fix(goals): 对话目标分析 JSON 解析失败分类与 json-repair 兜底（issue #259） - #260
Merged
Merged
Conversation
- guardrails validate_and_clean_json 解析前剥离思考标签；失败分类为 空响应/无 JSON 结构纯文本/JSON 损坏/字段校验失败；响应预览升为 WARNING 便于定位；json-repair（可选依赖）兜底恢复截断 JSON - 对话目标/意图分析：LLM 空响应与消毒后为空显式降级，不再流入 误导性的 JSON 解析报错 - 新增回归测试 test_guardrails_json_parsing.py
Contributor
Reviewer's Guide该 PR 改进对话目标分析的 LLM JSON 处理链路：先剥离推理内容，按响应失败类型提供可定位日志，并通过可选 json-repair 增强修复能力；目标管理器对空响应直接默认降级，同时补充回归测试并发布 4.2.6。 Sequence diagram for resilient goal analysis JSON parsingsequenceDiagram
participant LLM
participant GoalManager as ConversationGoalManager
participant Guardrails as GuardrailsManager
participant JsonRepair as json_repair
participant Logger
GoalManager->>LLM: analyze goal or intent
LLM-->>GoalManager: response
alt response is empty
GoalManager->>Logger: warning default fallback
else response has content
GoalManager->>Guardrails: validate_and_clean_json(response)
Guardrails->>Guardrails: remove_thinking_content(response)
alt valid JSON
Guardrails-->>GoalManager: parsed result
else malformed JSON
Guardrails->>Logger: warning category and preview
Guardrails->>JsonRepair: repair_json(text, return_objects=True)
alt repair succeeds
JsonRepair-->>Guardrails: repaired object
Guardrails-->>GoalManager: parsed result
else repair unavailable or fails
Guardrails-->>GoalManager: None
GoalManager->>Logger: default fallback
end
end
end
Flow diagram for categorized JSON parsing failuresflowchart TD
A[LLM response] --> B{Input empty?}
B -->|Yes| C[Log empty response]
C --> D[Return default fallback]
B -->|No| E[remove_thinking_content]
E --> F{Content remains?}
F -->|No| G[Log response empty after cleaning]
G --> D
F -->|Yes| H[Extract JSON structure]
H --> I{JSON parses and fields validate?}
I -->|Yes| J[Return validated result]
I -->|No JSON structure| K[Log pure-text response and preview]
I -->|JSON structure damaged| L[Log damaged JSON and preview]
K --> M[Attempt standard JSON repair]
L --> M
M --> N{Repair succeeds?}
N -->|Yes| J
N -->|No| O[_json_repair_fallback]
O --> P{json-repair succeeds?}
P -->|Yes| J
P -->|No| D
File-Level Changes
Assessment against linked issues
Possibly linked issues
Tips and commandsInteracting with Sourcery
Customizing Your ExperienceAccess your dashboard to:
Getting Help
|
![sourcery-ai[bot]](https://avatars.githubusercontent.com/in/48477?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Hey - I've found 1 issue
Prompt for AI Agents
Please address the comments from this code review:
## Individual Comments
### Comment 1
<location path="utils/guardrails_manager.py" line_range="620-623" />
<code_context>
+ logger.warning(f" [Guardrails] 常规 JSON 修复失败: {fix_error}")
+
+ # 最后兜底：json-repair（AstrBot 环境通常自带；未安装时静默跳过）
+ repaired = self._json_repair_fallback(cleaned_text)
+ if repaired is not None:
+ logger.info(f" [Guardrails] json-repair 兜底解析成功")
+ return repaired
+
+ logger.error(f" [Guardrails] JSON 解析与修复全部失败，返回 None 交由上层降级")
</code_context>
<issue_to_address>
**issue (bug_risk):** The json-repair fallback returns any non-empty repaired value without enforcing `expected_type` or requiring an object/array. With json-repair installed, a plain-text response can be returned as a string and logged as a successful parse, while a response expected to be an object can be returned as an array or scalar, bypassing the stated JSON-structure classification and potentially reaching callers with the wrong shape.
**Triggers:** When the initial JSON parse fails and the optional `json-repair` dependency is installed.
**Suggested fix:** After repair, validate the result against `expected_type` and reject scalar values or the wrong container type before returning it.
```suggestion
repaired = self._json_repair_fallback(cleaned_text)
if repaired is not None:
if expected_type == "object":
is_valid = isinstance(repaired, dict)
elif expected_type == "array":
is_valid = isinstance(repaired, list)
else:
is_valid = isinstance(repaired, (dict, list))
if not is_valid:
logger.warning(f" [Guardrails] json-repair 结果类型不符合预期: {type(repaired).__name__}")
return None
logger.info(f" [Guardrails] json-repair 兜底解析成功")
return repaired
```
</issue_to_address>Sourcery assessment
Approval pending. 1 finding to address first.
Blocking findings: utils/guardrails_manager.py:623
Comment on lines
+620
to
+623
| repaired = self._json_repair_fallback(cleaned_text) | ||
| if repaired is not None: | ||
| logger.info(f" [Guardrails] json-repair 兜底解析成功") | ||
| return repaired |
Contributor
There was a problem hiding this comment.
issue (bug_risk): The json-repair fallback returns any non-empty repaired value without enforcing expected_type or requiring an object/array. With json-repair installed, a plain-text response can be returned as a string and logged as a successful parse, while a response expected to be an object can be returned as an array or scalar, bypassing the stated JSON-structure classification and potentially reaching callers with the wrong shape.
Triggers: When the initial JSON parse fails and the optional json-repair dependency is installed.
Suggested fix: After repair, validate the result against expected_type and reject scalar values or the wrong container type before returning it.
Suggested change
| repaired = self._json_repair_fallback(cleaned_text) | |
| if repaired is not None: | |
| logger.info(f" [Guardrails] json-repair 兜底解析成功") | |
| return repaired | |
| repaired = self._json_repair_fallback(cleaned_text) | |
| if repaired is not None: | |
| if expected_type == "object": | |
| is_valid = isinstance(repaired, dict) | |
| elif expected_type == "array": | |
| is_valid = isinstance(repaired, list) | |
| else: | |
| is_valid = isinstance(repaired, (dict, list)) | |
| if not is_valid: | |
| logger.warning(f" [Guardrails] json-repair 结果类型不符合预期: {type(repaired).__name__}") | |
| return None | |
| logger.info(f" [Guardrails] json-repair 兜底解析成功") | |
| return repaired |
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
1 participant
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.
修复内容（issue #259）
对话目标分析中模型返回纯文本/思考内容时，
validate_and_clean_json连续报JSON 解析失败/JSON 修复失败: Expecting value: line 1 column 1 (char 0)，且日志无法定位原因。guardrails_manager
<think>等，推理型模型常见），不再干扰 JSON 提取json-repair（可选依赖，AstrBot 环境通常自带；未安装时静默跳过），可恢复截断 JSON 等常规正则修复覆盖不了的情况conversation_goal_manager
测试
tests/unit/test_guardrails_json_parsing.py（24 个用例）：空响应、纯文本分类、思考标签包裹、单引号/尾逗号修复、截断 JSON 兜底、字段校验失败、目标管理器空响应短路径版本
Closes #259
Summary by Sourcery
Improve conversation-goal analysis resilience and diagnostics for empty, explanatory, and malformed LLM responses.
Bug Fixes:
Enhancements:
Tests:
Chores: