeekAgent, GroqAgent, OpenRouterAgent])
+@pytest.mark.parametrize("status", [429, 500])
+@pytest.mark.asyncio
+async def test_provider_errors_emit_failure_not_success(agent_type, status):
+    response = httpx.Response(status, request=httpx.Request("POST", "https://example.invalid"))
+    error_type = openai.RateLimitError if status == 429 else openai.APIStatusError
+
+    async def create(**kwargs):
+        raise error_type("simulated failure", response=response, body=None)
+
+    agent = agent_type(api_key="test")
+    agent._client = SimpleNamespace(chat=SimpleNamespace(completions=SimpleNamespace(create=create)))
+    events = [event async for event in agent.chat([])]
+    assert len(events) == 1 and isinstance(events[0], LimitHit)
+    assert events[0].error_type == (LimitStatus.RATE_LIMITED if status == 429 else LimitStatus.UNKNOWN)