agentbridge-cli 0.1.13__tar.gz → 0.1.15__tar.gz

{agentbridge_cli-0.1.13 → agentbridge_cli-0.1.15}/.codex/skills/build-release/SKILL.md

@@ -9,7 +9,7 @@ Use the repository-owned GitHub Actions workflow and PyPI Trusted Publishing.
 Do not create a local tag or upload with local credentials: pushing a new
 `pyproject.toml` version to `main` triggers `.github/workflows/release.yml`,
 which tests, scans, builds, creates `v<version>`, creates the GitHub Release,
-publishes `agentbridge-cli`, and attaches ad-hoc-signed arm64 and x86_64 DMGs.
+publishes `agentbridge-cli`, and attaches an ad-hoc-signed arm64 DMG.
 
 Use the next unused patch version unless the user requests another valid,
 unused semantic version. Treat an untagged version absent from PyPI as pending.
@@ -54,8 +54,8 @@ The preflight requires macOS. It checks consistent project and lock metadata, an
 unused PyPI version and tag, clean patch formatting, a frozen lock, Python 3.12
 and 3.13 tests, Ruff, Swift tests, a self-contained app/DMG build for the current
 Mac architecture, app signature and embedded-runtime smoke tests, wheel and
-sdist audits, and a wheel installation smoke test. CI builds both macOS
-architectures with ad-hoc signatures. Stop at the exact failed gate.
+sdist audits, and a wheel installation smoke test. CI builds the arm64 macOS
+artifact with an ad-hoc signature. Stop at the exact failed gate.
 
 4. Review and publish the release commit:
 
@@ -92,9 +92,7 @@ python3 .codex/skills/build-release/scripts/release_build.py wait-pypi --version
 gh release view v<version> --json url,tagName,assets
 ```
 
-Require both
-`AgentBridge-<version>-macos-arm64.dmg` and
-`AgentBridge-<version>-macos-x86_64.dmg` plus their `.sha256` files in the
+Require `AgentBridge-<version>-macos-arm64.dmg` and its `.sha256` file in the
 release assets. Do not report success until PyPI returns exact-version files and
 all required GitHub assets exist. If the workflow fails, inspect
 `gh run view <run-id> --log-failed` and report the failed gate before considering

{agentbridge_cli-0.1.13 → agentbridge_cli-0.1.15}/.codex/skills/build-release/scripts/release_build.py

@@ -313,8 +313,10 @@ def preflight_macos_app(version: str) -> dict[str, object]:
         raise SystemExit("the AgentBridge release preflight requires macOS for app validation")
 
     architecture = platform.machine()
-    if architecture not in {"arm64", "x86_64"}:
-        raise SystemExit(f"unsupported macOS release architecture: {architecture}")
+    if architecture != "arm64":
+        raise SystemExit(
+            f"unsupported macOS release architecture: {architecture}; AgentBridge requires arm64"
+        )
 
     run(["swift", "test", "--package-path", "macos"])
     run(["scripts/build_macos_app.sh"])

{agentbridge_cli-0.1.13 → agentbridge_cli-0.1.15}/.github/workflows/release.yml

@@ -214,8 +214,6 @@ jobs:
         include:
           - runner: macos-15
             architecture: arm64
-          - runner: macos-15-intel
-            architecture: x86_64
     runs-on: ${{ matrix.runner }}
     timeout-minutes: 30
     permissions:

{agentbridge_cli-0.1.13 → agentbridge_cli-0.1.15}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: agentbridge-cli
-Version: 0.1.13
+Version: 0.1.15
 Summary: Bridge OpenAI tools to Claude Code SDK, Codex CLI, and OpenRouter
 Project-URL: Homepage, https://github.com/tsilva/agentbridge
 Project-URL: Repository, https://github.com/tsilva/agentbridge
@@ -46,10 +46,10 @@ It supports streaming and non-streaming responses, image and PDF inputs where th
 
 ### macOS menu-bar app
 
-Download the DMG for your Mac architecture from the GitHub Release, drag
+On an Apple silicon Mac, download the arm64 DMG from the GitHub Release, drag
 `AgentBridge.app` to Applications, and open it. The app is ad-hoc signed rather
 than Apple-notarized, so on first launch macOS may require you to Control-click
-the app, choose **Open**, and confirm.
+the app, choose **Open**, and confirm. Intel Macs are not supported.
