tunacode-cli 0.1.74__tar.gz → 0.1.75__tar.gz

tunacode_cli-0.1.75/.claude/delta/2026-02-22-enable-mypy-pre-commit-hook.md

@@ -0,0 +1,34 @@
+---
+title: Enable mypy at pre-commit stage (keep pre-push enforcement)
+type: delta
+link: enable-mypy-pre-commit-hook
+path: .pre-commit-config.yaml
+depth: 0
+seams: [S]
+ontological_relations:
+  - affects: [[tooling]]
+  - affects: [[typing-contracts]]
+  - affects: [[developer-workflow]]
+tags:
+  - mypy
+  - pre-commit
+  - quality-gate
+created_at: 2026-02-22T15:17:04-06:00
+updated_at: 2026-02-22T15:17:04-06:00
+uuid: 6a5c9232-30f7-4aa9-bf94-ecb0197d6263
+---
+
+# Enable mypy at pre-commit stage (keep pre-push enforcement)
+
+## Summary
+
+Updated `.pre-commit-config.yaml` so the `mypy` hook runs during both:
+
+- `pre-commit`
+- `pre-push`
+
+This catches type errors earlier at commit time while retaining the existing push-time gate.
+
+## Verification
+
+- `uv run pre-commit run mypy --files src/tunacode/ui/widgets/editor.py` ✅

tunacode_cli-0.1.75/.claude/delta/2026-02-22-mypy-type-contract-fixes.md

@@ -0,0 +1,47 @@
+---
+title: Resolve four mypy contract violations with source-level typing fixes
+type: delta
+link: mypy-type-contract-fixes
+path: src/tunacode
+depth: 1
+seams: [S, M]
+ontological_relations:
+  - affects: [[ui]]
+  - affects: [[core]]
+  - affects: [[tools]]
+  - affects: [[typing-contracts]]
+tags:
+  - mypy
+  - typing
+  - literals
+  - rich
+  - error-handling
+created_at: 2026-02-22T15:12:48-06:00
+updated_at: 2026-02-22T15:12:48-06:00
+uuid: febc9ce8-4061-4ee8-b2a1-770f0997a7ca
+---
+
+# Resolve four mypy contract violations with source-level typing fixes
+
+## Summary
+
+Fixed the reported mypy errors by tightening type contracts at their source:
+
+1. `editor.py`: added explicit Rich style narrowing so `Text.stylize()` always receives `StyleType`.
+2. `web_fetch.py`: marked `_handle_http_error()` as `NoReturn` because it always raises.
+3. `compaction/types.py`: made compaction status constants literal-typed (`Final[Literal[...]]`) and defined `CompactionStatus` as a type alias.
+4. `shell_runner.py`: unpacked `render_bash()` tool render tuple and returned only the renderable content as required by `RenderableType`.
+
+No shims, no fallback branches, and no cast-based suppression were introduced.
+
+## Files changed
+
+- `src/tunacode/ui/widgets/editor.py`
+- `src/tunacode/tools/web_fetch.py`
+- `src/tunacode/core/compaction/types.py`
+- `src/tunacode/ui/shell_runner.py`
+
+## Verification
+
+- `uv run mypy src/tunacode/ui/widgets/editor.py src/tunacode/tools/web_fetch.py src/tunacode/core/compaction/controller.py src/tunacode/ui/shell_runner.py src/tunacode/core/compaction/types.py` ✅
+- `uv run ruff check src/tunacode/ui/widgets/editor.py src/tunacode/tools/web_fetch.py src/tunacode/core/compaction/types.py src/tunacode/ui/shell_runner.py` ✅

