chore(deps): bump claude-agent-sdk to 0.1.58 with compat proxy enabled

Dev preview PR — combines the cli_path plumbing (#12741), the
in-process compat proxy (#12745), and the SDK bump in one branch so
we can dogfood the full upgrade end-to-end.

Changes:

* `claude-agent-sdk` -> 0.1.58 (bundled CLI 2.1.97).  Gets us all the
  blocked features:
    - `exclude_dynamic_sections` cross-user prompt cache hits
      (0.1.57) — directly amplifies #12725
    - `AssistantMessage.usage` per-turn token tracking (0.1.49) —
      cost attribution
    - `task_budget` (0.1.51) — per-task cost ceiling
    - `get_context_usage()` (0.1.52) — context window monitoring
    - MCP large-tool-result truncation fix (0.1.55)
    - MCP HTTP/SSE buffer leak fix (CLI 2.1.97) — known production
      memory creep
    - 429 retry exponential-backoff fix (CLI 2.1.97) — production
      rate-limit recovery
    - `--resume` cache miss regression fix (CLI 2.1.90)
    - SDK session quadratic-write fix (CLI 2.1.90)

* `ChatConfig.claude_agent_use_compat_proxy` default flipped from
  `False` -> `True`. The bundled CLI in 0.1.55+ injects the
  `context-management-2025-06-27` beta header which OpenRouter
  rejects (anthropics/claude-agent-sdk-python#789). The proxy strips
  it transparently. Disable explicitly only if you've pinned to a
  CLI version in `_KNOWN_GOOD_BUNDLED_CLI_VERSIONS_DIRECT`.

* `sdk_compat_test.py` pin assertion split into two known-good sets:
    - `_KNOWN_GOOD_BUNDLED_CLI_VERSIONS_DIRECT` — works without the
      proxy ({"2.1.63", "2.1.70"})
    - `_KNOWN_GOOD_BUNDLED_CLI_VERSIONS_VIA_PROXY` — works only with
      the compat proxy enabled ({"2.1.97"})
  The test now requires `claude_agent_use_compat_proxy=True` for
  proxy-only versions, so disabling the proxy on a fresh checkout
  with this PR fails fast with a clear error.

Operational rollout (when ready to ship beyond dev preview):

1. Merge #12741 (plumbing + reproduction test)
2. Merge #12745 (proxy module — opt-in default off)
3. Merge this PR (bumps SDK + flips default to on)
4. Watch production for the existing reproduction test running
   continuously as a regression guard
5. If anything goes wrong: revert this PR (proxy becomes opt-in
   again, SDK back to whichever version is in the previous merge)

Dev preview usage: deploy this branch with no env-var changes —
the proxy is on by default. The reproduction test will continue
to pass against the bundled CLI 2.1.97 when (and only when) the
proxy successfully strips the forbidden patterns.

