feat: Anthropic SDK backend + per-model backend selection by anticomputer · Pull Request #265 · GitHubSecurityLab/seclab-taskflow-agent

anticomputer · 2026-06-11T15:31:06Z

Summary

Adds anthropic_sdk as a third backend adapter and enables per-model backend selection so different models in a single taskflow can use different SDKs.

Changes

New backend: `anthropic_sdk`

Drives the native Anthropic Messages API (/v1/messages) via the official anthropic Python SDK. Implemented in sdk/anthropic_sdk/backend.py (284 lines), implementing the existing AgentBackend protocol.

Capabilities:

Streaming text output
Multi-turn MCP tool calling (sequential tool call execution)
Adaptive thinking with configurable reasoning.effort (low/medium/high/max)
stream_thinking model_settings option (opt-in, default off)
Error translation to existing Backend*Error types

Auth: The Anthropic SDK sends x-api-key by default. Providers that use Bearer auth (APIProvider.bearer_auth=True -- includes CAPI, GitHub Models, OpenAI, and any custom proxy registered with bearer_auth=True) get Authorization: Bearer via default_headers with a placeholder api_key. Unknown endpoints (e.g. direct api.anthropic.com) default to native SDK auth via x-api-key. Token resolution uses get_AI_token() (AI_API_TOKEN then COPILOT_TOKEN fallback) with optional per-model token env-var override.

Thinking: Uses thinking.type: adaptive with output_config.effort, matching CAPI requirements. CAPI currently returns encrypted thinking signatures (content not readable through the proxy), but the stream_thinking flag is ready for when thinking content is exposed.

Per-model backend selection

The backend key can now be set in per-model model_settings, not just globally on the model config document. This allows mixed-backend taskflows:

models:
  code_analysis: claude-opus-4.7
  general_tasks: gpt-5.4-mini
model_settings:
  code_analysis:
    backend: anthropic_sdk
    reasoning:
      effort: high
  general_tasks:
    api_type: responses
    backend: openai_agents

_resolve_task_model() now returns a 6th element (per-task backend override), which deploy_task_agents uses with fallback to the global backend.

Dependency changes

Both anthropic and github-copilot-sdk moved from optional to regular dependencies. With per-model backend config, any SDK could be needed at runtime, so optional installs no longer make sense.

Testing

All tests performed against CAPI with the pipeline AI_API_TOKEN:

Test	Result
Basic messages (no tools)	Pass
Streaming text output	Pass
Single-turn MCP tool calling (echo)	Pass
Multi-turn tool calls (3x echo)	Pass
Mixed-backend taskflow (Anthropic + OpenAI)	Pass
All reasoning efforts (low/medium/high/max)	Pass
No reasoning config	Pass
max_tokens override	Pass
Custom endpoint override	Pass
Invalid model (error handling)	Pass
Invalid backend name (error handling)	Pass
OpenAI backend regression	Pass
Default backend (no backend field) regression	Pass

Files changed

src/seclab_taskflow_agent/sdk/anthropic_sdk/ -- new backend adapter
src/seclab_taskflow_agent/sdk/__init__.py -- registry for third backend
src/seclab_taskflow_agent/models.py -- extended BackendSdk and ApiType Literals
src/seclab_taskflow_agent/runner.py -- per-model backend in _resolve_task_model(), threaded through deploy_task_agents()
pyproject.toml -- anthropic + copilot-sdk as regular deps
README.md -- updated backend docs
doc/GRAMMAR.md -- updated model_settings docs

Adds anthropic_sdk as a third backend adapter driving the native Anthropic Messages API (/v1/messages) via the official anthropic Python SDK. Supports streaming, MCP tool calling, and adaptive thinking with configurable reasoning effort. Key changes: - New backend: sdk/anthropic_sdk/backend.py implementing AgentBackend - Per-model backend selection via model_settings.backend (allows mixed backends in a single taskflow, e.g. Anthropic for code_analysis + OpenAI for general_tasks) - Both anthropic and github-copilot-sdk are now regular dependencies (not optional) since per-model backend config means any SDK could be needed at runtime - BackendSdk/ApiType Literals extended for anthropic_sdk/messages - _resolve_task_model() returns per-task backend override - stream_thinking model_settings option (opt-in, default off) - README and GRAMMAR.md updated with backend docs Auth: CAPI's /v1/messages expects Authorization: Bearer (not x-api-key); the adapter passes the bearer header via default_headers. Thinking: Uses adaptive thinking with output_config.effort. CAPI returns encrypted thinking signatures (content not readable); the stream_thinking flag is ready for when/if thinking content is exposed. Tested: basic messages, streaming, multi-turn tool calling via MCP, mixed-backend taskflows, all reasoning effort levels (low/medium/high/ max), error handling, openai_agents regression. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Pull request overview

Adds a third SDK adapter (anthropic_sdk) and allows selecting the backend per model so a single taskflow can mix providers/SDKs while sharing the same runner + MCP/tooling surface.

Changes:

Introduces AnthropicSDKBackend (Anthropic Messages API via anthropic SDK) with streaming and MCP tool-calling support.
Extends model resolution to support per-model backend overrides and threads that through agent deployment.
Promotes anthropic and github-copilot-sdk to regular dependencies and updates user-facing docs/grammar accordingly.

Show a summary per file

File	Description
src/seclab_taskflow_agent/sdk/anthropic_sdk/backend.py	New Anthropic SDK backend adapter implementing the shared backend protocol.
src/seclab_taskflow_agent/sdk/anthropic_sdk/init.py	Adds package marker/docstring for the new backend module.
src/seclab_taskflow_agent/sdk/init.py	Registers `anthropic_sdk` as a known backend in the lazy backend factory.
src/seclab_taskflow_agent/runner.py	Adds per-task/per-model backend override support to model resolution and deployment.
src/seclab_taskflow_agent/models.py	Extends `ApiType` and `BackendSdk` literals to include the new options.
README.md	Documents the new backend and per-model backend selection precedence.
pyproject.toml	Moves `anthropic` and `github-copilot-sdk` into core dependencies; removes the copilot extra.
doc/GRAMMAR.md	Updates grammar docs to include `backend` in `model_settings` and the new `messages` api_type.

Copilot's findings

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Files reviewed: 8/8 changed files
Comments generated: 7

- Remove unused json import (lint/CodeQL) - Validate reasoning.effort against allowed values upfront - Pass through temperature/top_p to Anthropic API - Add exclude_from_context support (stop after tool results) - Thread exclude_from_context into _AnthropicHandle Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

MCP tools can have description=None; the Anthropic API requires a valid string. Fall back to tool name when description is None. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Copilot's findings

Files reviewed: 8/8 changed files
Comments generated: 4

- Update doc examples to use claude-opus-4.7 and show api_type: messages - Add tests/test_sdk_anthropic_adapter.py (18 tests covering validate, tool conversion, token resolution, tool result parsing) - Fix test_runner.py: update _resolve_task_model unpacking to 6-tuple - Fix test_sdk_base.py: update backend resolution tests to match new behavior (endpoint no longer auto-selects copilot_sdk) - Add test for explicit anthropic_sdk backend selection Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Copilot's findings

Files reviewed: 11/11 changed files
Comments generated: 1

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Copilot's findings

Files reviewed: 11/11 changed files
Comments generated: 1

Allows the backend to work with both CAPI (Authorization: Bearer) and direct Anthropic endpoints (x-api-key) without code changes. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Access the MCP session directly to get the raw tool list, bypassing the openai-agents tool_filter which requires run_context/agent args not available outside its run loop. Apply blocked_tools filtering and namespace prefixing in our own code. Tested: blocked tool correctly hidden from model, unblocked tools work normally. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Copilot's findings

Files reviewed: 12/12 changed files
Comments generated: 3

The anthropic backend was reaching into openai-agents private attrs (`_obj`, `.session`) to bypass tool_filter at tool-enumeration time. This required duplicating the namespace-prefix logic that already lives on MCPNamespaceWrap and risked double-prefixing on the fallback path. Move the 'list tools without invoking the agent-side tool_filter' logic into MCPNamespaceWrap.list_tools_unfiltered(), where the wrapper already owns its namespace and session reference. The anthropic backend becomes a one-liner; double-prefix risk is eliminated; openai-agents internal access is centralized in one place (mcp_utils.py). Also bump default_model in the provider registry from gpt-4.1 to gpt-5.5 (Copilot and OpenAI direct), openai/gpt-4.1 to openai/gpt-5.5 (GitHub Models). Only affects callers who do not specify a model -- the audit pipeline always specifies models via model_config, so this is purely a fallback for community users. Tests added: tests/test_mcp_utils.py (6 tests covering prefix correctness, no-double-prefix, tool attribute preservation, missing-session error, caller-state isolation, regression of existing list_tools()). Tests updated: test_capi_extended.py (default_model assertions). 274 tests pass, ruff clean. Local audit on anticomputer/vulnerable-test-app produced 4 vulnerabilities (verifying MCP tools enumerated + called correctly). Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

- Remove quotes from _FakeTool return type (UP037) - Use raw string for regex pattern in pytest.raises match (RUF043) Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Copilot's findings

Files reviewed: 15/15 changed files
Comments generated: 2

The default_model for the OpenAI direct provider was bumped to gpt-5.5 in the bearer_auth refactor, but _OpenAIProvider.check_tool_calls()'s prefix allowlist still only matched gpt-3.5/gpt-4/o-series. This meant supports_tool_calls('gpt-5.5', ...) returned False, so list_tool_call_models() would omit the default model from the catalog output -- a contradiction with the model being the configured default. Add 'gpt-5' to the prefix tuple and a regression test covering gpt-5, gpt-5.5, gpt-5.5-mini, and a hypothetical gpt-5.6. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

README.md: Add the per-task backend override as the highest-precedence selection level. Tasks can put 'backend:' in their own model_settings block to override the model-level value, per _resolve_task_model(). doc/GRAMMAR.md: Tighten the 'passed through to the selected SDK backend' claim. openai_agents accepts the standard OpenAI parameter set, anthropic_sdk forwards a curated subset (temperature, top_p, reasoning, max_tokens, stream_thinking), and copilot_sdk consumes only its own exposed keys (e.g. reasoning_effort) and silently ignores the rest. Avoid misleading users about arbitrary key forwarding. mcp_utils.py: Make list_tools_unfiltered idempotent on the prefix. Strip an existing namespace prefix before re-applying so the method is safe to call repeatedly even if the underlying session somehow returns a cached/reused tool object whose name was previously namespaced. Uses str.removeprefix() (no-op when prefix is absent). Regression test added covering the previously-prefixed-input path. 276 tests pass, hatch fmt clean. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Copilot's findings