 
 You can also build the app locally on macOS 13 or later. This requires
 [uv](https://docs.astral.sh/uv/) and the Xcode Command Line Tools, but not an
@@ -177,6 +177,7 @@ uv build                                              # build wheel and source d
 - Claude clients are created lazily, reused by model, and capped by the worker count. Claude sessions do not load filesystem settings and run with built-in tools disabled.
 - Codex runs one ephemeral `codex exec` process per request in a temporary directory with read-only sandboxing, no approvals, and project rules ignored. Multimodal structured-output calls also ignore user config and disable execution and image-generation tools. Native image calls use the same strict profile, keep execution disabled, and enable the image-generation capability needed for the edit.
 - Claude and Codex function calls are represented through prompted JSON; OpenRouter tool calls pass through its SDK. Session logs and extracted image or PDF attachments are saved under `~/.config/agentbridge/logs/sessions` by default.
+- Streaming provider failures emit an OpenAI-shaped `error` object as an SSE `data` event before `[DONE]`; they are not returned as assistant message text.
 - Set `store: false` on chat requests to suppress session-log artifacts. The native image route requires `store: false`, accepts data URLs only, validates both rasters, locates the result from the structured Codex thread ID, and removes that thread's generated-image directory after the request.
 
 ## Publishing
@@ -185,8 +186,8 @@ Releases use the `Release` GitHub Actions workflow and PyPI Trusted Publishing
 for the `agentbridge-cli` project. The publisher is scoped to owner `tsilva`,
 repository `agentbridge`, workflow `release.yml`, and environment `pypi`; no
 PyPI API token is required. GitHub Releases contain the Python distributions
-plus ad-hoc-signed arm64 and x86_64 macOS DMGs and SHA-256 checksums. These DMGs
-are not Apple-notarized and require no Apple Developer credentials. Releases
+plus an ad-hoc-signed arm64 macOS DMG and SHA-256 checksum. The DMG is not
+Apple-notarized and requires no Apple Developer credentials. Releases
 through `0.1.10` remain available under the previous `agentbridge-py`
 distribution name.
 

{agentbridge_cli-0.1.13 → agentbridge_cli-0.1.15}/README.md

@@ -14,10 +14,10 @@ It supports streaming and non-streaming responses, image and PDF inputs where th
 
 ### macOS menu-bar app
 
-Download the DMG for your Mac architecture from the GitHub Release, drag
+On an Apple silicon Mac, download the arm64 DMG from the GitHub Release, drag
 `AgentBridge.app` to Applications, and open it. The app is ad-hoc signed rather
 than Apple-notarized, so on first launch macOS may require you to Control-click
-the app, choose **Open**, and confirm.
+the app, choose **Open**, and confirm. Intel Macs are not supported.
 
 You can also build the app locally on macOS 13 or later. This requires
 [uv](https://docs.astral.sh/uv/) and the Xcode Command Line Tools, but not an
@@ -145,6 +145,7 @@ uv build                                              # build wheel and source d
 - Claude clients are created lazily, reused by model, and capped by the worker count. Claude sessions do not load filesystem settings and run with built-in tools disabled.
 - Codex runs one ephemeral `codex exec` process per request in a temporary directory with read-only sandboxing, no approvals, and project rules ignored. Multimodal structured-output calls also ignore user config and disable execution and image-generation tools. Native image calls use the same strict profile, keep execution disabled, and enable the image-generation capability needed for the edit.
 - Claude and Codex function calls are represented through prompted JSON; OpenRouter tool calls pass through its SDK. Session logs and extracted image or PDF attachments are saved under `~/.config/agentbridge/logs/sessions` by default.
+- Streaming provider failures emit an OpenAI-shaped `error` object as an SSE `data` event before `[DONE]`; they are not returned as assistant message text.
 - Set `store: false` on chat requests to suppress session-log artifacts. The native image route requires `store: false`, accepts data URLs only, validates both rasters, locates the result from the structured Codex thread ID, and removes that thread's generated-image directory after the request.
 
 ## Publishing
@@ -153,8 +154,8 @@ Releases use the `Release` GitHub Actions workflow and PyPI Trusted Publishing
 for the `agentbridge-cli` project. The publisher is scoped to owner `tsilva`,
 repository `agentbridge`, workflow `release.yml`, and environment `pypi`; no
 PyPI API token is required. GitHub Releases contain the Python distributions
-plus ad-hoc-signed arm64 and x86_64 macOS DMGs and SHA-256 checksums. These DMGs
-are not Apple-notarized and require no Apple Developer credentials. Releases
+plus an ad-hoc-signed arm64 macOS DMG and SHA-256 checksum. The DMG is not
+Apple-notarized and requires no Apple Developer credentials. Releases
 through `0.1.10` remain available under the previous `agentbridge-py`
 distribution name.
 

agentbridge_cli-0.1.15/agentbridge/_build_info.py

@@ -0,0 +1 @@
+GIT_HASH = "690c228"

{agentbridge_cli-0.1.13 → agentbridge_cli-0.1.15}/agentbridge/dashboard.py

@@ -225,6 +225,8 @@ def _parse_log_file(path: Path) -> dict | None:
 
 def _get_recent_logs(limit: int = 20) -> list[dict]:
     """Read recent session log files, newest first."""
+    if limit <= 0:
+        return []
     log_dir = session_log_dir()
     if not log_dir.exists():
         return []
@@ -242,10 +244,12 @@ def _get_recent_logs(limit: int = 20) -> list[dict]:
     )
 
     results = []
-    for path in log_files[:limit]:
+    for path in log_files:
         parsed = _parse_log_file(path)
         if parsed is not None:
             results.append(parsed)
+            if len(results) >= limit:
+                break
 
     return results
 
@@ -366,10 +370,11 @@ def create_dashboard_router(
             if log.get("duration_ms") is None:
                 log["duration_ms"] = timing.get("duration_ms", 0)
             if log.get("input_tokens") is None:
-                usage_data = log.get("usage")
-                if usage_data:
-                    log["input_tokens"] = usage_data.get("input_tokens")
-                    log["output_tokens"] = usage_data.get("output_tokens")
+                usage_data = (
+                    log.get("usage") if isinstance(log.get("usage"), dict) else {}
+                )
+                log["input_tokens"] = usage_data.get("input_tokens")
+                log["output_tokens"] = usage_data.get("output_tokens")
 
         merged = active + completed
         return merged[:limit]
@@ -441,8 +446,8 @@ def create_dashboard_router(
         if parsed is None:
             raise HTTPException(status_code=404, detail="Request not found")
 
-        timing = parsed.get("timing", {})
-        usage = parsed.get("usage", {})
+        timing = parsed.get("timing") if isinstance(parsed.get("timing"), dict) else {}
+        usage = parsed.get("usage") if isinstance(parsed.get("usage"), dict) else {}
         return templates.TemplateResponse(
             request,
             "detail.html",

{agentbridge_cli-0.1.13 → agentbridge_cli-0.1.15}/agentbridge/server.py

@@ -686,6 +686,23 @@ OPENROUTER_TIMEOUT = int(
 )
 
 
+def _stream_error_event(
+    message: str,
+    *,
+    error_type: str = "server_error",
+    code: str | None = None,
+) -> str:
+    """Return a machine-readable OpenAI-shaped error in an SSE data event."""
+    payload = ErrorResponse(
+        error=ErrorDetail(
+            message=message,
+            type=error_type,
+            code=code,
+        )
+    )
+    return f"data: {payload.model_dump_json()}\n\n"
+
+
 class BridgeHTTPException(HTTPException):
     """HTTPException with request_id for error tracing."""
 
@@ -1863,18 +1880,13 @@ async def stream_openrouter_api(
         )
         dashboard_state.request_errored(request_id, f"{type(e).__name__}: {error}")
         _dashboard_handled = True
-        error_chunk = ChatCompletionChunk(
-            id=request_id,
-            created=created,
-            model=request.model,
-            choices=[
-                StreamChoice(
-                    delta=DeltaMessage(content="\n\n[Error: OpenRouter request failed.]"),
-                    finish_reason=None,
-                )
-            ],
+        yield _stream_error_event(
+            "OpenRouter request timed out."