tunacode_cli-0.1.75/.claude/delta/2026-02-23_parallel-tool-call-rollout.md

@@ -0,0 +1,34 @@
+---
+type: delta
+date: 2026-02-23
+status: done
+tags: [tinyagent, parallel-tool-calls, concurrency-cap, tui]
+---
+
+# Parallel Tool-call Rollout via tiny-agent-os 1.2.5
+
+## Scope completed
+
+- `tun-d506` — bumped `tiny-agent-os` to `>=1.2.5` and refreshed `uv.lock`.
+- `tun-q44p` — enforced a hard cap of 3 concurrent tool executions by wrapping all runtime `AgentTool.execute` handlers with a shared semaphore in `agent_config.py`.
+- `tun-5nny` — hardened `RequestOrchestrator` tool lifecycle handling:
+  - strict `tool_execution_start` args normalization (`dict` or JSON object string only)
+  - fail-loud `TypeError` on malformed payloads
+  - explicit duration semantics for parallel batches (suppress per-tool duration when batch execution is detected)
+- `tun-6qpn` — updated TUI status behavior for parallel starts/completions:
+  - status bar tracks multiple in-flight tools coherently
+  - callbacks now emit completion before last-action updates
+  - shell status path aligned with the same completion semantics
+
+## Repo truth updates
+
+- README migration note now reflects parallel batches with max 3 in-flight calls.
+- Changelog `Unreleased` section documents dependency cutover, runtime cap, orchestrator hardening, and UI behavior update.
+
+## Validation
+
+- `uv run ruff check --fix .`
+- `uv run ruff check .`
+- `uv run pytest tests/unit/core/test_tool_concurrency_limit.py tests/unit/core/test_request_orchestrator_parallel_tools.py tests/unit/ui/test_status_bar_layout.py tests/system/cli/test_repl_support.py tests/unit/ui/test_tool_panel_css_flow.py`
+- `uv run pytest`
+- `uv run python -c "import importlib.metadata as m; print(m.version('tiny-agent-os'))"` -> `1.2.5`

tunacode_cli-0.1.75/.claude/metadata/research/2026-02-23-parallel-tool-calls-rollout-plan.md

@@ -0,0 +1,117 @@
+---
+title: Parallel tool-call rollout plan for TunaCode
+type: metadata
+link: parallel-tool-calls-rollout-plan
+path: .claude/metadata/research/2026-02-23-parallel-tool-calls-rollout-plan.md
+depth: 2
+seams: [A]
+ontological_relations:
+  - affects: [[core-agents]]
+  - affects: [[ui]]
+  - affects: [[tools]]
+  - relates_to: [[tinyagent]]
+tags:
+  - tinyagent
+  - parallel-tool-calls
+  - rollout-plan
+  - tui
+created_at: 2026-02-23T14:54:29-06:00
+updated_at: 2026-02-23T14:54:29-06:00
+uuid: ac4317c5-b747-4efe-bd6f-039f4bdc3833
+---
+
+# Parallel tool-call rollout plan for TunaCode
+
+## Objective
+
+Adopt tinyagent parallel tool-call execution in TunaCode with no compatibility shim, while preserving correctness, UI clarity, and fail-loud behavior.
+
+## Code-state snapshot (from current repo)
+
+- Dependency floor is still `tiny-agent-os>=1.2.4` (`pyproject.toml:31`).
+- Lockfile currently resolves `tiny-agent-os` to `1.2.4` (`uv.lock` package stanza).
+- README still states sequential execution (`README.md:40`).
+- TunaCode runtime already consumes tinyagent tool lifecycle events:
+  - `tool_execution_start` handler (`src/tunacode/core/agents/main.py:396`)
+  - `tool_execution_end` handler (`src/tunacode/core/agents/main.py:417`)
+- UI callbacks are start/end based and status-bar driven:
+  - `build_tool_start_callback()` (`src/tunacode/ui/repl_support.py:220`)
+  - `build_tool_result_callback()` (`src/tunacode/ui/repl_support.py:189`)
+  - status methods (`src/tunacode/ui/widgets/status_bar.py:52`, `:55`)
+- Vendored tinyagent code is already on 1.2.5 semantics:
+  - parallel gather in `execute_tool_calls()` (`tinyAgent/tinyagent/agent_tool_execution.py:186`)
+  - all starts emitted before ends (`tinyAgent/tests/test_parallel_tool_execution.py`)
+  - post-batch steering semantics documented in `tinyAgent/CHANGELOG.md`.
+
+## Implementation plan
+
+### Phase 1 — Dependency cutover (no shim)
+
+1. Bump TunaCode dependency floor to `tiny-agent-os>=1.2.5`.
+2. Refresh lockfile with `uv lock` so resolved version is >=1.2.5.
+3. Verify runtime import path remains unchanged (TunaCode already uses tinyagent public API).
+
+**Acceptance**
+- `pyproject.toml` and `uv.lock` both reflect 1.2.5-compatible resolution.
+- `uv run python -c "import importlib.metadata as m; print(m.version('tiny-agent-os'))"` prints >=1.2.5 in project env.
+
+### Phase 2 — Runtime hardening for parallel batches
+
+1. **Normalize tool args at event ingress** in `RequestOrchestrator` start-handler.
+   - Replace cast-only behavior with explicit type checks.
+   - If args are invalid (non-dict JSON object), fail loud with a clear error.
+2. **Validate duration semantics** under batch completion ordering.
+   - Current per-tool duration is measured start→end event in TunaCode.
+   - In parallel mode, all end events are emitted after gather completion, so durations may converge to batch latency.
+   - Decide one policy and apply consistently:
+     - either keep current metric but label it as batch-relative,
+     - or suppress per-tool duration when multiple tools are active in a batch.
+3. Keep steering logic aligned to tinyagent contract (post-batch application).
+
+**Acceptance**
+- No callback crashes from malformed `args` payloads.
+- Tool duration presentation is intentional and documented in code comments/tests.
+
+### Phase 3 — UI behavior alignment for concurrent starts
+
+1. Verify status bar behavior when multiple start events arrive before end events.
+2. Keep existing minimal behavior (`running: <tool>`, then `last: <tool>`) for first cut, or upgrade to explicit concurrent state (count/list) in same PR if low risk.
+3. Ensure tool panels remain classed correctly (`running/completed/failed`) through `tool_panel_smart()`.
+
+**Acceptance**
+- No UI regressions in tool panel rendering.
+- Status bar stays coherent during multi-tool turns.
+
+### Phase 4 — Tests and gates
+
+Add focused tests around TunaCode event handling (not tinyagent internals):
+
+1. `RequestOrchestrator` handles parallel-style order:
+   - start(A), start(B), end(A), end(B)
+   - registry and callbacks remain correct per tool.
+2. `RequestOrchestrator` arg-normalization path:
+   - invalid args payload triggers deterministic failure.
+3. Optional UI test for status bar transitions under batched starts.
+
+Run gates:
+- `uv run pytest`
+- `uv run ruff check --fix .`
+- `uv run ruff .`
+
+### Phase 5 — Repo truth updates
+
+1. Update README statement that still claims sequential-only tool execution.
+2. Add changelog entry describing the cutover and any intentional UI semantics.
+
+## Risks to actively watch
+
+1. **Concurrent mutating tools** (e.g., two file writes/edits in one assistant turn) can create race-like behavior in shared file/cache state.
+2. **Metric trust** if per-tool durations look precise but are batch-relative.
+3. **User visibility** during concurrent tool starts if status bar only shows one tool name.
+
+## Suggested rollout order
+
+1. Dependency bump + lockfile
+2. Runtime hardening + tests
+3. UI polish (if needed)
+4. README/changelog updates

{tunacode_cli-0.1.74 → tunacode_cli-0.1.75}/.pre-commit-config.yaml

@@ -48,7 +48,7 @@ repos:
       - id: ruff-format  # Re-enabled - ruff handles linting, ruff-format handles formatting
         exclude: models_registry\.json$
 
-  # Type checking with mypy (runs on push, not commit - too slow for commit)
+  # Type checking with mypy (runs on commit and push)
   - repo: https://github.com/pre-commit/mirrors-mypy
     rev: v1.14.1
     hooks:
@@ -59,7 +59,7 @@ repos:
           - 'types-requests'
         args: ['--ignore-missing-imports', '--no-strict-optional']
         exclude: '(conftest\.py|src/tunacode/core/agents/|^tests/|^scripts/)'
-        stages: [pre-push]
+        stages: [pre-commit, pre-push]
 
 
   # Local hooks for project-specific checks

{tunacode_cli-0.1.74 → tunacode_cli-0.1.75}/CHANGELOG.md

@@ -7,6 +7,23 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.1.75] - 2026-02-23
+
+### Added
+- Tool panel CSS flow with status-based classes (running/completed/failed)
+- Compaction awareness indicator in context panel
+- CSS tint styling for file tool states (read/update)
+- Parallel tool lifecycle coverage: capped-concurrency execution test, RequestOrchestrator parallel batch tests, and status-bar callback sequencing tests
+
+### Changed
+- Enhanced `/compact` command with error handling and user feedback
+- Updated read_file renderer to support new hashline format (1:ab|content)
+- Improved tool panel rendering with semantic CSS classes
+- Bumped `tiny-agent-os` dependency to `>=1.2.5` and refreshed lockfile resolution
+- Enforced max 3 in-flight tool executions via shared tool semaphore wrapping
+- Hardened RequestOrchestrator tool-start arg normalization and made batch-mode duration reporting explicit (suppressed for multi-tool batches)
+- Updated status bar running-state behavior to stay coherent while multiple tools are active
+
 ## [0.1.74] - 2026-02-21
 
