validator-run

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

/validator-run

Execute the autonomous verification suite.

执行自主验证套件。

Procedure

流程

Step 1 - Run Agent Validator

步骤1 - 运行Agent Validator

If the caller requests a specific review to be enabled, append

--enable-review <name>

to the run command for each requested review.

Run

agent-validate run

using

Bash

with

timeout: 300000

. ALWAYS wait for and read the full command output before proceeding — the command typically takes 1-2 minutes. Verify you can see a
Status:
line in the output before continuing.

如果调用者要求启用特定评审，请为每个请求的评审在运行命令后追加

--enable-review <name>

。

使用

Bash

运行

agent-validate run

，设置

timeout: 300000

。在继续操作前务必等待并读取完整的命令输出——该命令通常需要1-2分钟。继续前请确认输出中包含
Status:
行。

Step 2 - Check Status

步骤2 - 检查状态

NEVER assume success — you must see an explicit

Status:

line before continuing. Check it and route accordingly:

```
Status: Passed
```
→ Go to Step 8.
```
Status: Passed with warnings
```
→ Go to Step 8.
```
Status: Failed
```
→ Continue to Step 3. You MUST continue — do not stop here.
```
Status: Retry limit exceeded
```
→ Go to Step 8.
No status line visible → Known issue: Bun can drop all stdout/stderr when LLM review subprocesses run. Read the console log file to get the status: find the latest
```
console.*.log
```
in the validator log directory (e.g.,
```
validator_logs/console.1.log
```
) and look for the
```
Status:
```
line there. If no console log is found there, also check
```
validator_logs/previous/
```
for logs from the most recent archived run. If no console log exists in either location, the command may have timed out or failed to run — re-run with a longer timeout or investigate the error. Do NOT proceed as if it passed.

绝不要假设验证成功——必须看到明确的

Status:

行才能继续。检查状态并按如下路径处理：

```
Status: Passed
```
→ 进入步骤8。
```
Status: Passed with warnings
```
→ 进入步骤8。
```
Status: Failed
```
→ 继续步骤3。必须继续操作，请勿在此停止。
```
Status: Retry limit exceeded
```
→ 进入步骤8。
未显示状态行 → 已知问题： 当LLM评审子进程运行时，Bun可能会丢失所有stdout/stderr输出。请读取控制台日志文件获取状态：在验证器日志目录中找到最新的
```
console.*.log
```
文件（例如
```
validator_logs/console.1.log
```
），并在其中查找
```
Status:
```
行。如果该目录下没有控制台日志，也请检查
```
validator_logs/previous/
```
中最近存档运行的日志。如果两个位置都没有控制台日志，说明命令可能超时或运行失败——请延长超时时间重新运行或排查错误。请勿当作验证通过继续操作。

Step 3 - Extract Failures

步骤3 - 提取失败信息

Required when status is Failed:

Infer the log directory from the file paths in the console output (e.g., if output references
```
validator_logs/check_._lint.1.log
```
, the log directory is
```
validator_logs/
```
)
Extract log failures using the first available strategy: a. Task tool (Claude Code):
```
Task
```
with
```
subagent_type="general-purpose"
```
,
```
model="haiku"
```
,
```
prompt=
```
the Extract Prompt (from the Appendix below) +
```
"\n\nLog directory: <inferred path>"
```
. Task calls MUST be synchronous — NEVER use
```
run_in_background: true
```
. b. Subagent delegation: If your environment supports delegating work to a subagent but not the Task tool, delegate the Extract Prompt instructions with the log directory to a subagent for processing. c. Inline fallback: If no subagent capability is available, follow the Extract Prompt instructions yourself to read the log files and produce the compact failure summary.

当状态为Failed时需执行此步骤：

从控制台输出中的文件路径推断日志目录（例如，如果输出引用
```
validator_logs/check_._lint.1.log
```
，则日志目录为
```
validator_logs/
```
）
使用以下优先策略提取日志失败信息： a. 任务工具（Claude Code）：调用
```
Task
```
，设置
```
subagent_type="general-purpose"
```
、
```
model="haiku"
```
，
```
prompt
```
为下方附录中的提取提示语 +
```
"\n\nLog directory: <推断路径>"
```
。任务调用必须同步——请勿设置
```
run_in_background: true
```
。 b. 子代理委托：如果您的环境支持将工作委托给子代理但不支持任务工具，请将提取提示语说明和日志目录委托给子代理处理。 c. 内联回退：如果没有子代理功能，请自行按照提取提示语说明读取日志文件并生成简洁的失败摘要。

Step 4 - Report Failures

步骤4 - 报告失败信息

Print the compact failure summary returned from Step 3.