autogpt_platform/backend/backend/copilot/config.py

@@ -187,21 +187,26 @@ class ChatConfig(BaseSettings):
         "(same pattern as `api_key` / `base_url`).",
     )
     claude_agent_use_compat_proxy: bool = Field(
-        default=False,
+        default=True,
         description="Run the in-process OpenRouter compatibility proxy "
         "(`backend.copilot.sdk.openrouter_compat_proxy`) in front of the "
         "Claude Code CLI. The proxy strips `tool_reference` content "
         "blocks and the `context-management-2025-06-27` beta header / "
         "field from outgoing requests so newer SDK / CLI versions stop "
-        "tripping OpenRouter's stricter validation. Orthogonal to "
-        "`claude_agent_cli_path` — the override picks the binary, the "
-        "proxy rewrites whatever the binary sends. Reads from "
-        "`CHAT_CLAUDE_AGENT_USE_COMPAT_PROXY` or the unprefixed "
-        "`CLAUDE_AGENT_USE_COMPAT_PROXY` environment variable (same "
-        "pattern as `claude_agent_cli_path`). Only takes effect when "
-        "the session has an Anthropic-compatible upstream to forward "
-        "to — direct-Anthropic sessions skip the proxy entirely to "
-        "avoid silently re-routing through OpenRouter.",
+        "tripping OpenRouter's stricter validation. Defaults to True "
+        "because the bundled CLI in `claude-agent-sdk >= 0.1.55` requires "
+        "the proxy. Orthogonal to `claude_agent_cli_path` — the override "
+        "picks the binary, the proxy rewrites whatever the binary sends. "
+        "Disable explicitly only if you've pinned `claude-agent-sdk` to "
+        "a version whose bundled CLI is in "
+        "`_KNOWN_GOOD_BUNDLED_CLI_VERSIONS_DIRECT` (2.1.63 or 2.1.70). "
+        "Reads from `CHAT_CLAUDE_AGENT_USE_COMPAT_PROXY` or the "
+        "unprefixed `CLAUDE_AGENT_USE_COMPAT_PROXY` environment "
+        "variable (same pattern as `claude_agent_cli_path`). Only "
+        "takes effect when the session has an Anthropic-compatible "
+        "upstream to forward to — direct-Anthropic sessions skip the "
+        "proxy entirely to avoid silently re-routing through "
+        "OpenRouter.",
     )
     use_openrouter: bool = Field(
         default=True,

autogpt_platform/backend/backend/copilot/sdk/sdk_compat_test.py

@@ -232,27 +232,82 @@ def test_sdk_exports_hook_event_type(hook_event: str):
 # version, so the SDK Python API surface and the CLI binary version can
 # be picked independently.
 
-# CLI versions verified to work against OpenRouter from production
-# traffic.  When upstream lands a fix and we can confirm a newer version
-# works, add it to this set rather than blanket-removing the assertion.
-_KNOWN_GOOD_BUNDLED_CLI_VERSIONS: frozenset[str] = frozenset({"2.1.63"})
+# CLI versions verified to work against OpenRouter directly (no compat
+# proxy required) — bisected via the reproduction test in
+# `cli_openrouter_compat_test.py`.  Bundled CLI versions outside this
+# set are still allowed but ONLY when the compat proxy is enabled (see
+# the second known-good set below + the test below).
+_KNOWN_GOOD_BUNDLED_CLI_VERSIONS_DIRECT: frozenset[str] = frozenset(
+    {
+        "2.1.63",  # claude-agent-sdk 0.1.45 — original pin from PR #12294.
+        "2.1.70",  # claude-agent-sdk 0.1.47 — first version with the
+        #          tool_reference proxy detection fix; bisect-verified
+        #          OpenRouter-safe in #12742.
+    }
+)
+
+# CLI versions verified to work against OpenRouter ONLY when the
+# in-process `openrouter_compat_proxy` is enabled (which strips the
+# `tool_reference` content blocks and `context-management-2025-06-27`
+# beta from outgoing requests).  Without the proxy these CLI versions
+# trip OpenRouter's stricter validation and return 400.
+_KNOWN_GOOD_BUNDLED_CLI_VERSIONS_VIA_PROXY: frozenset[str] = frozenset(
+    {
+        "2.1.97",  # claude-agent-sdk 0.1.58 — needs `claude_agent_use_compat_proxy=True`
+        #          due to the upstream regression in
+        #          anthropics/claude-agent-sdk-python#789.
+    }
+)
+
+# Aggregate set used by the assertion below — the test allows EITHER
+# a directly-known-good CLI OR a proxy-known-good CLI when the proxy
+# is enabled in the active config.
+_KNOWN_GOOD_BUNDLED_CLI_VERSIONS: frozenset[str] = (
+    _KNOWN_GOOD_BUNDLED_CLI_VERSIONS_DIRECT | _KNOWN_GOOD_BUNDLED_CLI_VERSIONS_VIA_PROXY
+)
 
 
 def test_bundled_cli_version_is_known_good_against_openrouter():
     """Pin the bundled CLI version so accidental SDK bumps cause a loud,
-    fast failure with a pointer to the OpenRouter compatibility issue."""
+    fast failure with a pointer to the OpenRouter compatibility issue.
+
+    A CLI version that's only safe via the compat proxy is allowed only
+    when ``ChatConfig.claude_agent_use_compat_proxy`` is enabled.
+    """
     from claude_agent_sdk._cli_version import __cli_version__
 
