Token导航 LogoToken导航TokenDH.com
研究检索执行命令github未标认证来源可访问许可证需确认审计通过

fixbugfixbug 搜索

Agent Skill

fixbug 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

451

周安装

19

GitHub Stars

2

下载量

158
CodexClaudeCursorGemini CLI

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

unknown

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:fixbug(fixbug 搜索)
来源仓库:https://github.com/tercel/code-forge
仓库路径:skills/fixbug
安装命令:
npx skills add https://github.com/tercel/code-forge --skill fixbug
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/tercel/code-forge --skill fixbug

简介

fixbug 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中快速定位候选结果。

  • 适用于需要根据关键词或任务场景从来源线索中获取信息的场景。
  • 通过 npx skills add 命令安装,需结合原始 README 核验具体用法。
  • 安装前建议确认权限范围、维护状态及是否触发联网或文件读写操作。
  • fixbug 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

SKILL.md

Code Forge — Fixbug

Systematically debug and fix bugs with interactive trace-back to upstream documents (task descriptions, plans, requirements).

When to Use

  • Encountered a bug or unexpected behavior in a feature
  • Need to diagnose whether the root cause is in code, task description, plan, or requirements
  • Want to fix the bug with TDD and keep upstream documents in sync

Examples

/code-forge:fixbug login page returns 500 error    # Diagnose and fix from description
/code-forge:fixbug @issues/bug-123.md              # Fix from bug report file
/code-forge:fixbug --review user-auth              # Batch-fix all issues from review report
/code-forge:fixbug --review                        # Auto-detect review report and fix

Workflow

Bug Input → Context Scan → Feature Association → Root Cause Diagnosis → Trace-back Confirmation → TDD Fix → Doc Sync → Summary

Detailed Steps

@../shared/configuration.md


Step 1: Receive Bug Description

Accept input in three modes:

  • Prompt text (e.g., /code-forge:fixbug login page returns 500 error) — use the text as bug description
  • File reference (e.g., /code-forge:fixbug @issues/bug-123.md) — read the file for bug details
  • Review mode (/code-forge:fixbug --review [feature-name]) — read the review report and batch-fix all issues. See Step 1R below.

1.0 Path-Like Input Guard

Before treating input as prompt text, check if it looks like a file/directory path without @. If the input does NOT start with @ and is NOT --review, but matches ANY of these patterns, it is likely a path the user forgot to prefix with @:

  • Contains / (e.g., issues/bug-123.md, ../bugs/report.md)
  • Starts with . (e.g., ./bug-report.md)
  • Ends with .md (e.g., bug-123.md)
  • Matches an existing file or directory on disk

Action: Use AskUserQuestion:

Your input looks like a file path: "{input}"
Did you mean to use file mode? (file paths require an @ prefix)
  • Options:

- "Yes, use as file path" → prepend @ and treat as file reference - "No, treat as bug description" → continue as prompt text


If no input provided, use AskUserQuestion to ask: "Describe the bug you encountered."

For prompt text and file reference modes, store the bug description and continue to Step 2.

For --review mode, go to Step 1R.


Step 1R: Review Mode Setup

Locate and parse the review report. The review report lives in the current conversation context (displayed by a prior /code-forge:review run). A saved file on disk is the fallback.

Source priority:

  1. Conversation context (primary): Look for the most recent review report output in the current conversation. The report starts with # Code Review: or # Project Review: and contains the structured dimension sections. If found, use it directly.
  2. Saved file (fallback): Only if no review report exists in the current conversation:

- With feature name (--review my-feature): read {output_dir}/{feature_name}/review.md - Without feature name (--review): search {output_dir}/project-review.md first, then scan {output_dir}/*/review.md - If multiple found: use AskUserQuestion to let user select

  1. Neither found: Show error — "No review report found in this session. Run /code-forge:review first, or /code-forge:review --save to persist to disk."
  2. Parse issues from the report:

- Extract all issues with severity blocker, critical, and warning - Skip suggestion-level issues (these are optional improvements, not bugs) - Group issues by file for efficient fixing - Sort by severity: blockers first, then criticals, then warnings

  1. Present issue summary and confirm: Display: Review Report: {feature_name or "project"} Found {N} issues to fix: Blockers: {count} Criticals: {count} Warnings: {count} Issues: 1. [{severity}] {file}:{line} — {title} 2. [{severity}] {file}:{line} — {title}... Use AskUserQuestion: "Fix all {N} issues? Or enter issue numbers to fix selectively (e.g., 1,3,5)."