打印步骤3返回的简洁失败摘要。

Step 5 - Fix

步骤5 - 修复问题

Fix issues reasonably supported by the feedback or likely intended by the human. When skipping an issue, briefly state what was skipped and why.

Valid reasons to skip:

Purely stylistic or subjective preference
The human would not want it changed

You MUST NOT skip for these reasons:

"Issue is pre-existing" — you MUST fix it unless you have another valid reason
"Issue is out of scope" — you MUST address valid feedback even if it requires refactoring or non-trivial changes

Apply this guidance to each failure and fix accordingly:

CHECK failures with Fix Skill: invoke the named skill
CHECK failures with Fix Instructions: follow the instructions
REVIEW violations: fix or skip per the guidance above

根据反馈合理修复问题，或按照人类的预期进行修复。跳过问题时，请简要说明跳过的内容及原因。

可跳过的合理原因：

纯风格或主观偏好问题
人类不希望修改该内容

以下原因禁止跳过问题：

"问题已存在"——除非有其他合理原因，否则必须修复
"问题超出范围"——即使需要重构或重大修改，也必须处理有效的反馈

针对每个失败问题，按以下指导修复：

带有修复技能的CHECK失败：调用指定的skill
带有修复说明的CHECK失败：遵循说明操作
REVIEW违规：根据上述指导修复或跳过

Step 6 - Update Review Decisions

步骤6 - 更新评审决策

For REVIEW violations you addressed:

Update review decisions using the first available strategy (same as Step 3): a. Task tool (Claude Code):
```
Task
```
with
```
subagent_type="general-purpose"
```
,
```
model="haiku"
```
,
```
prompt=
```
the Update Prompt (from the Appendix below) + log directory + decisions list. Task calls MUST be synchronous — NEVER use
```
run_in_background: true
```
. b. Subagent delegation: Delegate the Update Prompt instructions with the log directory and decisions to a subagent. c. Inline fallback: Follow the Update Prompt instructions yourself to update the review JSON files.

对于已处理的REVIEW违规：

使用以下优先策略更新评审决策（与步骤3相同）： a. 任务工具（Claude Code）：调用
```
Task
```
，设置
```
subagent_type="general-purpose"
```
、
```
model="haiku"
```
，
```
prompt
```
为下方附录中的更新提示语 + 日志目录 + 决策列表。任务调用必须同步——请勿设置
```
run_in_background: true
```
。 b. 子代理委托：将更新提示语说明、日志目录和决策委托给子代理。 c. 内联回退：自行按照更新提示语说明更新评审JSON文件。

Step 7 - Re-run Verification

步骤7 - 重新运行验证

NEVER skip this step — if the run failed, you MUST fix and re-run. Run the same command from Step 1 (including any

--enable-review

flags) again with

Bash

and

timeout: 300000

. The tool detects existing logs and automatically switches to verification mode. Go back to Step 2 to check the status line and repeat.

绝不要跳过此步骤——如果运行失败，必须修复后重新运行。再次使用

Bash

运行步骤1中的相同命令（包括所有

--enable-review

参数），设置

timeout: 300000

。工具会检测现有日志并自动切换到验证模式。返回步骤2检查状态行并重复流程。

Step 8 - Summarize Session

步骤8 - 总结会话

Provide a summary of the session:

Final Status: (Passed / Passed with warnings / Retry limit exceeded)
Issues Fixed: (list key fixes)
Issues Skipped: (list skipped items and reasons)
Outstanding Failures: (if retry limit exceeded, list unverified fixes and remaining issues)

提供会话总结：

最终状态：（Passed / Passed with warnings / Retry limit exceeded）
已修复问题：（列出关键修复内容）
已跳过问题：（列出跳过项及原因）
未解决失败：（如果超出重试次数，列出未验证的修复内容和剩余问题）

Appendix: Subagent Prompts

附录：子代理提示语

Extract Prompt

提取提示语

You are an EXTRACT subagent. Your job is to read validator log files and return a compact error summary.

您是一名EXTRACT子代理。您的工作是读取验证器日志文件并返回简洁的错误摘要。

Input

输入

You receive a log directory path as your only input.

