跳到正文
cccskills
免费在 GitHub 公开

idle

The idle work loop - enumerate open issues and triage rows, classify fixable vs blocked, pick by value-vs-budget, dispatch. Standing user directive; invoke whenever otherwise idle.

中文简介正在准备中。当前显示原文说明。

查看安装方法

包含的文件(1)

  • SKILL.md1.9 KB

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

Idle Work Loop

Standing directive: there is no "waiting" state. Between lane completions, after a release, after a merge train drains — pick the next item and work it. New problems found along the way become issues at identification time, keeping the queue self-replenishing.

1. Enumerate

powershell.exe -Command 'gh issue list -R jamie8johnson/cqs --state open --limit 50 --json number,title,labels'
grep -c '| open |' docs/audit-triage.md   # plus the rows themselves if any

2. Classify

  • Blocked-external (upstream PRs, dependency releases, missing hardware, user-parked) — skip. Known standing members: cuvs upstream chain, ort RC, anything needing a Mac or native Windows.
  • Umbrella/design — only enter with explicit scoping; closing umbrellas orphans phases.
  • Fixable — everything else. Note effort (easy / medium / hard) and value.

3. Pick

Judgment call, biased by: user-stated priorities first; then highest-value-per-token. Mind the rate window — heavy lanes (schema work, parser work, large refactors) want ~25%+ of a 5-hour window remaining; near a reset, prefer trivial sweeps or solo-fixable items. If the issue list is dry, run /archeo to refill it, or check docs freshness (/docs-review if a week stale).

4. Dispatch

  • Implementation → lane-implementer agent (opus), one branch per coherent batch; worktree isolation.
  • Risky lanes (live scoring path, schema migrations, cross-surface signatures) → fable code-reviewer pass before landing.
  • Security-adjacent → opus for the analysis lane.
  • Land via /land; close linked issues on merge (verify, don't assume).

5. Loop

On lane completion: land it, file its residuals, return to step 1. Update tears when the board changes shape.

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

archeo

免费

Comment-archaeology sweep - find work items hiding in TODO/FIXME/for-now comments and plan-doc ledgers, judge them, file issues, queue trivial fixes. Good idle-loop activity.

中文简介正在准备中。当前显示原文说明。

jamie8johnson/cqs172026年7月14日 更新

audit

免费

Run a multi-category code audit on the cqs codebase. Spawns parallel agents per batch.

中文简介正在准备中。当前显示原文说明。

jamie8johnson/cqs172026年7月14日 更新

Run before modifying a function — impact + test-map + explain in one pass, presented as a modification checklist.

中文简介正在准备中。当前显示原文说明。

jamie8johnson/cqs172026年7月14日 更新

Trust-boundary security audit of cqs — fans out unbiddable-auditor agents across 6 categories (RT-INJ/RT-FS/RT-RES/RT-DATA/RT-RELAY/RT-EXFIL), attacker mindset, run-the-attack PoC + regression guard.

中文简介正在准备中。当前显示原文说明。

jamie8johnson/cqs172026年7月14日 更新

Run after making changes, before committing — reviews the current git diff for impact and risk via cqs review.

中文简介正在准备中。当前显示原文说明。

jamie8johnson/cqs172026年7月14日 更新

cqs

免费

Unified cqs CLI dispatcher — semantic code search, call graph navigation, impact analysis, quality tools, notes, and infrastructure. Use `/cqs <command> [args]` for any cqs operation.

中文简介正在准备中。当前显示原文说明。

jamie8johnson/cqs172026年7月14日 更新

查看 jamie8johnson 的全部技能

报告此技能的问题