Skip to main content
ClaudeWave
ES, Spanish
Skill2.9k estrellas del repoactualizado 28d ago

bug-capture

bug-capture converts conversational bug reports into durable GitHub issues using the project's domain language rather than implementation details. Use it when a user describes a defect, requests issue logging, or conducts QA testing, as it explores the codebase for vocabulary context, checks for duplicates, and files issues without exposing file paths or line numbers that become stale after refactors.

Instalar en Claude Code
Copiar
git clone --depth 1 https://github.com/rohitg00/pro-workflow /tmp/bug-capture && cp -r /tmp/bug-capture/skills/bug-capture ~/.claude/skills/bug-capture
Después abre una sesión nueva de Claude Code; el skill carga automáticamente.

SKILL.md

# bug-capture

Turn a conversation into an issue that still reads correctly after a
major refactor.

## Flow

### 1. Listen, then clarify minimally

Let the user describe the problem in their own words. Ask at most two
short clarifying questions, drawn from:

- Expected behavior vs. actual behavior.
- Concrete reproduction steps if not already implied.
- Frequency: deterministic, intermittent, or one-off.

If the description already answers these, skip straight to filing. Over-
interviewing is a tax the reporter pays for your uncertainty.

### 2. Explore in parallel

While the user is answering, start a background exploration of the
relevant area. The goal is **not** to propose a fix. The goal is to
absorb the project's own vocabulary — the nouns and verbs the codebase
uses for this feature — so the issue reads like it was written by a
maintainer.

If the repo has a glossary file (common names: GLOSSARY.md,
UBIQUITOUS_LANGUAGE.md, docs/domain.md), read it first.

### 3. Check for duplicates before filing

Run `gh issue list --search "<key phrase>" --state all --limit 10`. If a
live or recently closed issue matches, surface it to the user and ask
whether to add a comment instead of opening a new issue. Do not silently
skip filing.

### 4. Decide: single issue or breakdown

Break down when the report contains two or more independent failure
modes that a different contributor could fix in parallel. Keep as one
when every symptom traces to a single wrong behavior.

For a breakdown, file in dependency order so each child issue can
reference a real parent number, and mark honest `Blocked by` links.
Avoid inventing dependencies to make the tree look tidier.

### 5. File with `gh issue create`

File without asking the user to review the draft. Send back the URLs.

#### Single-issue template

```
## What happened
<observed behavior, in domain terms>

## What I expected
<expected behavior>

## Reproduction
1. <step>
2. <step>
3. <step>

## Context
<anything that narrows where the bug lives, in domain terms — e.g.
"only affects the import path, not the export path">
```

#### Child-issue template

```
## Parent
#<parent-number>

## What is wrong
<one behavior, narrow slice>

## What I expected
<expected for this slice>

## Reproduction
1. <step>

## Blocked by
<#issue or "none — independent">

## Context
<notes that apply only to this slice>
```

### 6. Rules that apply to every issue body

- No file paths, line numbers, function names, or PR numbers. These go
  stale. Describe behavior, not code.
- Use the project's domain nouns, not generic tech terms. "The sync
  worker drops the patch" beats "`applyPatch()` throws".
- Reproduction steps are mandatory. If you cannot derive them, go back
  to the user once before filing.
- Thirty-second read target. Cut anything that does not help a
  maintainer decide whether to pick it up.

### 7. Keep going

After each issue, print the URL and ask whether there is a next one. Do
not batch multiple reports into one filing pass — each bug deserves its
own scoped issue.
agent-teamsSkill

Coordinate multiple Claude Code sessions as a team — lead + teammates with shared task lists, mailbox messaging, and file-lock claiming. Patterns for team sizing, task decomposition, and when to use teams vs sub-agents vs worktrees.

auto-setupSkill

Auto-configure quality gates, hooks, and settings for a new project. Detects project type and sets up appropriate tooling. Use when onboarding a new codebase.

batch-orchestrationSkill

Decompose large-scale changes into independent units and spawn parallel agents in isolated worktrees. Use for migrations, refactors, codemods, and any change touching 10+ files with the same pattern.

compact-guardSkill

Smart context compaction with state preservation. Saves critical files, task progress, and working state before compaction, restores after. Use before manual compact or when auto-compact triggers.

context-engineerSubagent

Analyzes and optimizes context window usage across sessions. Use when context feels bloated, sessions run slow, or approaching compaction limits.

context-engineeringSkill

Master the four operations of context engineering — Write, Select, Compress, Isolate. Manage token budgets, compaction strategies, and context partitioning to keep AI sessions sharp and efficient.

context-optimizerSkill

Optimize token usage and context management. Use when sessions feel slow, context is degraded, or you're running out of budget.

cost-analystSubagent

Analyze session token usage and cost patterns. Identify expensive operations and recommend optimizations. Use to understand and reduce session costs.