您仅会收到一个日志目录路径作为输入。

Process

流程

List files directly under the log directory
Find the highest-numbered
```
console.N.log
```
file (e.g.,
```
console.3.log
```
>
```
console.2.log
```
)
Read it and find all lines containing
```
[FAIL]
```
For each
```
[FAIL]
```
line, extract the referenced file path
Read each referenced file:
- .log
  files (check gates): Extract error output. Look for
```
--- Fix Instructions ---
```
  sections and
```
--- Fix Skill: <name> ---
```
  sections. Include their full content.
- .json
  files (review gates): Parse the JSON. Find violations where
```
status
```
  is
```
"new"
```
  . For each, extract:
```
file
```
  ,
```
line
```
  ,
```
issue
```
  ,
```
priority
```
  ,
```
fix
```
  .

列出日志目录下的直接文件
找到编号最高的
```
console.N.log
```
文件（例如
```
console.3.log
```
>
```
console.2.log
```
）
读取该文件并找到所有包含
```
[FAIL]
```
的行
对于每个
```
[FAIL]
```
行，提取引用的文件路径
读取每个引用的文件：
- .log
  文件（检查关卡）：提取错误输出。查找
```
--- Fix Instructions ---
```
  部分和
```
--- Fix Skill: <name> ---
```
  部分。包含其完整内容。
- .json
  文件（评审关卡）：解析JSON。找到
```
status
```
  为
```
"new"
```
  的违规项。对于每个违规项，提取：
```
file
```
  、
```
line
```
  、
```
issue
```
  、
```
priority
```
  、
```
fix
```
  。

Output Format

输出格式

Return a plain-text summary using EXACTLY this format:

For check failures:

text

CHECKS:
[fail] <gate_label>
<concise error description>
Fix Instructions: <extracted text if present, otherwise omit this line>
Fix Skill: <skill name if present, otherwise omit this line>

For review failures:

text

REVIEWS:
[<priority>] <gate_label>
<file>:<line> - <issue summary>
Fix: <fix suggestion>

If there are no failures of a type, omit that section entirely.

使用以下精确格式返回纯文本摘要：

检查失败：

text

CHECKS:
[fail] <gate_label>
<简洁错误描述>
Fix Instructions: <提取的文本（如果存在，否则省略此行）>
Fix Skill: <技能名称（如果存在，否则省略此行）>

评审失败：

text

REVIEWS:
[<priority>] <gate_label>
<file>:<line> - <问题摘要>
Fix: <修复建议>

如果某类没有失败项，请完全省略该部分。

Example

示例

Example Input

示例输入

Log directory:

validator_logs/

The directory contains:

```
console.2.log
```
```
check_src_lint.2.log
```
```
review_src_code-quality_claude@1.2.json
```

console.2.log contains:

text

[START] check:src:lint
[FAIL]  check:src:lint (1.23s) - Exited with code 1
      Log: validator_logs/check_src_lint.2.log
[START] review:src:code-quality (claude@1)
[FAIL]  review:src:code-quality (claude@1) (5.42s) - Found 2 violations
      Review: validator_logs/review_src_code-quality_claude@1.2.json

check_src_lint.2.log contains:

text

[2026-02-15T10:23:45.123Z] Starting check: lint
Executing command: bun run lint
Working directory: /Users/user/project/src

src/helpers.ts:3:5 - error: Unexpected var, use let or const instead

Command failed: bun run lint
Result: fail - Exited with code 1

--- Fix Instructions ---
Replace all `var` declarations with `const` or `let`.

review_src_code-quality_claude@1.2.json contains:

json

{
  "adapter": "claude",
  "status": "fail",
  "violations": [
    {
      "file": "src/main.ts",
      "line": 45,
      "issue": "Missing error handling for async database call",
      "fix": "Wrap in try-catch block",
      "priority": "high",
      "status": "new"
    },
    {
      "file": "src/utils.ts",
      "line": 10,
      "issue": "Function exceeds 50 lines",
      "fix": "Extract helper methods",
      "priority": "medium",
      "status": "fixed"
    }
  ]
}

日志目录:

validator_logs/

该目录包含：

```
console.2.log
```
```
check_src_lint.2.log
```
```
review_src_code-quality_claude@1.2.json
```

console.2.log 包含：

text

[START] check:src:lint
[FAIL]  check:src:lint (1.23s) - Exited with code 1
      Log: validator_logs/check_src_lint.2.log
[START] review:src:code-quality (claude@1)
[FAIL]  review:src:code-quality (claude@1) (5.42s) - Found 2 violations
      Review: validator_logs/review_src_code-quality_claude@1.2.json

check_src_lint.2.log 包含：

text

[2026-02-15T10:23:45.123Z] Starting check: lint
Executing command: bun run lint
Working directory: /Users/user/project/src

src/helpers.ts:3:5 - error: Unexpected var, use let or const instead

Command failed: bun run lint
Result: fail - Exited with code 1

--- Fix Instructions ---
Replace all `var` declarations with `const` or `let`.