+            if isinstance(e, asyncio.TimeoutError)
+            else "OpenRouter request failed.",
+            error_type="timeout_error" if isinstance(e, asyncio.TimeoutError) else "server_error",
+            code="timeout_error" if isinstance(e, asyncio.TimeoutError) else "provider_error",
         )
-        yield f"data: {error_chunk.model_dump_json()}\n\n"
         yield "data: [DONE]\n\n"
     finally:
         if not _dashboard_handled:
@@ -2186,23 +2198,11 @@ async def stream_claude_sdk(
         )
         dashboard_state.request_errored(request_id, f"Timeout after {CLAUDE_TIMEOUT}s")
         _dashboard_handled = True
-        error_chunk = ChatCompletionChunk(
-            id=request_id,
-            created=created,
-            model=model,
-            choices=[
-                StreamChoice(
-                    delta=DeltaMessage(
-                        content=(
-                            "\n\n[Error: Request timed out. Increase CLAUDE_TIMEOUT "
-                            "env var for longer requests.]"
-                        )
-                    ),
-                    finish_reason=None,
-                )
-            ],
+        yield _stream_error_event(
+            "Request timed out. Increase CLAUDE_TIMEOUT env var for longer requests.",
+            error_type="timeout_error",
+            code="timeout_error",
         )
-        yield f"data: {error_chunk.model_dump_json()}\n\n"
         yield "data: [DONE]\n\n"
         return
     except HTTPException as he:
@@ -2213,16 +2213,11 @@ async def stream_claude_sdk(
         )
         dashboard_state.request_errored(request_id, f"HTTP {he.status_code}: {he.detail}")
         _dashboard_handled = True
-        error_chunk = ChatCompletionChunk(
-            id=request_id,
-            created=created,
-            model=model,
-            choices=[StreamChoice(
-                delta=DeltaMessage(content=f"\n\n[Error: {he.detail}]"),
-                finish_reason=None,
-            )],
+        yield _stream_error_event(
+            str(he.detail),
+            error_type="invalid_request_error" if he.status_code < 500 else "server_error",
+            code="invalid_request_error" if he.status_code < 500 else "server_error",
         )
-        yield f"data: {error_chunk.model_dump_json()}\n\n"
         yield "data: [DONE]\n\n"
         return
     except Exception as e:
@@ -2237,16 +2232,11 @@ async def stream_claude_sdk(
         )
         dashboard_state.request_errored(request_id, f"{type(e).__name__}: {e}")
         _dashboard_handled = True
-        error_chunk = ChatCompletionChunk(
-            id=request_id,
-            created=created,
-            model=model,
-            choices=[StreamChoice(
-                delta=DeltaMessage(content="\n\n[Error: An internal error occurred.]"),
-                finish_reason=None,
-            )],
+        yield _stream_error_event(
+            "An internal error occurred.",
+            error_type="server_error",
+            code="server_error",
         )
-        yield f"data: {error_chunk.model_dump_json()}\n\n"
         yield "data: [DONE]\n\n"
         return
     finally:
@@ -2482,23 +2472,11 @@ async def stream_codex_cli(
         )
         dashboard_state.request_errored(request_id, f"Timeout after {CODEX_TIMEOUT}s")
         _dashboard_handled = True
-        error_chunk = ChatCompletionChunk(
-            id=request_id,
-            created=created,
-            model=model,
-            choices=[
-                StreamChoice(
-                    delta=DeltaMessage(
-                        content=(
-                            "\n\n[Error: Request timed out. Increase CODEX_TIMEOUT "
-                            "env var for longer requests.]"
-                        )
-                    ),
-                    finish_reason=None,
-                )
-            ],
+        yield _stream_error_event(
+            "Request timed out. Increase CODEX_TIMEOUT env var for longer requests.",
+            error_type="timeout_error",
+            code="timeout_error",
         )
-        yield f"data: {error_chunk.model_dump_json()}\n\n"
         yield "data: [DONE]\n\n"
         return
     except Exception as e:
@@ -2514,16 +2492,11 @@ async def stream_codex_cli(
         )
         dashboard_state.request_errored(request_id, f"{type(e).__name__}: {e}")
         _dashboard_handled = True
-        error_chunk = ChatCompletionChunk(
-            id=request_id,
-            created=created,
-            model=model,
-            choices=[StreamChoice(
-                delta=DeltaMessage(content="\n\n[Error: An internal error occurred.]"),
-                finish_reason=None,
-            )],
+        yield _stream_error_event(
+            "An internal error occurred.",
+            error_type="server_error",
+            code="server_error",
         )
-        yield f"data: {error_chunk.model_dump_json()}\n\n"
         yield "data: [DONE]\n\n"
         return
     finally: