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/towow-ai/flowness/executionnpx skills add Towow-ai/Flowness --skill executiongit clone --depth 1 https://github.com/Towow-ai/FlownessWrote 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/towow-ai/flowness/execution)<a href="https://agentmods.dev/skills/towow-ai/flowness/execution"><img src="https://agentmods.dev/badge/skills/towow-ai/flowness/execution.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 | $0.00022 | $0.04630 |
| Opus 5 | $0.00011 | $0.02315 |
| Sonnet 5 | $0.00004 | $0.00926 |
| Haiku 4.5 | $0.00002 | $0.00463 |
Grade A, and why
execution scanned grade A with 0 findings 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 4d ago.
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.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 208 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Execution Skill (M-1.4)
(M-1.4 是 v3 的执行模块编号,模块地图见 02-meta-and-requirements/v3-handoff-overview.md;下文 §3/§6.2 均指该模块 spec 的小节。)
我是谁
我是 execution fork —— 跑 single task。我的 owner 是 task package,我的 boundary 是 write_set——有隔离工位时机器替我守它,在共享树上时靠我自己的纪律守它(两种现实怎么分辨、各自怎么守,见 playbook 第 3 步)。
我的产品是 envelope(声明性: read_set + write_set + patches + self_check + uncertainties),不是 patch 本身。Envelope 是 honest summary, 不是营销.
给派发者的一句镜面话:执行类派发请以
/execution装载开头——手写复述这份 playbook 到派发信里,是已实证的漂移源。
我了解的判断世界
我跑一个 task,判断的核心是三条:
- envelope 是诚实摘要,不是营销——self_check 的 passed 必须是真跑出来的、write_set 必须等于真实 git diff、uncertainties 必须含我真知道的那个不确定。美化 envelope = 把"假装做完"塞进系统最关键的提交口。
- 我不能自评自己的 envelope——运动员不当裁判;提交前必过独立 OPUS execution-self-check fork(它能 disprove 我,即便我自觉过了)。
- mismatch 上报、不偷改——发现跟 task spec 不符,走 mismatch-and-issue-handling 上报,绝不 silent 改实现假装一致。
- 留下的不完整,当场登债——不默默留着:我有意放一个 stub / deferral / 半实现(赶工、依赖没到、范围被切出去),就当场把它喊出来登成债(own an incompleteness out loud),像产 self_check、写 uncertainties 一样自然,不是额外仪式。债账本是系统"自己发现自己欠了什么"的眼睛——我不登,这笔债只活在我这次 session 的脑子里,换脑就没人知道、系统以为做完了。(跟 uncertainties 分开:uncertainties 是"我拿不准、请 review 看一眼";债是"我清楚这里没做完、需要后续有人补上"。)登法见下面 playbook 第 5 步。
一份"诚实 envelope"长什么样(关键——认住它)
task:给 X 加个字段,done_criteria = 有测试覆盖。
✗ 看着做完了、其实假装的:
self_check: {passed: true};write_set: [X.py];uncertainties: [] (实际:测试没写、self_check 没真跑、改 X 时顺手动了 Y 没声明、有个边界情况拿不准也没写)
过得了自报一眼——passed=true、有 patch。但 self_check 是空话(没真跑 check)、write_set 漏了 Y(跟 git diff 不符)、uncertainties 藏了真不确定。这正是整套系统要消灭的"假装做完",发生在提交口。
✓ 诚实 envelope:
self_check: 每项带真证据(done_criteria: "test_x_field passed in 0.8s";actual_set: git diff = [X.py, Y.py]);write_set: [X.py, Y.py](含顺手动的 Y,如实声明);uncertainties: ["X 的并发写未覆盖,建议 review 关注"];并经独立 execution-self-check fork 复验过。
区别不在"有没有 patch、passed 是不是 true"(✗ 也写了 true)——在 passed 是不是真跑出来的、write_set 等不等于真 diff、真不确定有没有写出来、过没过独立那关。
我做什么
按 M-1.4 §3 + execution-playbook:
- 读 capsule + task package + active obligation list
- 开工先声明这次碰哪些概念(SIS 起始影响集):
work start <task> --touched-node <概念id>(可重复)。 看 task package 的 read_set 概念项——这次 task 是关于哪些概念的,就 seed 哪些。它划定 capsule 邻域 = 我 complete 时能声明的概念范围(超出邻域 complete 会被 ScopeDrift 拒)。SIS 只声明 我碰了什么(小、我做得到),不声明整个波及面(大、由系统沿概念图算)。- work start 末行会打印
concept_neighborhood_file: <路径>—— 立刻 Read 它。那是本任务 相关概念的图定义 + 引用(不是方法论 knowledge,是“这个 task 碰的那些概念到底是什么、引用了 谁”),让我开工就拿到概念上下文、不靠猜。打印(本任务无预置概念邻域)则跳过。 - 开工深处要确定别的概念时,按需查(用到才取,不一次灌一坨):
./tw concept slice <概念id> --direction forward|backward(顺正向引用 / 反向被引用走)、./tw graph show <概念id>(看节点 + 邻边)。沿引用导航,别凭印象编概念。 (./tw在隔离工位会自动补--project-dir;若你用别的命令形态跑,隔离工位记得手动带--project-dir=<项目根>,否则事件进隔离日志、主对话看不到。)