 ### Changed

{tunacode_cli-0.1.74 → tunacode_cli-0.1.75}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: tunacode-cli
-Version: 0.1.74
+Version: 0.1.75
 Summary: Your agentic CLI developer.
 Project-URL: Homepage, https://tunacode.xyz/
 Project-URL: Repository, https://github.com/alchemiststudiosDOTai/tunacode
@@ -31,7 +31,7 @@ Requires-Dist: rich<15.0.0,>=14.2.0
 Requires-Dist: ruff>=0.14.0
 Requires-Dist: textual-autocomplete>=4.0.6
 Requires-Dist: textual<5.0.0,>=4.0.0
-Requires-Dist: tiny-agent-os>=1.2.4
+Requires-Dist: tiny-agent-os>=1.2.5
 Requires-Dist: typer>=0.15.0
 Provides-Extra: dev
 Requires-Dist: autoflake>=2.0.0; extra == 'dev'
@@ -97,7 +97,7 @@ A terminal-based AI coding agent with a NeXTSTEP-inspired interface.
 **Recent Breaking Changes:**
 
 - Session persistence format changed - existing sessions may not load correctly
-- Tool execution is sequential — parallel tool calls are on the roadmap
+- Tool execution runs in parallel batches with a strict cap of 3 in-flight tool calls
 - The `docs/` directory has been restructured
 
 For full details, see the [CHANGELOG](CHANGELOG.md).
@@ -182,7 +182,7 @@ Slash commands are command objects in `tunacode.ui.commands`; each one is a `Com
 | `/help` | Show available commands |
 | `/clear` | Clear transient agent state while preserving message history. |
 | `/compact` | Force context compaction |
-| `/debug` | Toggle debug logging to screen |
+| `/debug` | Toggle debug logging to screen (includes parallel tool-call lifecycle lines) |
 | `/model` | Open model picker or switch model |
 | `/theme` | Open theme picker or switch theme |
 | `/resume` | List, load, or delete persisted sessions. |
@@ -191,6 +191,15 @@ Slash commands are command objects in `tunacode.ui.commands`; each one is a `Com
 | `/exit` | Exit TunaCode |
 | `exit` | Legacy alias for exit |
 
+### Confirm Parallel Tool Calls
+
+Run `/debug` to enable lifecycle logs. During agent execution, parallel batches are reported with lines prefixed by:
+
+- `[LIFECYCLE] Parallel tool calls active: ...`
+- `[LIFECYCLE] Parallel tool calls update: ...`
+- `[LIFECYCLE] Parallel tool calls complete`
+
+If no `Parallel tool calls` lifecycle lines appear, that request did not execute a parallel tool batch.
 ## Tools
 
 <img src="docs/images/tool-calls.png" alt="Tool calls in tunacode" width="600"/>

{tunacode_cli-0.1.74 → tunacode_cli-0.1.75}/README.md

@@ -37,7 +37,7 @@ A terminal-based AI coding agent with a NeXTSTEP-inspired interface.
 **Recent Breaking Changes:**
 
 - Session persistence format changed - existing sessions may not load correctly
-- Tool execution is sequential — parallel tool calls are on the roadmap
+- Tool execution runs in parallel batches with a strict cap of 3 in-flight tool calls
 - The `docs/` directory has been restructured
 
 For full details, see the [CHANGELOG](CHANGELOG.md).
@@ -122,7 +122,7 @@ Slash commands are command objects in `tunacode.ui.commands`; each one is a `Com
 | `/help` | Show available commands |
 | `/clear` | Clear transient agent state while preserving message history. |
 | `/compact` | Force context compaction |
-| `/debug` | Toggle debug logging to screen |
+| `/debug` | Toggle debug logging to screen (includes parallel tool-call lifecycle lines) |
 | `/model` | Open model picker or switch model |
 | `/theme` | Open theme picker or switch theme |
 | `/resume` | List, load, or delete persisted sessions. |
@@ -131,6 +131,15 @@ Slash commands are command objects in `tunacode.ui.commands`; each one is a `Com
 | `/exit` | Exit TunaCode |
 | `exit` | Legacy alias for exit |
 
+### Confirm Parallel Tool Calls
+
+Run `/debug` to enable lifecycle logs. During agent execution, parallel batches are reported with lines prefixed by:
+
+- `[LIFECYCLE] Parallel tool calls active: ...`
+- `[LIFECYCLE] Parallel tool calls update: ...`
+- `[LIFECYCLE] Parallel tool calls complete`
+
+If no `Parallel tool calls` lifecycle lines appear, that request did not execute a parallel tool batch.
 ## Tools
 
