Preserve contributor head and current main reasoning/errors. Reject malformed capability declarations and retry discovery; actual localhost HTTP ReAct round-trip verified.
json.loads raises RecursionError on deeply nested input, which escaped the
complete-JSON check and failed the request. Treat it like any other input
that is not a complete JSON document.
Signed-off-by: FenjuFu <fufenjupku@gmail.com>
With prefilled reasoning enabled, a structured answer such as
{"closing":"</think>"} was cut at the tag inside its string value.
Reasoning followed by a closing tag is never a valid JSON document, so a
reply that parses as one complete object or array is now returned
unchanged. Also annotate the tests and helpers this PR adds.
Signed-off-by: FenjuFu <fufenjupku@gmail.com>
Only an answered /api/show lookup is cached now, and the cache is replaced rather than mutated. A failed lookup sends no tools for that request and is retried on the next one, so a transient outage no longer disables tools for the provider's lifetime. Also annotate the test helpers and tests this PR adds.
Signed-off-by: FenjuFu <fufenjupku@gmail.com>
Tools are now forwarded only when the selected model declares tool support:
catalogued models answer from ModelInfo.supports_tools (llama3.2 and mistral
are tool models on ollama.com, codellama is not), and any other model is
looked up once through /api/show, whose capabilities list is what the
installed model declares. A failed lookup sends no tools.
Native Ollama tool calls often have no id, so each one gets a unique id, and
tool results are sent with tool_name resolved from the call they answer, so
several results stay attributable.
Signed-off-by: FenjuFu <92919259+FenjuFu@users.noreply.github.com>
Strip a balanced <think>...</think> block that opens the reply by default and
leave every other reply untouched, so answers that mention the tags (a lone
or reversed closing tag, JSON string values) are never truncated.
The close-only form produced by chat templates that prefill <think> is now
handled only when the OpenAI or Ollama provider is configured for it, via
prefilled_reasoning=True or OPENAI_PREFILLED_REASONING /
OLLAMA_PREFILLED_REASONING. Hosted providers keep the default.
Signed-off-by: FenjuFu <92919259+FenjuFu@users.noreply.github.com>
OllamaProvider built the /api/chat payload without the tools from
LLMInput, so Ollama models were never told any tools existed and a
ReActAgent backed by Ollama never ran one. The response side already
parsed message.tool_calls.
Send the tools in the OpenAI function format that /api/chat accepts,
matching the other providers.
Fixes#3274
Signed-off-by: FenjuFu <fufenjupku@gmail.com>
Self-hosted reasoning models served through vLLM or llama.cpp without a
reasoning parser, or through Ollama without a thinking template, return
their chain of thought inside message content. The OpenAI-compatible
providers and the Ollama provider passed it through verbatim, so
LLMOutput.content and ReActAgent final answers began with the reasoning.
Models whose chat template prefills <think> in the generation prompt
(Spark-X2.5, Qwen3 thinking variants) only emit the closing tag, so the
content arrives as "...</think>answer" with no opening tag.
Add strip_reasoning() and apply it in the OpenAI, Astraflow, Atlas and
Ollama providers. A <think> that does not open the output is kept as a
literal, and unterminated reasoning is left unchanged.
Signed-off-by: FenjuFu <fufenjupku@gmail.com>
- claw askClaude: prefer native .exe, route only .cmd/.bat through cmd.exe
with a quoted command line (DEP0190-safe, same pattern as
mcp-health-check); never exec .ps1 directly (not executable without
powershell); model allowlist retained so the shell line carries no
attacker metacharacters
- executor sync path: close un-awaited coroutine before returning the
generic failure, eliminating RuntimeWarning noise
* ci: add ruff + mypy to the Python CI job and fix pyproject tool config
The python-tests job runs pytest but not lint/type checks, and the ruff
and mypy configuration in pyproject.toml was silently broken, so neither
tool could run at all.
- add ruff and mypy steps to the existing python-tests job
- fix invalid pyproject keys: [tool.ruff] src-path -> src,
[tool.mypy] src_paths -> mypy_path
- ignore ruff UP042 (the (str, Enum) mixin is intentional)
- resolve ruff findings (unused/unsorted imports) across src and tests
- fix mypy errors in tools/executor.py and prompt/builder.py
* fix(ci): satisfy Python lint after main refresh
---------
Co-authored-by: haelyra <49814733+haelyra@users.noreply.github.com>
* fix(llm/providers/claude): attach cache_control to system block, not top-level
The Anthropic Messages API does not accept `cache_control` as a top-level
request parameter — it is a per-content-block field. Passing it at the top
level raises `TypeError` in the Python SDK (which validates kwargs against
`messages.create()`'s signature) or a `400 unknown_parameter` from the API,
so every ClaudeProvider.generate() call fails.
Move `cache_control: {"type": "ephemeral"}` onto the last system-prompt
block so ephemeral prompt caching still works when a system prompt is
present, and drop it when there isn't one (nothing to cache).
Existing tests didn't catch this because `FakeMessages.create(**_params)`
accepted anything and ignored the kwargs. FakeMessages now records
`last_params`, and two regression tests assert that:
- `cache_control` never appears as a top-level param, and
- when a system prompt is set, `cache_control` rides on the last block.
Fixes#2512
* test(claude_provider): split composite isinstance+truthiness assertion (PT018)
Ruff PT018 flagged the combined `isinstance(system, list) and system` check
in `test_generate_does_not_pass_cache_control_as_top_level_param`. Split
it into two focused asserts (`isinstance(system, list)` then `assert system`)
so a failure points at the exact violation instead of a compound condition.
Behavior unchanged; 6/6 tests still pass.
Addresses CodeRabbit review on #2515.
* fix: add .gitattributes to force LF line endings for text files
npm run command-registry:check (part of npm test) fails on a fresh clone
on Windows with the common core.autocrlf=true setting: git checks out
docs/COMMAND-REGISTRY.json with CRLF, but generate-command-registry.js
always writes LF, so the strict string comparison in checkRegistry()
never matches. Forcing LF via .gitattributes makes checkouts consistent
across platforms regardless of a contributor's local autocrlf setting.
* fix: normalize CRLF line endings to LF per .gitattributes
pyproject.toml, src/llm/__init__.py, src/llm/prompt/builder.py,
src/llm/providers/claude.py, and tests/test_builder.py had CRLF line
endings committed to the repo, inconsistent with the rest of the
codebase. Renormalized via 'git add --renormalize .' now that
.gitattributes enforces eol=lf.
---------
Co-authored-by: Affaan Mustafa <me@affaanmustafa.com>
Replace invalid default model IDs (e.g. claude-sonnet-4-7) with current
claude-sonnet-4-6, claude-opus-4-8, and claude-haiku-4-5. Route system
messages to the API system field, enable ephemeral prompt caching, omit
temperature for Opus 4.7/4.8, and surface cache usage metrics. Update the
CLI model picker to match.
Co-authored-by: Vladimir Đuranović <vlada@MacBook-Pro.local>
Co-authored-by: Cursor <cursoragent@cursor.com>
The OpenAI-compatible API can return HTTP 200 with an empty choices list
or choices[0].message = None (content-filtered responses on Gemini,
overwhelmed Ollama instances). Without a guard, both sites raise an
unhandled IndexError or AttributeError crashing the provider.
Added guard in OpenAIProvider.generate() and AstraFlowProvider.generate().
- add Vite and Redis pattern skills from closed stale PRs
- add frontend-slides support assets
- port skill-comply runner fixes and LLM prompt/provider regressions
- harden agent frontmatter validation and sync catalog counts
haelyra