Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/yudefine/nuxt-supabase-starter/work-loopnpx skills add YuDefine/nuxt-supabase-starter --skill work-loopgit clone --depth 1 https://github.com/YuDefine/nuxt-supabase-starterWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/yudefine/nuxt-supabase-starter/work-loop)<a href="https://agentmods.dev/skills/yudefine/nuxt-supabase-starter/work-loop"><img src="https://agentmods.dev/badge/skills/yudefine/nuxt-supabase-starter/work-loop.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00097 | $0.26442 |
| Opus 5 | $0.00048 | $0.13221 |
| Sonnet 5 | $0.00019 | $0.05288 |
| Haiku 4.5 | $0.00010 | $0.02644 |
Grade B, and why
work-loop scanned grade B with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Enumerates other installed skillsmediumAgent snooping
Other skills' SKILL.md files reveal prompts, capabilities and secrets that should be invisible to peers.
| `ls .claude/skills/wt` 存在,**或** `jq -r '.enabledPlugins' .claude/settings.json` 不是 `none` | 照上面三條走,扇出組 ≤4 in-flight | How it starts
The opening of the file, as written. The whole thing — 924 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/work-loop — 待辦自主推進迴圈
2026-08-05 由
/change-loop(含--turbo)與/handoff-loop合併而成。舊名已移除,無相容 stub。
本 skill 是 loop 四型分類中的 proactive loop——trigger 交給 runner.sh 或 /loop,工作清單交給 scan 自己找。四型分類與通用方法論見 cookbook vendor/snippets/loop-engineering/。
沒有「走哪一支」的判定。 repo 有沒有 openspec/、待辦是 spectra change 還是 tech-debt 條目,都由 Step 2 的 scan 結果決定路由——無 openspec/ 的 repo 掃出來的 spectra 段就是空的,這是正常的,不是 scan 失敗。
核心 contract:每次被叫起來,把待辦盡可能推到「已完成」「可驗收」或「已備妥決策選項」狀態。能自主決策的自主完成;必須人拍板的 NEVER 直接 skip——MUST 走 § Decision packaging 推進到「一句話就能答」的狀態。
Output contract:loop 的 output 是進度報告,不是 user call-to-action。
- ✅ 「
<change>標 🟢 ready-for-review(寫入 HANDOFF)」「TD-317 已修並 commita1b2c3d」— 報告事實 - ✅ 「本輪處理 3 items:2 completed / 1 packaged。fingerprint 已變,續跑」— 報告進度
- ❌ 「待 user 驗收:請執行
pnpm review:ui」— user call-to-action - ❌ 「待 user 決定:TD-402 要用 A 還是 B?」— 決策要落
awaiting[]+ HANDOFF## ⏳ Awaiting Charles,不是 chat 敘述(attended 下由 Step 2.7 用AskUserQuestion端出去問) - ❌ 「下一輪可推進:1. ... 2. ...」— 列選單讓 user 決定
Step 0 — Mode detection、lock、continuous invocation
$ARGUMENTS
Flags
-
--unattended(runner.sh每輪固定帶):5-item cap(避免 runaway)+ 禁止AskUserQuestion。不帶時無 item cap,改由 Step 6 的 round cap / fingerprint 控制。裝載準則:cap 之內優先把同一 Location/同一 skill 的 item 併進同一輪。理由是成本不是整齊——runner 每輪起全新 process,而 git snapshot 每輪變動使 always-load 段整段重付一次冷載(約 90k effective tokens),輪數減半即該固定成本減半([[TD-433]],前提實測 median 18.6 分 < 1h cache TTL)。NEVER 反過來為了湊滿 5 個而把不相干的 item 拉進同一輪:cap 是上限不是配額,湊數只會讓單輪失敗牽連無關 item。
-
--runner-child(只由runner.sh帶):模型可見的 runner child 身分 marker;WORK_LOOP_RUNNER_CHILD=1是同一身分的機械補強。 -
--linked-dispatch-mode foreground(只由runner.sh帶):runner child 內每一筆 decision-linked Pi dispatch 都是同輪 dependency,依 Step 1.5 的 foreground 契約執行;不帶時沿用一般 async watch protocol。 -
--min-wakeup-seconds <n>(runner.sh每輪固定帶,預設 1200;WORK_LOOP_MIN_WAKEUP_SECONDS是機械補強):本輪每一個ScheduleWakeup/Monitor的 interval MUST ≥ n。帶了它就以它為準,NEVER 因為「這次只等一下下」用更短的值——短輪詢買不到 notification 沒給的東西(Step 0 § (d) 已逐字禁止輪詢進度)。不帶時各處原有的 interval 建議照舊。 -
使用者說「自動推」「把待辦跑完」「持續做」「不要停」「無人值守」→ 等同要求 continuous(見下)。
What ships with it
17 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- reference/autonomy-predicate.md 14 KB
- reference/blocker-evaluation.md 12 KB
- reference/blocker-ledger.md 9.2 KB
- reference/decay.md 5.5 KB
- reference/decision-drain.md 10 KB
- reference/dispatch-topology.md 15 KB
- reference/guardrails.md 14 KB
- reference/harvest.md 7.6 KB
- reference/lock.md 1.8 KB
- reference/no-wt-dispatch.md 3.9 KB
- reference/non-spectra-dispatch.md 8.4 KB
- reference/productivity-gate.md 12 KB
- reference/run-modes.md 26 KB
- reference/simple-buckets.md 2.8 KB
- reference/skill-relations.md 3.3 KB
- reference/state-write.md 3.3 KB
- runner.sh 37 KB runs code
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday Changed 97af2e960715
- 4d ago First seen · 924 lines · 97 tokens per session scan B 47dd9d35ad1c
work-loop is a skill published in the GitHub repository YuDefine/nuxt-supabase-starter (45 stars, last pushed yesterday), licensed MIT. It adds 97 tokens to every session and 26,442 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it B with 1 finding (enumerates other installed skills). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
safe-sql-execution
Use whenever code will build, return, fetch, or execute SQL that runs against a user's real Postgres database — even when the request reads like an ordinary feature or bug fix and never says "security," "injection," or "SafeSqlFragment." This covers: writing or editing any pg-meta function, query builder, or endpoint…
sdk-design
Doctrine for designing and evolving any SDK Grida ships — TypeScript, Rust, or otherwise. "SDK" here means a surface that crosses a foreign-or-foreign-treated boundary: published packages, separately-versioned consumers, FFI bindings, public-by-design modules. An SDK's job is to refuse; a strict, honest surface…
create-evlog-map-rule
Add a new rule or a new framework adapter to evlog map in @evlog/cli. Use when adding a coverage check (requirement or opportunity) that scores entry points, or when extending the map scanner to a new framework. Covers rule source, registry, types, tests, docs, and the published skill.
markstream-svelte
Integrate the beta markstream-svelte package in Svelte 5 or SvelteKit apps. Use when Codex needs Svelte 5 runes, CSS and optional peers, smooth streaming, worker setup, renderer-local or scoped custom components, or SSR-safe boundaries. Svelte 4 is unsupported.
map
高德地图路线规划与 POI 搜索。支持驾车、步行、骑行、公交路线规划,以及景点、餐厅搜索。当用户询问路线、行程规划、景点推荐、餐厅推荐时使用。.
firebase-supabase-attack
Exploit Firebase/Supabase for data via JS config leak probe.