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 skills add kanfu-panda/pdlc-skills --skill pdlc-loop-rungit clone --depth 1 https://github.com/kanfu-panda/pdlc-skillsWrote 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/kanfu-panda/pdlc-skills/pdlc-loop-run)<a href="https://agentmods.dev/skills/kanfu-panda/pdlc-skills/pdlc-loop-run"><img src="https://agentmods.dev/badge/skills/kanfu-panda/pdlc-skills/pdlc-loop-run/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/kanfu-panda/pdlc-skills/pdlc-loop-run"><img src="https://agentmods.dev/badge/skills/kanfu-panda/pdlc-skills/pdlc-loop-run.svg" alt="Reviewed on agentmods" width="80" 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 5.1 | $0.00028 | $0.01121 |
| Opus 5 | $0.00014 | $0.00561 |
| Sonnet 5 | $0.00006 | $0.00224 |
| Haiku 4.5 | $0.00003 | $0.00112 |
Grade A, and why
pdlc-loop-run 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 12d 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.
What it actually says
收敛循环引擎(loop-run)
把机械收敛段 tdd → implement → review 烧成一个自主循环:从 current_stage 出发逐阶段自动推进,直到 review_done 或被 blocked。这是 /pdlc-loop-next 之上的高层引擎——用户不必自己写 bash 循环。
⛔ 终态即
review_done(设计如此):本引擎只覆盖机械收敛段。到达review_done即成功停机,交人工决定是否/pdlc-ship。绝不自动进入pdlc-ship/pdlc-deploy——发布/部署是不可逆·外发操作,永远留人(--autonomous对它们无效)。prd/design 的关键取舍同样不在本引擎范围内。
两种运行形态
- 默认推荐 · 外部 Runbook(真进程隔离):长跑 / 过夜 / 多 feature 并行时,用独立进程逐轮跑,每轮全新进程 = 真 fresh context 最可信。本命令可打印该 Runbook 脚本(见 usage-guide「自主循环 Runbook」)。
- 便捷 · 插件内 Task 版(本命令默认行为):适合短收敛。每个 stage 派发给一个 fresh Task subagent(context 相对隔离),返回后读状态机决定推进 / 停 / block。
循环算法(Task 版)
- 从
$ARGUMENTS取功能ID;--max-steps取迭代上限(缺省 4)。读docs/.pdlc-state/<功能ID>.json。 - 循环,每轮:
- 按
/pdlc-loop-next的映射判定下一条命令:pdlc-tdd/pdlc-implement/pdlc-review/done/blocked。 done→ 成功停机,输出<<<PDLC done stage=review>>>,提示交人工/pdlc-ship。blocked→ 停机交还人类,输出<<<PDLC blocked reason="...">>>。- 否则用 Task 工具派发该命令到一个 fresh subagent,带
--autonomous,模型取目标 skill frontmatter 的recommended_model(无则继承)。 - subagent 返回后重新读状态机
last_phase_result:ok=false→ fail-stop:停机、不重跑同一 stage,输出 blocked 哨兵。current_stage未推进(违反 IRON LAW 第 6 条)→ stuck-stop:停机报错。ok=true且已推进 → 记一步,继续下一轮。
- 步数 >
--max-steps→ 上限停机(防病态空转烧 token)。
- 按
- 输出循环小结(跑了几步、终态、每步
last_phase_result摘要)。
护栏(不可协商)
- 迭代上限:收敛段线性前进
tdd→implement→review(3 段),默认上限 4 = 3 段 + 1 段容错余量(供「首轮 block、人工修好后从原 stage 续跑一次」)。不给更多余量。 - 只前进不回炉:循环只在不同 stage 间前进,永不原地重跑同一 stage(与 IRON LAW「修复单次不递归」一致——单 stage 内的修复仍是单次)。
- fail-stop / stuck-stop:任一 stage
ok=false或状态未推进即停,绝不空转。 - 预算:外部 Runbook 形态必须配
--max-budget-usd(见 usage-guide)。自主循环持续烧 token,护栏是硬要求。
与 IRON LAW 的关系
本引擎不违反「修复单次不递归」:单个 stage 的自检-修复仍单次;引擎做的是跨 stage 前进 + 失败即停 + 上限,不是「重试同一步」。
功能ID: $ARGUMENTS
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.
- 12d ago First seen · 55 lines · 28 tokens per session scan A 2deb51555129
pdlc-loop-run is a skill published in the GitHub repository kanfu-panda/pdlc-skills (13 stars, last pushed today), licensed MIT. It adds 28 tokens to every session and 1,121 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 2026-08-31.
Other skills, from other repositories
execute
Dispatch and execute implementation plans with TDD and checkpoints. Use when plan is ready. Parallel by default for independent tasks.
tdd
Strict test-driven development for behavior changes. Requires verified RED before production code, minimal GREEN, and refactor only after passing tests.
tdd-xfail
Use when fixing a bug through strict xfail reproduction and needing proof that the test fails for the intended reason.
tdd-fix
TDD bug-fix workflow — reproduce a bug as a failing test, find root cause, fix, and verify.
tdd
Guides feature work and bug fixes through a red-green-refactor loop using vertical slices. Use when the user wants test-first development, acceptance criteria implemented safely, or one behavior delivered at a time.
aidex-bugfix
Use when fixing a bug, resolving a reported issue, or when something is broken and needs a test-driven fix — investigate root cause, write a failing regression test (RED), implement the minimum fix, confirm the test passes (GREEN), then commit test and fix together. Fires on "fix this bug", "this is broken", "it's not…