菜单
Use when the user says "bug:", "feature:", "feat:", "fix:", "issue:", "chore:", or explicitly "/issue". File a GitHub issue without interrupting the current workflow. Run as a background Task so the main conversation continues unblocked.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
基于 SOC 职业分类
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/flavordrake/mobissh --skill issue命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
| name | issue |
| description | Use when the user says "bug:", "feature:", "feat:", "fix:", "issue:", "chore:", or explicitly "/issue". File a GitHub issue without interrupting the current workflow. Run as a background Task so the main conversation continues unblocked. |
Process reference:
.claude/process.mddefines the label taxonomy, workflow states, and conventions that this skill must follow.
File a GitHub issue from an in-conversation observation. The user types something like
bug: default WSS URL is missing /ssh and expects it handled without derailing current work.
Spawn a general-purpose agent with the prompt from .claude/agents/issue-manager.md.
Use model: "sonnet", run_in_background: true. Custom subagent_types are broken in
file-based discovery (see .claude/rules/agents.md).
Background agents auto-deny permissions not pre-approved in settings.json. If the
agent fails on permissions, run it in foreground instead or file directly in the main
conversation.
Extract from the user's message. Apply exactly one Type label per .claude/process.md:
| Prefix | Type label | Title prefix |
|---|---|---|
| bug: | bug | bug: |
| fix: | bug | fix: |
| feature: | feature | feature: |
| feat: | feature | feat: |
| chore: | chore | chore: |
| issue: | (classify from description) | (classify from description) |
If the prefix is issue: or /issue, read the description and pick the best type label.
Add Domain labels if the description matches keywords:
| Keywords | Label |
|---|---|
| touch, gesture, swipe, pinch, scroll | touch |
| ios, safari, webkit, iphone, ipad | ios |
| security, vault, credential, encrypt | security |
| ux, ui, layout, mobile, keyboard, panel | ux |
| image, sixel, kitty, iterm | image |
Add Shape labels when applicable:
| Condition | Label |
|---|---|
| Issue needs research before code can be written | spike |
| Issue needs emulator/device validation | device |
| Issue is too large for one bot pass | composite |
Do NOT apply delegation labels (bot, divergence) -- those are managed by /delegate
and /integrate respectively.
Build a concise issue body. No filler.
scripts/gh-ops.sh version
Code: abc1234 | Server: 0.1.0:def5678 (or server not running).
If they differ, flag it as (STALE -- server hasn't been restarted).Check for recent test artifacts in either test results directory. If the report JSON
was modified within 30 minutes (stat -c %Y vs date +%s), include relevant evidence:
Appium tests (primary):
Legacy CDP emulator tests:
Both paths:
4. Per-test screenshots: check test-results/*-android-emulator/ for relevant PNGs
If a recording exists but no frames have been extracted (no frames/ directory or it's
empty), run scripts/review-recording.sh to generate uniform frame samples. This
is useful when the issue relates to visual behavior that screenshots alone don't capture.
Only include artifacts clearly connected to the issue. If nothing is recent or relevant, skip this section entirely.
If the issue is actionable (bug with clear reproduction, feature with clear scope), append:
@claude <Specific instruction for what to investigate or implement.
Reference relevant files, test commands, or prior issues.>
Do NOT add @claude for research issues (spike), things needing real-device validation
(device), or vague requests that need scoping. The /delegate skill handles enriching
issues with full delegation context and applying the bot label -- this step is just a
lightweight hint for obviously bot-ready issues.
Before filing, check for existing issues:
scripts/gh-ops.sh search "<key phrase from title>"
If a match exists, warn the user and ask whether to file or comment on the existing one.
Write the body to a temp file, then file using the wrapper script. Body template:
Filed while working on <context> (<branch>).
<Problem statement or feature description with technical details.>
## Reproduction
<Steps to reproduce, or user need for features.>
## Version
<output from `scripts/gh-ops.sh version`>
<If mismatched: "(STALE -- server hasn't been restarted)">
## Test Evidence
<Only if recent artifacts exist (Step 3). Otherwise omit this section entirely.>
- Run: <pass/fail summary from report.json>
- Failed: <test names and error snippets>
- Frames: <relevant filenames from frames/ or review/>
- Recording: test-results/emulator/recording.mp4 (<timestamp>)
@claude <Specific bot instructions, if actionable (Step 4).>
Write the composed body to a temp file, then file:
# Write body to temp file (use the Write tool, not a heredoc)
# Then file:
scripts/gh-file-issue.sh --title "<prefix>: <title>" --label "<type>" --label "<domain>" --body-file /tmp/issue-body.md
Important: Use the Write tool to create /tmp/issue-body.md, then call the script.
Do NOT use heredocs or $(cat <<EOF) in the bash command -- that's what causes
per-command approval prompts. One Write + one scripts/gh-file-issue.sh call.
Only include --label flags for labels that apply. Type is always one. Domain and shape
are zero or more. See .claude/process.md for the full label taxonomy.
Output: Filed: <issue-url>
If filing failed, report the error. Do not retry silently.
bug:) -- ask for at least a one-phrase descriptiongh not authenticated -- report error immediately.claude/process.md -- all are pre-created on the repoUse when the user says "delegate", "assign bot work", "dispatch issues", "triage open issues", "send to bot", or explicitly "/delegate". Scans open GitHub issues, classifies which are bot-delegatable, enriches them with design direction and context, and assigns via @claude comments. Analyzes prior bot failures and decomposes large issues.
Use when the user says "crystallize", "compact", "compile", "distill", or asks to reduce a skill's reliance on LLM prose reasoning. Phase-transitions deterministic operations out of skill docs into real scripts with real test data. Rewrites the SKILL.md to be shorter and call the scripts instead of describing the logic. Aggressive about the boundary: LLM does judgment, prose, ambiguity resolution. Scripts do math, time comparison, parsing, counting, aggregation, schema validation.
This skill should be used when the user asks to "test on device", "test on emulator", "run emulator", "launch AVD", "test PWA", "test on Android", "test on mobile", "verify on real device", "check on phone", or discusses testing a feature on an actual device or emulator rather than headless Playwright. Also use when validating features that headless browsers cannot cover (biometric, PWA install, Chrome autofill, touch gestures, password managers). Use proactively when a feature has been implemented that touches any of these capabilities.