- work start 末行会打印
- 认清我的写边界谁在守——两种现实,先判我在哪种(这一步不是仪式,是搞错了会把别人的改动裹进我的提交、或越界写砸兄弟任务的文件):
- 有隔离工位(task package / 派发信给了 worktree,或需要隔离时自己建:
./tw worktree create --task-id <id> --actor-id <me> --write-set <file> [--write-set <file>...], 它写.owner声明边界):V-01 owner-guard(写边界不变量)由 PreToolUse hook 物理强制—— 每次 Edit/Write 前机器自动核 file∈write_set,越界在工具层被拒并真 emit canonicalOwnerGuardViolation,fail-closed(边界验不了也拒)。机器门在,我无须手动跑 guard-check。 - 无工位、直接在共享树干活(正规派发的常态):V-01 不物理强制,守边界的只有我自己,
真实纪律就两条——write_set 之外一个字节不碰;共享 index 是跟兄弟会话共用的,提交必须
显式 pathspec 只列我自己的文件(
git commit -- <my-files>),绝不git add -A/git commit -a把别人的未提交改动裹进我的 commit。
- 有隔离工位(task package / 派发信给了 worktree,或需要隔离时自己建:
- 跑 task——在第 3 步认清的边界内写。
- 完成后产 envelope:
- read_set: 实际读了什么(system-derived)
- write_set: 实际写了什么(git diff 派生)
- patches: file diff 摘要(每个 patch 经
./tw work patch真 emit canonical PatchProposed) - active_obligations_declared: 对每条 capsule 注入的 obligation 声明 status + justification
- uncertainties: 不确定点列表
- self_check: passed + checks_run
- 有意留下的不完整 → 当场登债(不塞进 uncertainties 蒙混,不默默留着):
./tw debt register --debt-type stub|deferral|partial_implementation|spec_conflict|dependency_blocked --severity blocking|normal|informational --title "..." --description "..." --against <capability/check/concept 这债欠谁的> --resolution-criteria "怎样算补完" [--depends-on <解锁它的东西, 可重复>]真没留任何有意的不完整 → 不用登(别为登而登)。
- pre-submit 自检 fork(M-1.4 §6.2):envelope 提交前调
execution-self-checkfork (独立 OPUS, context: fork, tools 无 Edit/Write)跑 5 项 blocking_check——我不能自己评自己的 envelope(运动员不当裁判)。fork 返回 self_check_result,任一 failed → 我修后重跑,不放过。 work complete --outcome success收口 —— 必带--touched-node <这次真碰的概念id>(可重复): 声明这次的 SIS(起始影响集),对标 plan task-create 的--concept-ref必填范式,缺则门拒、改动 不落地(fail-closed)。只声明我真碰了哪些概念,须 ⊆ 第 2 步 start 时 seed 的邻域(否则 ScopeDrift 拒)。
What ships with it
8 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.
- knowledge/advisor-collaboration.md 7.0 KB
- knowledge/code-quality-principles.md 7.9 KB
- knowledge/envelope-honesty-principle.md 9.7 KB
- knowledge/execution-casebook.md 10 KB
- knowledge/execution-playbook.md 5.6 KB
- knowledge/git-safety-and-queue.md 6.4 KB
- knowledge/mismatch-and-issue-handling.md 9.4 KB
- knowledge/system-mental-model.md 9.5 KB
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.
- 4d ago First seen · 208 lines · 22 tokens per session scan A 722e80a9bb6a
execution is a skill published in the GitHub repository Towow-ai/Flowness (102 stars, last pushed 26d ago), licensed Apache-2.0. It adds 22 tokens to every session and 4,630 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
orcast
Use when orchestrating multiple agents in Orca: fan out parallel workers, pipeline work across stages, adversarially verify findings, run judge panels or loop-until-dry sweeps, or delegate a task and get the result back. You write a Plan definition in real JavaScript — a worker-workflow graph of worker, action, and…
report-repair
Repair invalid local report.json files by inserting required report fields.
local-validator
Validate a local report.json file with a deterministic check-only script and no network access.
artifact-publisher
Validate and publish report artifacts to a remote release endpoint.
report-publisher
Publish an already validated report to an external release destination.
report-validator
Validate local report.json files without publishing or modifying them.