- "all" / "yes" → process all issues - Comma-separated numbers → process only selected issues - "cancel" → abort

  1. Batch execution: For each selected issue (or group of issues in the same file):

- Set the issue's title + description + suggestion as the bug description - The review already provides file, line, description, and suggestion — use these directly as the diagnosis. Only spawn a diagnostic sub-agent (Step 4) if the review's suggestion is vague or the fix is non-obvious. - Execute Step 6 (TDD Fix) directly: - 6.1 Write a regression test targeting the specific issue - 6.2 Implement the fix (use the review's suggestion as guidance) - 6.3 Run full test suite - 6.4 Commit with message: fix: {issue title} ({file}:{line}) - Skip Steps 2 (Context Scan), 3 (Feature Association), 5 (Upstream Trace-back), and 7 (Doc Sync) — review issues are code-level (Level 1), and file context is already known from the report. - After fixing each issue, display a brief progress line: ✅ [{i}/{N}] Fixed: {title} ({file}:{line})

  1. Update state.json (if associated with a feature): Add a single aggregated fix record to the fixes array: {"bug": "review fixes — {N} issues", "root_cause_level": 1, "root_cause": "Code-level issues identified by review", "fixed_files": ["path/to/file1.ext", "path/to/file2.ext"], "commits": ["abc1234", "def5678"], "doc_updates": [], "fixed_at": "ISO timestamp"}
  2. Display batch summary: Review Fixes Complete: {feature_name or "project"} Fixed: {fixed_count}/{total_count} issues Blockers: {fixed_blockers}/{total_blockers} Criticals: {fixed_criticals}/{total_criticals} Warnings: {fixed_warnings}/{total_warnings} {If any failed:} ⚠ Could not fix: - {title} ({file}:{line}) — {reason} Commits: {commit hashes} Next steps: /code-forge:review {feature_name} Re-run review to verify fixes /code-forge:status {feature_name} View updated progress

Review mode ends here — do NOT continue to Step 2.


Step 2: Project Context Scan

Scan the project codebase to understand the context:

  1. Use Glob to get project structure overview
  2. Use Grep to search for keywords from the bug description in the codebase
  3. Identify files and modules likely related to the bug
  4. Note the tech stack and testing framework used

Store a concise context summary (~500 words).


Step 3: Associate Feature (If Exists)

Attempt to associate the bug with an existing code-forge feature:

  1. Search both {output_dir}/*/state.json and .code-forge/tmp/*/state.json for all features
  2. For each feature, check if the bug-related files overlap with the feature's task files (read tasks/*.md → "Files Involved" sections)
  3. Match found → load the feature's plan.md and relevant tasks/*.md as additional context. Note the feature name.
  4. No match found → mark as standalone bug. Skip upstream trace-back (Steps 5 and 7). Proceed with code-only fix.

If multiple features match, use AskUserQuestion to let user select the most relevant one.


Step 4: Root Cause Diagnosis

Offload to sub-agent for deep analysis.

Spawn an Agent tool call with:

  • subagent_type: "general-purpose"
  • description: "Diagnose bug root cause"

Sub-agent prompt must include:

  • Bug description
  • Project context summary from Step 2
  • Feature context (plan.md, task files) if associated in Step 3
  • Instruction to: reproduce the bug, identify the root cause, classify the root cause level

Root cause levels:

LevelDescriptionExample
1Code bugLogic error, boundary miss, typo, wrong variable
2Incomplete task descriptionTask.md steps missing a case, wrong acceptance criteria
3Plan design flawArchitecture doesn't handle a scenario, missing component
4Incomplete requirement docOriginal feature doc missing a requirement

Sub-agent must return:

ROOT_CAUSE_LEVEL: <1-4>
ROOT_CAUSE_SUMMARY: <1-2 sentence description>
AFFECTED_FILES:
- path/to/file.ext: <what's wrong>
UPSTREAM_DOCS_AFFECTED: (only if Level >= 2)
- Level 2: tasks/<task>.md — <what's missing/wrong>
- Level 3: plan.md — <what's missing/wrong>
- Level 4: {input_dir}/<feature>.md — <what's missing/wrong>
PROPOSED_FIX: <brief fix description>
REGRESSION_TEST: <what test to write>

Step 5: Interactive Trace-back Confirmation

This step only runs if Level >= 2 AND the bug is associated with a feature.

