alanhuangyoo f5cf25a8b6

fix(summarization): keep a Human Input Card reply as the current request (#5416) ...

Summarization rescues the latest user message by id so the current
request survives compaction (#4882). It picked that message with
is_real_user_message, which rejects every hide_from_ui HumanMessage,
including Human Input Card replies. When a run starts from a card
reply, the older visible request was kept and the user's answer was
summarized away.

Use is_genuine_user_message, which accepts hidden messages that carry a
valid human_input_response and still skips other hidden injections.

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

2026-09-14 10:45:31 +08:00

..

__init__.py

feat: add create_deerflow_agent SDK entry point (Phase 1) (#1203)

2026-03-29 15:31:18 +08:00

_bounded_dict.py

fix(subagents): unify guardrail caps on additive stop_reason + add token_budget (#3875 Phase 2) (#3980)

2026-07-08 22:26:06 +08:00

AGENTS.md

fix(agents): keep the token budget across goal continuations of a run (#5410)

2026-09-14 10:41:22 +08:00

audit_context.py

feat(observability): persist deferred tool promotions (#5183)

2026-09-05 14:03:00 +08:00

clarification_middleware.py

fix(clarification): drop sibling tool calls before interrupt (#4908)

2026-08-24 07:37:35 +08:00

configured_extensions.py

feat(extensions): allow constructor kwargs on config-declared middlewares (#5312)

2026-09-12 10:57:40 +08:00

dangling_tool_call_middleware.py

fix(middleware): skip raw tool-call fallback when invalid view carries the same call (#4693)

2026-08-08 21:07:55 +08:00

deferred_tool_filter_middleware.py

feat(extensions): let an out-of-tree extension observe what the agent did (#4863)

2026-08-23 09:57:12 +08:00

delegation_ledger.py

fix(subagents): preserve actionable acceptance gaps after compaction (#5287)

2026-09-08 16:11:49 +08:00

durable_context_middleware.py

feat: add opt-in task notes and compacted history recall (#5382)

2026-09-12 21:01:46 +08:00

dynamic_context_middleware.py

fix(memory): enforce backend read failure policy (#4726)

2026-09-06 09:01:33 +08:00

input_sanitization_middleware.py

fix(gateway): close two input-sanitization bypasses (#5375)

2026-09-14 09:54:42 +08:00

llm_error_handling_middleware.py

fix(llm): release owned recovery probe on cancellation (#5197)

2026-09-06 22:37:35 +08:00

loop_detection_middleware.py

fix(agents): isolate loop detection state by run (#5344)

2026-09-11 13:45:52 +08:00

mcp_routing_middleware.py

feat(observability): persist deferred tool promotions (#5183)

2026-09-05 14:03:00 +08:00

memory_middleware.py

feature(gateway): issue request trace ids unconditionally (#5119)

2026-09-01 16:49:39 +08:00

message_utils.py

fix(gateway): close two input-sanitization bypasses (#5375)

2026-09-14 09:54:42 +08:00

model_length_finish_reason_middleware.py

fix: surface length-capped model responses (#4309)

2026-07-26 14:43:08 +08:00

model_length_termination_detectors.py

fix: surface length-capped model responses (#4309)

2026-07-26 14:43:08 +08:00

read_before_write_middleware.py

fix(agents): make read-before-write cancellation lock-safe (#5395)

2026-09-13 19:55:58 +08:00

receipt_verification.py

feat(harness): subagent receipt citation verification (#5076)

2026-08-29 16:56:56 +08:00

safety_finish_reason_middleware.py

feat(extensions): let an out-of-tree extension observe what the agent did (#4863)

2026-08-23 09:57:12 +08:00

safety_termination_detectors.py

fix(runtime): suppress tool execution when provider safety-terminates with tool_calls (#3035)

2026-05-22 21:20:28 +08:00

sandbox_audit_middleware.py

fix(sandbox): judge command substitution by position in audit middleware (#4623)

2026-08-01 21:44:34 +08:00

skill_activation_middleware.py

feat(extensions): let an out-of-tree extension observe what the agent did (#4863)

2026-08-23 09:57:12 +08:00

skill_context.py

feat: emit structured runtime metadata (follow-up#3887) (#3906)

2026-07-04 11:27:19 +08:00

skill_tool_policy_middleware.py

fix(subagents): isolate callbacks and activate skills lazily (#4497)

2026-07-28 23:29:14 +08:00

subagent_limit_middleware.py

feat(subagents): add unified capacity and durable batch execution (#4998)

2026-08-25 07:49:38 +08:00

summarization_middleware.py

fix(summarization): keep a Human Input Card reply as the current request (#5416)

2026-09-14 10:45:31 +08:00

system_message_coalescing_middleware.py

feat(extensions): let an out-of-tree extension observe what the agent did (#4863)

2026-08-23 09:57:12 +08:00

terminal_response_middleware.py

feat(extensions): let an out-of-tree extension observe what the agent did (#4863)

2026-08-23 09:57:12 +08:00

thread_data_middleware.py

fix(runtime): honor LangGraph Server identity for user-scoped data (#4538)

2026-07-28 22:59:14 +08:00

title_middleware.py

feat(title): use filenames for attachment-only conversations (#5304)

2026-09-12 09:22:27 +08:00

todo_middleware.py

feat(extensions): let an out-of-tree extension observe what the agent did (#4863)

2026-08-23 09:57:12 +08:00

token_budget_middleware.py

fix(agents): keep the token budget across goal continuations of a run (#5410)

2026-09-14 10:41:22 +08:00

token_usage_middleware.py

fix(subagents): isolate background tasks from reused tool call IDs (#4758)

2026-08-12 09:25:05 +08:00

tool_call_args.py

feat(agents): elide superseded write_file payloads from model-bound requests (#5374)

2026-09-13 18:04:51 +08:00

tool_call_metadata.py

fix(middleware): sync raw tool call metadata (#2757)

2026-05-08 10:08:53 +08:00

tool_error_handling_middleware.py

feat: add opt-in task notes and compacted history recall (#5382)

2026-09-12 21:01:46 +08:00

tool_output_budget_middleware.py

feat(agents): elide superseded write_file payloads from model-bound requests (#5374)

2026-09-13 18:04:51 +08:00

tool_output_synopsis.py

feat(tool):Add structured synopsis for oversized tool output previews (#3377)

2026-07-16 16:41:04 +08:00

tool_progress_middleware.py

fix(agents): isolate loop detection state by run (#5344)

2026-09-11 13:45:52 +08:00

tool_promotion_audit_middleware.py

feat(observability): persist deferred tool promotions (#5183)

2026-09-05 14:03:00 +08:00

tool_receipt_middleware.py

feat(harness): subagent receipt citation verification (#5076)

2026-08-29 16:56:56 +08:00

tool_receipt.py

feat(harness): subagent receipt citation verification (#5076)

2026-08-29 16:56:56 +08:00

tool_result_meta.py

fix(agents): normalize Command-wrapped tool results (#4977)

2026-08-30 15:37:38 +08:00

tool_result_sanitization_middleware.py

fix(security): sanitize MCP-sourced tool results through the same trust boundary (#4839)

2026-09-01 09:49:32 +08:00

tool_transform_meta.py

feat(extensions): let an out-of-tree extension observe what the agent did (#4863)

2026-08-23 09:57:12 +08:00

uploads_middleware.py

fix(runtime): honor LangGraph Server identity for user-scoped data (#4538)

2026-07-28 22:59:14 +08:00

view_image_middleware.py

fix(view-image): read active sandbox images from sandbox (#5306)

2026-09-10 15:48:55 +08:00