Files reviewed: 15/15 changed files
Comments generated: 2

Adds 'cache_control: {type: ephemeral}' to messages.stream() calls. The API auto-places a cache breakpoint at the longest cacheable prefix (tools + system + accumulated messages) and moves it forward on each turn -- multi-turn agent loops get cache reads on every turn after the first. Default-on because all current Claude models support cache_control and CAPI accepts it (validated end-to-end against claude-mythos-5 via CAPI on 2026-06-12). Callers pointed at proxies that strip / reject cache_control can opt out with 'prompt_caching: false' in model_settings. A string value (e.g. 'prompt_caching: 1h') sets a custom TTL. Local validation against anticomputer/vulnerable-test-app on the same audit pipeline, same model config, only changing prompt_caching: metric | off | on | delta --------------------+-----------+-----------+------------ requests | 60 | 62 | +2 (noise) input tokens fresh | 909,806 | 124 | -99.99% cache read tokens | 0 | 728,079 | new cache write tokens | 0 | 210,261 | new output tokens | 42,300 | 44,933 | similar vulnerabilities | 4 | 5 | +1 est. mythos cost | $11.21 | $5.60 | -50% Same or better audit quality, half the token cost. Real audits with larger system prompts + more tool definitions amortize the cache writes over more reads, so production savings are typically larger than 50%. Tests added: - prompt_caching default-on emits cache_control - prompt_caching=False suppresses cache_control (opt-out) - prompt_caching='1h' includes the ttl field 23 tests pass total, hatch fmt clean. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Reviewer was correct that blocked_tools was effectively a no-op in the anthropic_sdk backend: taskflow YAML supplies raw tool names like 'read_file', but list_tools_unfiltered() returns namespace-prefixed names like '{hash}read_file'. The old 'tool.name not in blocked' check never matched, silently letting every blocked tool through. This is the security bug the reviewer flagged on PR #265. Fix: match the raw name against the un-prefixed portion of each tool's namespaced name, in addition to the literal name. The mcp_server_map keys stay namespaced because that's what Anthropic sends in tool_use. Regression tests: - raw 'read_file' filters out '{hash}read_file' (the bug case) - already-namespaced names still match (backwards compat) doc/GRAMMAR.md: also fix an inaccuracy the reviewer flagged in the same review pass -- the docs claimed copilot_sdk 'silently ignores' unsupported model_settings keys, but it actually raises BackendCapabilityError on 'temperature' and 'parallel_tool_calls' at validate() time. Updated wording to distinguish 'ignored' (anthropic_sdk) from 'rejected' (copilot_sdk) so users aren't surprised by a hard fail when they expected a silent drop. 281 tests pass, hatch fmt clean. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Coworker review flagged that gpt-5.5 is not a viable default: - gpt-5 family models require the responses API, but APIProvider has no api_type field to signal that — callers using the default would silently hit the wrong endpoint shape - GitHub Models never received gpt-5.5; gpt-4.1 is what's still supported there, so 'openai/gpt-5.5' would 404 - Most callers specify models explicitly via model_config anyway, so the default is only a fallback safety net — keep it on a model that exists on all three providers Reverts the registry defaults and dataclass default; keeps the gpt-5 prefix in _OpenAIProvider._CHAT_PREFIXES (direct OpenAI API does serve gpt-5 family, and the prefix check is independent of default selection). Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

anticomputer · 2026-06-15T14:57:22Z

Thanks for the catch — reverted the default_model bump in c6ef3ae. You're right on both counts: gpt-5 family needs the responses API (which APIProvider doesn't signal, so callers hitting the default would silently get the wrong endpoint shape), and GitHub Models doesn't carry gpt-5.5 so openai/gpt-5.5 would 404. Kept gpt-5 in _OpenAIProvider._CHAT_PREFIXES since the direct OpenAI API does serve it and that allowlist is independent of which default we ship. Agreed it's a small thing given Taskflows almost always specifies the model, but better to leave the safety net pointing at something that actually works on all three providers.

Copilot

Copilot's findings

Files reviewed: 15/15 changed files
Comments generated: 4

Three behavior fixes flagged by review or by re-reading the diff: 1. 4xx exception mapping (reviewer-flagged): previously only anthropic.BadRequestError (400) was mapped to BackendBadRequestError. Auth (401), permission (403), not-found (404), conflict (409), unprocessable (422) all fell through to BackendUnexpectedError and surfaced as 'Agent Exception' instead of a clean request error. Catch anthropic.APIStatusError and map any 4xx status to BackendBadRequestError; 5xx still falls through to BackendUnexpectedError (the request was well-formed). 2. Empty-token failure mode: build() now raises BackendBadRequestError with a clear message when no API token can be resolved, instead of either leaking RuntimeError from get_AI_token() or letting the Anthropic client be constructed with an empty 'Bearer ' header (which produces an opaque 401 mid-stream much later). 3. Stale module docstring in sdk/__init__.py: said 'Two backends are supported' and referenced the removed '[copilot]' optional-extra. Updated to reflect the current three-backend reality. Test cleanup (reviewer-flagged): - DRY'd 3x duplicate _FakeStreamCtx boilerplate in the prompt-caching tests into a single _make_fake_client() helper at the top of the file. The helper uses a proper empty async iterator class instead of the 'return; yield' empty-generator pattern the reviewer flagged as awkward. Added regression coverage: - test_build_raises_bad_request_when_no_token_available - test_4xx_api_status_errors_map_to_bad_request (parameterized over 400/401/403/404/409/422) - test_5xx_api_status_errors_map_to_unexpected 281 -> 289 passing; lint clean (hatch fmt --linter --check). Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Copilot's findings

Files reviewed: 15/15 changed files
Comments generated: 2

anticomputer · 2026-06-15T16:35:10Z

+    for c in content:
+        text = getattr(c, "text", None)
+        if text:
+            parts.append(text)
+    return "\n".join(parts) if parts else str(result)


Good catch — fixed in 4f1f440. Changed if text: to if text is not None: so explicit empty strings are preserved verbatim. The str(result) fallback now only fires when there are genuinely no text-bearing content blocks at all (not when all blocks have empty text). Added two regression tests covering an only-empty result and an empty among non-empty siblings.

anticomputer · 2026-06-15T16:35:11Z

+    # Clear every token-source env var the standard chain consults
+    for var in ("AI_API_TOKEN", "OPENAI_API_KEY", "AZURE_OPENAI_API_KEY",
+                "ANTHROPIC_API_KEY", "GITHUB_TOKEN", "GH_TOKEN"):
+        monkeypatch.delenv(var, raising=False)


Real flakiness risk — fixed in 4f1f440. Trimmed the cargo-culted list of unrelated API keys down to just the two variables that capi.get_AI_token() actually consults (AI_API_TOKEN and COPILOT_TOKEN) and explicitly delete both. Docstring updated to call out the dependency on the token chain so the next person editing get_AI_token knows to update the test too.

Two more reviewer-flagged issues: 1. _call_tool_result_to_text() dropped empty TextContent The truthy check 'if text:' treated TextContent(text='') the same as text=None and skipped it. With an only-empty content list, parts would be [] and the helper fell through to the str(result) fallback (which is a noisy repr of the result object) instead of returning the actual empty result the tool reported. Fix: 'if text is not None:' preserves explicit empty strings; the str(result) fallback now only fires when there are no text-bearing blocks at all. 2. test_build_raises_bad_request_when_no_token_available was flaky The test cleared a long list of API key env vars (defensive cargo cult) but missed COPILOT_TOKEN, which is the second variable that capi.get_AI_token() consults. On runners with COPILOT_TOKEN set (e.g. CI envs authed to Copilot), the test would unexpectedly find a token and the assertion would fail. Simplified to clear only the two vars the chain actually consults: AI_API_TOKEN and COPILOT_TOKEN. +2 regression tests for empty-string preservation; 291 passing. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot AI review requested due to automatic review settings June 11, 2026 15:31

Copilot started reviewing on behalf of anticomputer June 11, 2026 15:31 View session

github-advanced-security AI found potential problems Jun 11, 2026

View reviewed changes

Comment thread src/seclab_taskflow_agent/sdk/anthropic_sdk/backend.py Fixed

Comment thread src/seclab_taskflow_agent/sdk/anthropic_sdk/backend.py Fixed

Copilot AI reviewed Jun 11, 2026

View reviewed changes

anticomputer and others added 2 commits June 11, 2026 11:38

fix: handle None tool descriptions in Anthropic tool conversion

ed4412f

MCP tools can have description=None; the Anthropic API requires a valid string. Fall back to tool name when description is None. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot AI review requested due to automatic review settings June 11, 2026 15:41

Copilot started reviewing on behalf of anticomputer June 11, 2026 15:41 View session

Copilot AI reviewed Jun 11, 2026

View reviewed changes

Comment thread src/seclab_taskflow_agent/runner.py

Comment thread src/seclab_taskflow_agent/runner.py

Comment thread pyproject.toml

Comment thread src/seclab_taskflow_agent/sdk/anthropic_sdk/backend.py

github-advanced-security AI found potential problems Jun 11, 2026

View reviewed changes

Comment thread tests/test_sdk_anthropic_adapter.py Fixed

fix: lint errors in test file (unused imports, N803 camelCase)

b6f0057

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot AI review requested due to automatic review settings June 11, 2026 16:01

Copilot started reviewing on behalf of anticomputer June 11, 2026 16:02 View session