review_src_code-quality_claude@1.2.json 包含：

json

{
  "adapter": "claude",
  "status": "fail",
  "violations": [
    {
      "file": "src/main.ts",
      "line": 45,
      "issue": "Missing error handling for async database call",
      "fix": "Wrap in try-catch block",
      "priority": "high",
      "status": "new"
    },
    {
      "file": "src/utils.ts",
      "line": 10,
      "issue": "Function exceeds 50 lines",
      "fix": "Extract helper methods",
      "priority": "medium",
      "status": "fixed"
    }
  ]
}

Example Output

示例输出

text

CHECKS:
[fail] check:src:lint
src/helpers.ts:3:5 - error: Unexpected var, use let or const instead
Fix Instructions: Replace all `var` declarations with `const` or `let`.

REVIEWS:
[high] review:src:code-quality (claude@1)
src/main.ts:45 - Missing error handling for async database call
Fix: Wrap in try-catch block

Note: The

src/utils.ts:10

violation was omitted because its status is

"fixed"

, not

"new"

text

CHECKS:
[fail] check:src:lint
src/helpers.ts:3:5 - error: Unexpected var, use let or const instead
Fix Instructions: Replace all `var` declarations with `const` or `let`.

REVIEWS:
[high] review:src:code-quality (claude@1)
src/main.ts:45 - Missing error handling for async database call
Fix: Wrap in try-catch block

注意：

src/utils.ts:10

的违规项被省略，因为其状态为

"fixed"

，而非

"new"

。

Rules

规则

Do NOT summarize or editorialize — copy error details verbatim where possible
Do NOT skip any
```
[FAIL]
```
entries
Keep the output compact — one entry per check failure, one entry per review violation (3 lines each)
For review violations, only include those with
```
status: "new"
```
— skip
```
"fixed"
```
and
```
"skipped"
```

请勿总结或编辑——尽可能直接复制错误细节
请勿跳过任何
```
[FAIL]
```
条目
保持输出简洁——每个检查失败项占一条目，每个评审违规项占一条目（各占3行）
对于评审违规项，仅包含
```
status: "new"
```
的项——跳过
```
"fixed"
```
和
```
"skipped"
```
的项

Update Prompt

更新提示语

You are an UPDATE subagent. Your job is to update review JSON files with fix/skip decisions.

您是一名UPDATE子代理。您的工作是用修复/跳过决策更新评审JSON文件。

Input

输入

You receive:

A log directory path
A list of decisions, each with:
```
file
```
,
```
line
```
,
```
issue_prefix
```
,
```
status
```
("fixed" or "skipped"), and
```
result
```
(brief description)

您会收到：

日志目录路径
决策列表，每个决策包含：
```
file
```
、
```
line
```
、
```
issue_prefix
```
、
```
status
```
（"fixed" 或 "skipped"）和
```
result
```
（简要描述）

Process

流程

For each decision:

Find the matching
```
.json
```
file in the log directory by scanning for a violation that matches on
```
file
```
(exact) AND
```
line
```
(exact) AND where
```
issue
```
starts with the provided
```
issue_prefix
```
. If multiple violations match, use the first unprocessed one (status still
```
"new"
```
)
Read the JSON file
Find the matching violation in the
```
violations
```
array
Set
```
"status"
```
to the provided status value
Set
```
"result"
```
to the provided result string
Write the updated JSON back to the same file path

对于每个决策：

在日志目录中查找匹配的
```
.json
```
文件，方法是扫描与
```
file
```
（完全匹配）、
```
line
```
（完全匹配）且
```
issue
```
以提供的
```
issue_prefix
```
开头的违规项。如果有多个匹配项，使用第一个未处理的项（状态仍为
```
"new"
```
）
读取JSON文件
在
```
violations
```
数组中找到匹配的违规项
将
```
"status"
```
设置为提供的状态值
将
```
"result"
```
设置为提供的结果字符串
将更新后的JSON写回原文件路径

