Repository navigation
fix(agent): harden tool-call path against malformed relay responses - #4569
Merged
Merged
Conversation
Combine malformed tool-call handling with placeholder filtering and a no-tools fallback so a relay that returns tool_use blocks with null id/name/input can no longer crash a turn or permanently wedge a session. Adapted to the ContextGovernor architecture (context governance now lives in nanobot/agent/context_governance.py, not runner.py): - ToolCallRequest.has_valid_name(): single source of truth for "usable name" (non-empty string). - tool_hints.format_tool_hints(): skip tool calls with a non-string/empty name instead of raising AttributeError on the whole turn. - ContextGovernor.strip_placeholder_assistant_messages() and strip_malformed_tool_calls() (plus the _tool_call_name_is_valid helper): history-cleaning staticmethods invoked at the START of prepare_for_model() — strip_placeholder, then strip_malformed, then the existing drop_orphan/backfill chain. Both only repair the model-facing copy and leave persisted history untouched (return a copy, or the same list when nothing changes). Also wired into runner's minimal-repair path. - AgentRunner._drop_malformed_tool_calls(): returns (dropped, all_dropped, original_finish_reason); clears finish_reason to "stop" when all calls are dropped. - AgentRunner._malformed_tool_call_retry_messages() + _request_model malformed_retry flag: when an all-dropped tool_calls response comes back, retry once with a corrective note; if the retry STILL comes back all-dropped, fall back to _request_no_tools for graceful text degradation. Tests for the history-cleaning methods live with ContextGovernor in tests/agent/test_runner_governance.py; response-layer and tool-hint tests stay on AgentRunner / tool_hints. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Re-bin
approved these changes
Jun 28, 2026
Re-bin
left a comment
Re-bin
left a comment
Collaborator
There was a problem hiding this comment.
LGTM. This is a focused defensive fix for malformed tool-call names: bad calls are filtered before execution/persistence, polluted model-facing history is cleaned, and all-malformed responses get one corrective retry before falling back without tools.
I pushed a small follow-up to fix ruff formatting and align the retry note / PR body with the intentionally narrow guard. Local targeted validation passed:
- python -m pytest tests/agent/test_runner_governance.py tests/agent/test_tool_hint.py -q
- python -m ruff check nanobot/agent/context_governance.py nanobot/agent/runner.py nanobot/providers/base.py nanobot/utils/tool_hints.py tests/agent/test_runner_governance.py tests/agent/test_tool_hint.py
m11y
added a commit
to m11y/nanobot
that referenced
this pull request
Jul 28, 2026
Temporary diagnostic to locate the intermittent upstream relay bug where a tool_use block comes back with a missing/non-string name (finish_reason still 'tool_calls'), forcing HKUDS#4569's drop->retry->no-tools fallback and surfacing degenerate replies like "call". On both the streaming and non-streaming Anthropic paths, when any returned tool_use has an empty name, dump the full request (system + messages + tool names), the full response (per-block raw dump, so we can see whether id/input survive), and the streamed content_block_start events (to tell whether the relay drops the name in the SSE stream itself or only in the assembled message) to ~/.nanobot/logs/malformed_toolcall_*.json. On by default; opt out with NANOBOT_NO_MALFORMED_CAPTURE=1. Never raises. To be replaced by the real recovery fix once the captured data pins down what survives. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
Some upstream relays can emit malformed tool-use responses where the tool name is missing or not a usable string. If such a call reaches execution or gets persisted into history, later turns can crash, fail provider validation, or keep replaying the same broken tool-call pattern. Compaction placeholder assistant messages can also nudge the model back toward failed tool-call demonstrations.
Changes
Hardens the tool-call path in two focused places:
ToolCallRequest.has_valid_name()and drops response tool calls whose name is missing or empty before execution/persistence.ContextGovernorby removing malformed persisted tool calls and placeholder assistant messages before request shaping. Persisted transcript data is left untouched.Provider parsers already synthesize missing tool IDs where needed, and argument shape validation remains in
ToolRegistry, so this PR keeps the guard intentionally narrow around the malformed-name failure mode.Testing
python -m pytest tests/agent/test_runner_governance.py tests/agent/test_tool_hint.py -qpython -m ruff check nanobot/agent/context_governance.py nanobot/agent/runner.py nanobot/providers/base.py nanobot/utils/tool_hints.py tests/agent/test_runner_governance.py tests/agent/test_tool_hint.py