Copilot AI reviewed Jun 11, 2026

View reviewed changes

Comment thread tests/test_runner.py

test: add backend extraction coverage to _resolve_task_model tests

266c54b

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

anticomputer marked this pull request as ready for review June 11, 2026 17:17

Copilot AI review requested due to automatic review settings June 11, 2026 17:17

anticomputer requested review from JarLob, Kwstubbs, kevinbackhouse, p- and sylwia-budzynska as code owners June 11, 2026 17:17

Copilot started reviewing on behalf of anticomputer June 11, 2026 17:17 View session

Copilot AI reviewed Jun 11, 2026

View reviewed changes

Comment thread tests/test_sdk_anthropic_adapter.py

anticomputer and others added 2 commits June 11, 2026 15:01

fix: pass real token as api_key instead of placeholder

4417279

Allows the backend to work with both CAPI (Authorization: Bearer) and direct Anthropic endpoints (x-api-key) without code changes. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot AI review requested due to automatic review settings June 11, 2026 19:15

Copilot started reviewing on behalf of anticomputer June 11, 2026 19:15 View session

p- previously approved these changes Jun 12, 2026

View reviewed changes

style: use ternary for token resolution (ruff SIM108)

ed19781

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot AI review requested due to automatic review settings June 12, 2026 15:12

anticomputer dismissed p-’s stale review via ed19781 June 12, 2026 15:12

Copilot started reviewing on behalf of anticomputer June 12, 2026 15:13 View session

Copilot AI reviewed Jun 12, 2026

View reviewed changes

Comment thread src/seclab_taskflow_agent/sdk/anthropic_sdk/backend.py Outdated

Comment thread README.md Outdated

Comment thread doc/GRAMMAR.md Outdated

anticomputer and others added 2 commits June 12, 2026 12:06

style: fix hatch fmt lint errors in test_mcp_utils

00978f2

- Remove quotes from _FakeTool return type (UP037) - Use raw string for regex pattern in pytest.raises match (RUF043) Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot AI review requested due to automatic review settings June 12, 2026 16:09

Copilot started reviewing on behalf of anticomputer June 12, 2026 16:10 View session

Copilot AI reviewed Jun 12, 2026

View reviewed changes

Comment thread src/seclab_taskflow_agent/capi.py

Comment thread src/seclab_taskflow_agent/mcp_utils.py

anticomputer and others added 2 commits June 12, 2026 12:43

Copilot AI review requested due to automatic review settings June 12, 2026 17:21

Copilot started reviewing on behalf of anticomputer June 12, 2026 17:21 View session

Copilot AI reviewed Jun 12, 2026

View reviewed changes

Comment thread src/seclab_taskflow_agent/sdk/anthropic_sdk/backend.py Outdated

Comment thread doc/GRAMMAR.md Outdated

p- previously approved these changes Jun 15, 2026

View reviewed changes

Copilot AI review requested due to automatic review settings June 15, 2026 14:52

anticomputer dismissed p-’s stale review via f42c06b June 15, 2026 14:52

Copilot started reviewing on behalf of anticomputer June 15, 2026 14:53 View session

Copilot AI reviewed Jun 15, 2026

View reviewed changes

Comment thread tests/test_sdk_anthropic_adapter.py Outdated

Comment thread tests/test_sdk_anthropic_adapter.py Outdated

Comment thread tests/test_sdk_anthropic_adapter.py Outdated

Comment thread src/seclab_taskflow_agent/sdk/anthropic_sdk/backend.py

Copilot AI review requested due to automatic review settings June 15, 2026 15:39

Copilot started reviewing on behalf of anticomputer June 15, 2026 15:39 View session

Copilot AI reviewed Jun 15, 2026

View reviewed changes

Conversation

anticomputer commented Jun 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Changes

New backend: anthropic_sdk

Per-model backend selection

Dependency changes

Testing

Files changed

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Copilot's findings

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot's findings

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot's findings

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot's findings

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot's findings

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot's findings

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot's findings

Uh oh!

Uh oh!

Uh oh!

anticomputer commented Jun 15, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot's findings

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot's findings

Uh oh!

anticomputer Jun 15, 2026

Choose a reason for hiding this comment

Uh oh!

anticomputer commented Jun 11, 2026 •

edited

Loading

New backend: `anthropic_sdk`