Present the root cause analysis to the user, then confirm upstream updates level by level.

5.1 Present Analysis

Display:

Root Cause Analysis

Level: {level} — {level_description}
Summary: {root_cause_summary}

Affected code:
  {affected_files list}

Upstream documents affected:
  {upstream_docs_affected list}

5.2 Per-Level Confirmation

For each affected upstream level (from lowest to highest), use AskUserQuestion:

"Root cause traced to {level_name}: {doc_path} — {what's wrong}. Update this document?"

Options:

  • "Yes, update this document" → mark for update in Step 7
  • "No, code fix only" → skip this level, do not modify this document
  • "Show proposed changes" → display the proposed diff for this document, then re-ask

5.3 Generate Fix Plan

Compile the fix plan:

  • Code changes to make (always)
  • Upstream documents to update (based on user confirmations)
  • Regression test to write

Step 6: TDD Fix

Execute the fix following TDD methodology:

6.1 Write Regression Test

Write a test that reproduces the bug:

  • Test must FAIL with the current code (proving the bug exists)
  • Test must describe the expected correct behavior

Run the test to verify it fails.

6.2 Implement Fix

Make the minimal code changes to fix the bug.

Run the regression test to verify it passes.

6.3 Run Full Test Suite

Run the project's full test suite to ensure no regressions:

  • If tests pass: continue
  • If tests fail: investigate and fix before proceeding

6.4 Commit Fix

Commit the code changes with a descriptive message:

fix: {brief description of bug fix}

Step 7: Upstream Document Sync

This step only runs if upstream documents were confirmed for update in Step 5.

For each confirmed document update:

7.1 Show Diff Preview

Before modifying each document, show the proposed changes in diff format:

--- a/{doc_path}
+++ b/{doc_path}
@@ ... @@
- old content
+ new content

Ask user: "Apply this change?" (Yes / No / Edit manually)

7.2 Apply Changes

  • Level 2 (task.md): Update the task's steps, acceptance criteria, or files involved
  • Level 3 (plan.md): Update architecture, task breakdown, or risk sections
  • Level 4 (feature doc): Update requirements or scope

7.3 Commit Document Updates

Commit upstream document changes separately from code fix:

docs: update {doc_path} — traced from bug fix

Step 8: Update state.json

If the bug is associated with a feature:

  1. Read the feature's state.json
  2. Add a fixes array (if not present) to the feature-level metadata
  3. Append a fix record: {"bug": "brief bug description", "root_cause_level": 2, "root_cause": "brief root cause", "fixed_files": ["path/to/file.ext"], "commits": ["abc1234"], "doc_updates": ["tasks/auth-logic.md"], "fixed_at": "ISO timestamp"}
  4. Update state.json updated timestamp

If standalone bug (no associated feature): skip this step.


Step 9: Summary

Display fix summary:

Bug Fix Complete

Bug: {description}
Root Cause: Level {level} — {summary}

Code Changes:
  {files changed}

Regression Test:
  {test file and name}

Document Updates:
  {list of updated docs, or "none"}

Commits:
  {commit hashes}

Next steps:
  /code-forge:status {feature}    View updated progress
  /code-forge:review {feature}    Review all changes

适合场景

01

用户想查找某类 Agent Skill 时

02

需要根据任务场景推荐可安装能力包时

03

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

Codex

34.32%
按下载量换算54

Claude

27.83%
按下载量换算44

Cursor

17.94%
按下载量换算28

Gemini CLI

10.14%
按下载量换算16

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

执行命令

安装流程涉及命令执行,可能通过 npx skills add https://github.com/tercel/code-forge --skill fixbug 联网下载 Skill 或依赖。用户安装前应确认命令来源、仓库内容和执行环境。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills