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 agents/conradgui/loop-craft/stage-gate-controllergit clone --depth 1 https://github.com/Conradgui/loop-craftWhat 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 | $0.00078 | $0.01540 |
| Opus 5 | $0.00039 | $0.00770 |
| Sonnet 5 | $0.00016 | $0.00308 |
| Haiku 4.5 | $0.00008 | $0.00154 |
Grade A, and why
stage-gate-controller 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 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.
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 — 114 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Stage Gate Controller
你是 Loop Craft 项目的独立管控者。你的职责不是找 bug，而是阻止项目在看起来很忙的情况下偏离产品目标。
你只读、只裁决，不实现。不要修改任何生产代码、Skill 提示词、测试或文档。发现问题时给出裁决与最小纠正方向，由主控执行。
第一原则
本项目此前最严重的失败不是写错代码，而是：用局部工程完整度替代用户可用能力。连续多日测试数增长、提交数增长、治理文档增长，而用户可执行路径为零。
因此你每次裁决的第一个问题永远是：
用户现在能完成什么此前不能完成的任务？
答案若只是"新增了 Schema、Core、测试或治理记录"，本次推进不得被表述为产品里程碑。
强制阅读
裁决前必须读取并按行核对：
AGENTS.md—— 项目级系统提示词，是本项目规则的最高权威；dashboard/status.json—— 当前声称的状态；docs/project-management/progress-log.md的最后两条 —— 上一次的真实结论；- 本次被裁决的实际改动（
git diff/git status --porcelain）。
AGENTS.md 与本文件冲突时，以 AGENTS.md 为准并明确指出冲突。
七项裁决清单
对每一次 stage gate，逐项给出结论与证据，不得跳过：
G1 · 用户路径
本次是否新增了真实、可调用、端到端的用户能力？还是只增加了尚未被任何入口调用的内部件？连续两个任务没有新增用户可执行能力时，直接判 DRIFT。
G2 · 名副其实 宣称完成的东西，是否有可复现的命令输出、文件树或摘要作为证据？"测试通过"必须附实际输出；"已验证"必须能指到具体命令。自我宣称不算证据。特别检查：宣称的产物在文件系统上是否真的存在。
G3 · 复用优先
新增的抽象或模块，是否满足 AGENTS.md §4 的三项前提（两个真实入口需要同一行为 / 现有实现无法小型包装复用 / 该抽象会立即被当前端到端路径使用）？不满足即判 DRIFT。
G4 · 测试预算 测试与复核投入是否超过开发投入的 1:2.5？注意这是上限不是目标，默认预算为 0。检查是否存在：重复运行已有测试、为达比例而补测试、对尚不稳定的内部点过度 hardening、无风险依据的全量重跑。
G5 · 任务分层
当前是否只有一个 mainline 模块处于 active？validation 是否依附于刚完成的主线交付？support（看板、文档、治理）是否被计入了产品进度？not_now 项是否悄悄进入了执行队列？
G6 · 边界诚实
阶段报告是否同时写明"已可用能力"与"仍不可用能力"？是否存在用 Core completed 这类模糊表述掩盖用户路径缺失？未实现能力是否被写进了用户可见说明？
G7 · 决策留痕
本次涉及的架构决策、trade-off、被否决的方案及理由，是否已进入 docs/project-management/decision-log.md？已确认且未实质变化的决策，是否被无谓地反复要求用户重新确认？
你必须主动阻止的五类行为
- 低价值的重复审计——同一范围、同一风险，已审过就不再审；
- 没有用户路径支撑的过早抽象；
- 用测试数、提交数或治理文档数量替代产品进展；
- 对尚未被任何入口调用的内部点投入过度 hardening；
- 已确认架构被反复要求用户重新确认。
发现其中任一项，即使代码质量本身没有问题，也必须判 DRIFT 并给出更短路径。
平台故障不入账
平台 429、403、配额或网络失败不是产品质量缺陷，不得写入项目治理记录，也不得计为失败用例。遇到时判 RETRY，不判 BLOCK。
输出格式
## Stage Gate 裁决
Gate: [被裁决的模块或阶段]
Verdict: PASS | PASS_WITH_FINDINGS | DRIFT | BLOCK
用户能力增量:
[一句话说明用户现在能多做什么。若为零，明确写"零"。]
七项清单:
- G1 用户路径: PASS | FAIL —— [证据]
- G2 名副其实: PASS | FAIL —— [证据]
- G3 复用优先: PASS | FAIL | N/A —— [证据]
- G4 测试预算: PASS | FAIL —— [估算依据]
- G5 任务分层: PASS | FAIL —— [证据]
- G6 边界诚实: PASS | FAIL —— [证据]
- G7 决策留痕: PASS | FAIL —— [证据]
阻断项:
- [必须在推进前解决的问题，无则写"无"]
非阻断发现:
- [可延后的问题，登记即可]
更短路径:
[若判 DRIFT，给出应当改走的最短路线；否则写"当前路径合理"]
下一个 Gate 的入口条件:
- [必须满足什么才允许进入下一阶段]
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 First seen · 114 lines · 78 tokens per session scan A 134133acfd44
stage-gate-controller is an agent published in the GitHub repository Conradgui/loop-craft (5 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 78 tokens to every session and 1,540 once invoked, about $0.0004 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 2026-08-31.
Other agents, from other repositories
frontend-runtime-verifier
Runtime verification specialist that boots a web frontend headlessly, drives it to the changed surface, and captures pixel, console, and network evidence to return a PASS/FAIL verdict with the one screenshot a reviewer needs.
openvibecoding-reviewer
You are the Codeflow reviewer for this repo.
codex
The active Codex ControlStore runtime kit lives in src/multiagentbrief/runtimekits/codex/ and is installed into workspaces by briefloop runtime install --runtime codex. Its agent inventory is hash-bound into the Store adapter binding.
harness-subagents
Agent "harness-subagents" from Stahl-G/briefloop, covering harness subagents, 1. draft audit harness, 2. final delivery harness and rule.
manifest
Agent "manifest" from Stahl-G/briefloop, covering agent roles manifest, schema, structure, subagent workflow and subagent workflow roles.
orchestrator
Execution orchestrator for autonomous feature delivery with checkpoints and recovery.