-    assert __cli_version__ in _KNOWN_GOOD_BUNDLED_CLI_VERSIONS, (
+    from backend.copilot.config import ChatConfig
+
+    cfg = ChatConfig()
+    proxy_enabled = cfg.claude_agent_use_compat_proxy
+
+    if __cli_version__ in _KNOWN_GOOD_BUNDLED_CLI_VERSIONS_DIRECT:
+        return  # safe with or without the proxy
+
+    if __cli_version__ in _KNOWN_GOOD_BUNDLED_CLI_VERSIONS_VIA_PROXY:
+        assert proxy_enabled, (
+            f"Bundled Claude Code CLI version {__cli_version__!r} is only "
+            "OpenRouter-safe when `claude_agent_use_compat_proxy` is "
+            "enabled, but the active ChatConfig has the proxy disabled. "
+            "Either set `COPILOT__CLAUDE_AGENT_USE_COMPAT_PROXY=true` or "
+            "downgrade `claude-agent-sdk` to a version whose bundled CLI "
+            f"is in {sorted(_KNOWN_GOOD_BUNDLED_CLI_VERSIONS_DIRECT)!r}. "
+            "See https://github.com/anthropics/claude-agent-sdk-python/issues/789."
+        )
+        return
+
+    raise AssertionError(
         f"Bundled Claude Code CLI version is {__cli_version__!r}, which is "
-        f"not in the OpenRouter-known-good set "
-        f"{sorted(_KNOWN_GOOD_BUNDLED_CLI_VERSIONS)!r}. "
+        f"not in any OpenRouter-known-good set "
+        f"({sorted(_KNOWN_GOOD_BUNDLED_CLI_VERSIONS)!r}). "
         "If you intentionally bumped `claude-agent-sdk`, verify the new "
         "bundled CLI works with OpenRouter against the reproduction test "
         "in `cli_openrouter_compat_test.py`, then add the new CLI version "
-        "to `_KNOWN_GOOD_BUNDLED_CLI_VERSIONS`. If you cannot make the "
-        "bundled CLI work, set `claude_agent_cli_path` to a known-good "
-        "binary instead and skip the bundled one. See "
+        "to either `_KNOWN_GOOD_BUNDLED_CLI_VERSIONS_DIRECT` (works "
+        "without the proxy) or `_KNOWN_GOOD_BUNDLED_CLI_VERSIONS_VIA_PROXY` "
+        "(works only with `claude_agent_use_compat_proxy=true`). If you "
+        "cannot make the bundled CLI work either way, set "
+        "`claude_agent_cli_path` to a known-good binary instead. See "
         "https://github.com/anthropics/claude-agent-sdk-python/issues/789 "
         "and https://github.com/Significant-Gravitas/AutoGPT/pull/12294."
     )

autogpt_platform/backend/poetry.lock

@@ -1,4 +1,4 @@
-# This file is automatically @generated by Poetry 2.2.1 and should not be changed by hand.
+# This file is automatically @generated by Poetry 2.1.4 and should not be changed by hand.
 