Rules

规则

Do NOT modify any fields other than
```
status
```
and
```
result
```
Do NOT modify violations that don't match the provided decisions
Preserve all other JSON structure and key ordering
If a violation cannot be found, report it in your response but continue with other decisions
Write the JSON with 2-space indentation

请勿修改
```
status
```
和
```
result
```
以外的任何字段
请勿修改与提供的决策不匹配的违规项
保留所有其他JSON结构和键的顺序
如果找不到违规项，请在响应中报告，但继续处理其他决策
写入JSON时使用2空格缩进

Example

示例

Example Input

示例输入

Log directory:

validator_logs/

Decisions:

file:

src/main.ts

, line: 45, issue_prefix:

Missing error handling

, status:

fixed

, result:

Added try-catch around database call

file:

src/utils.ts

, line: 10, issue_prefix:

Function exceeds

, status:

skipped

, result:

Stylistic preference, function is readable as-is

The log directory contains

review_src_code-quality_claude@1.2.json

json

{
  "adapter": "claude",
  "status": "fail",
  "violations": [
    {
      "file": "src/main.ts",
      "line": 45,
      "issue": "Missing error handling for async database call",
      "fix": "Wrap in try-catch block",
      "priority": "high",
      "status": "new"
    },
    {
      "file": "src/utils.ts",
      "line": 10,
      "issue": "Function exceeds 50 lines",
      "fix": "Extract helper methods",
      "priority": "medium",
      "status": "new"
    }
  ]
}

日志目录:

validator_logs/

决策：

file:

src/main.ts

, line: 45, issue_prefix:

Missing error handling

, status:

fixed

, result:

Added try-catch around database call

file:

src/utils.ts

, line: 10, issue_prefix:

Function exceeds

, status:

skipped

, result:

Stylistic preference, function is readable as-is

日志目录包含

review_src_code-quality_claude@1.2.json

：

json

{
  "adapter": "claude",
  "status": "fail",
  "violations": [
    {
      "file": "src/main.ts",
      "line": 45,
      "issue": "Missing error handling for async database call",
      "fix": "Wrap in try-catch block",
      "priority": "high",
      "status": "new"
    },
    {
      "file": "src/utils.ts",
      "line": 10,
      "issue": "Function exceeds 50 lines",
      "fix": "Extract helper methods",
      "priority": "medium",
      "status": "new"
    }
  ]
}

Example Output (what you write to the JSON file)

示例输出（写入JSON文件的内容）

After updating,

review_src_code-quality_claude@1.2.json

becomes:

json

{
  "adapter": "claude",
  "status": "fail",
  "violations": [
    {
      "file": "src/main.ts",
      "line": 45,
      "issue": "Missing error handling for async database call",
      "fix": "Wrap in try-catch block",
      "priority": "high",
      "status": "fixed",
      "result": "Added try-catch around database call"
    },
    {
      "file": "src/utils.ts",
      "line": 10,
      "issue": "Function exceeds 50 lines",
      "fix": "Extract helper methods",
      "priority": "medium",
      "status": "skipped",
      "result": "Stylistic preference, function is readable as-is"
    }
  ]
}

更新后，

review_src_code-quality_claude@1.2.json

变为：

json

{
  "adapter": "claude",
  "status": "fail",
  "violations": [
    {
      "file": "src/main.ts",
      "line": 45,
      "issue": "Missing error handling for async database call",
      "fix": "Wrap in try-catch block",
      "priority": "high",
      "status": "fixed",
      "result": "Added try-catch around database call"
    },
    {
      "file": "src/utils.ts",
      "line": 10,
      "issue": "Function exceeds 50 lines",
      "fix": "Extract helper methods",
      "priority": "medium",
      "status": "skipped",
      "result": "Stylistic preference, function is readable as-is"
    }
  ]
}

Example Response

示例响应

text

Updated 2 violations:
- src/main.ts:45 — set to fixed
- src/utils.ts:10 — set to skipped

text

Updated 2 violations:
- src/main.ts:45 — set to fixed
- src/utils.ts:10 — set to skipped

Output

输出

Return a brief confirmation listing each decision applied:

text

Updated <N> violations:
- <file>:<line> — set to <status>

If any decisions could not be matched, add:

text

Unmatched decisions:
- <file>:<line> — <issue_prefix> (not found in any JSON file)

返回简要确认信息，列出每个已应用的决策：

text

Updated <N> violations:
- <file>:<line> — set to <status>

如果有任何决策无法匹配，请添加：

text

Unmatched decisions:
- <file>:<line> — <issue_prefix> (not found in any JSON file)