 <img src="docs/images/tool-calls.png" alt="Tool calls in tunacode" width="600"/>

{tunacode_cli-0.1.74 → tunacode_cli-0.1.75}/docs/modules/core/core.md

@@ -22,7 +22,7 @@ The engine. Takes a user message, routes it through a tinyagent `Agent`, handles
 |------|---------|
 | `main.py` | `RequestOrchestrator` -- the main request lifecycle. `process_request()` is the public entry point. Handles: history coercion, pre-request compaction, streaming event dispatch, abort cleanup, empty-response intervention, context-overflow retry. |
 | `agent_components/__init__.py` | Re-exports from sub-modules. |
-| `agent_components/agent_config.py` | `get_or_create_agent()` -- builds or retrieves a cached tinyagent `Agent`. Configures: system prompt, tools, model, stream function, API key resolver, compaction transform. `invalidate_agent_cache()` clears both module and session caches after abort/timeout. |
+| `agent_components/agent_config.py` | `get_or_create_agent()` -- builds or retrieves a cached tinyagent `Agent`. Configures: system prompt, tools, model, stream function, API key resolver, compaction transform. `invalidate_agent_cache()` clears both module and session caches after abort/timeout. `_build_tools()` constructs the tool list (bash, discover, read_file, hashline_edit, web_fetch, write_file). Validation functions: `_coerce_request_delay()`, `_coerce_global_request_timeout()`, `_compute_agent_version()`. |
 | `agent_components/agent_helpers.py` | Human-readable tool descriptions for UI panels. `create_empty_response_message()` builds the intervention prompt when the model returns nothing. |
 | `agent_components/state_transition.py` | `AgentStateMachine` -- thread-safe FSM with states: `USER_INPUT -> ASSISTANT -> TOOL_EXECUTION -> RESPONSE`. `AGENT_TRANSITION_RULES` defines valid edges. |
 | `resume/sanitize.py` | Cleans persisted session messages for safe resume (removes dangling tool calls, fixes structural violations). |
@@ -131,6 +131,29 @@ CompactionController.check_and_compact(messages, max_tokens)
 
 `StateManager.load_session()` deserializes and separates thought entries from message history.
 
+### System Prompt
+
+The system prompt defines TunaCode's identity and operational rules for the tinyagent framework.
+
+**Location:** `src/tunacode/prompts/system_prompt.md`
+
+**Loading mechanism:** `load_system_prompt()` in `agent_config.py` reads the markdown file at runtime and appends dynamic context from `load_tunacode_context()`.
+
+**Dynamic context:** `load_tunacode_context()` loads the user's `AGENTS.md` guide file (cached) and injects it into the prompt under the `<user_context>` section.
+
+**Tool philosophy:** Tools are described by purpose and intent, not by function signature. The tinyagent framework provides JSON schemas separately. This keeps the prompt focused on *when* and *why* to use each tool:
+
+| Tool | Purpose Description |
+|------|---------------------|
+| `discover` | Natural-language code search and repository exploration |
+| `read_file` | Read file contents with content-hash tagged lines |
+| `hashline_edit` | Edit existing file using hash-validated line references |
+| `write_file` | Create a new file (fails if exists; read first, then hashline_edit) |
+| `bash` | Execute shell commands for tests, linting, git, builds |
+| `web_fetch` | Fetch public web content as readable text |
+
+**Agent version hashing:** `_compute_agent_version()` generates a cache key from configuration that affects agent behavior: `max_retries`, `tool_strict_validation`, `request_delay`, `global_request_timeout`, `max_tokens`.
+
 ## Why
 
 The `RequestOrchestrator` class exists to keep the streaming event loop testable and the callback wiring explicit. Each event type has its own handler method -- no giant switch statement.
@@ -138,3 +161,5 @@ The `RequestOrchestrator` class exists to keep the streaming event loop testable
 Compaction is request-scoped (one compaction per request at most) to avoid compacting the same history repeatedly when the model makes multiple turns.
 
 The `StateManagerProtocol` breaks the circular dependency between session state and agent creation -- agents need state, state stores agents.
+
+tinyagent provides the agent framework (migrated from pydantic-ai), handling the underlying event streaming, tool schema generation, and message protocol conversion.