 [[package]]
 name = "agentmail"
@@ -909,17 +909,18 @@ files = [
 
 [[package]]
 name = "claude-agent-sdk"
-version = "0.1.45"
+version = "0.1.58"
 description = "Python SDK for Claude Code"
 optional = false
 python-versions = ">=3.10"
 groups = ["main"]
 files = [
-    {file = "claude_agent_sdk-0.1.45-py3-none-macosx_11_0_arm64.whl", hash = "sha256:26a5cc60c3a394f5b814f6b2f67650819cbcd38c405bbdc11582b3e097b3a770"},
-    {file = "claude_agent_sdk-0.1.45-py3-none-manylinux_2_17_aarch64.whl", hash = "sha256:decc741b53e0b2c10a64fd84c15acca1102077d9f99941c54905172cd95160c9"},
-    {file = "claude_agent_sdk-0.1.45-py3-none-manylinux_2_17_x86_64.whl", hash = "sha256:7d48dcf4178c704e4ccbf3f1f4ebf20b3de3f03d0592086c1f3abd16b8ca441e"},
-    {file = "claude_agent_sdk-0.1.45-py3-none-win_amd64.whl", hash = "sha256:d1cf34995109c513d8daabcae7208edc260b553b53462a9ac06a7c40e240a288"},
-    {file = "claude_agent_sdk-0.1.45.tar.gz", hash = "sha256:97c1e981431b5af1e08c34731906ab8d4a58fe0774a04df0ea9587dcabc85151"},
+    {file = "claude_agent_sdk-0.1.58-py3-none-macosx_11_0_arm64.whl", hash = "sha256:69197950809754c4f06bba8261f2d99c3f9605b6cc1c13d3409d0eb82fb4ee64"},
+    {file = "claude_agent_sdk-0.1.58-py3-none-macosx_11_0_x86_64.whl", hash = "sha256:75d60883fc5e2070bccd8d9b19505fe16af8e049120c03821e9dc8c826cca434"},
+    {file = "claude_agent_sdk-0.1.58-py3-none-manylinux_2_17_aarch64.whl", hash = "sha256:7bf4eb0f00ec944a7b63eb94788f120dfb0460c348a525235c7d6641805acc1d"},
+    {file = "claude_agent_sdk-0.1.58-py3-none-manylinux_2_17_x86_64.whl", hash = "sha256:650d298a3d3c0dcdde4b5f1dbf52f472ff0b0ec82987b27ffa2a4e0e72928408"},
+    {file = "claude_agent_sdk-0.1.58-py3-none-win_amd64.whl", hash = "sha256:2c2130a7ffe06ed4f88d56b217a5091c91c9bcb1a69cfd94d5dcf0d2946d8c55"},
+    {file = "claude_agent_sdk-0.1.58.tar.gz", hash = "sha256:77bee8fd60be033cb870def46c2ab1625a512fa8a3de4ff8d766664ffb16d6a6"},
 ]
 
 [package.dependencies]
@@ -8928,4 +8929,4 @@ cffi = ["cffi (>=1.17,<2.0) ; platform_python_implementation != \"PyPy\" and pyt
 [metadata]
 lock-version = "2.1"
 python-versions = ">=3.10,<3.14"
-content-hash = "da61798b73758b9292fc1933268d488fbe739dc1fbf5c6586cd0c76a3411eb2e"
+content-hash = "c4cc6a0a26869a167ce182b178224554135d89d8ffa4605257d17b3f495cdf59"

autogpt_platform/backend/pyproject.toml

@@ -18,7 +18,7 @@ apscheduler = "^3.11.1"
 autogpt-libs = { path = "../autogpt_libs", develop = true }
 bleach = { extras = ["css"], version = "^6.2.0" }
 cachetools = "^5.5.0"
-claude-agent-sdk = "0.1.45"  # see copilot/sdk/sdk_compat_test.py for capability checks
+claude-agent-sdk = "0.1.58"  # latest stable; bundled CLI 2.1.97 ships the broken context-management beta and REQUIRES the openrouter_compat_proxy. See sdk_compat_test.py.
 click = "^8.2.0"
 cryptography = "^46.0"
 discord-py = "^2.5.2"