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/ryanzhao1011/workframe/self-iterationnpx skills add ryanzhao1011/workframe --skill self-iterationgit clone --depth 1 https://github.com/ryanzhao1011/workframeWrote 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/ryanzhao1011/workframe/self-iteration)<a href="https://agentmods.dev/skills/ryanzhao1011/workframe/self-iteration"><img src="https://agentmods.dev/badge/skills/ryanzhao1011/workframe/self-iteration.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 | $0.00023 | $0.04596 |
| Opus 5 | $0.00012 | $0.02298 |
| Sonnet 5 | $0.00005 | $0.00919 |
| Haiku 4.5 | $0.00002 | $0.00460 |
Grade B, and why
self-iteration scanned grade B with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Reads agent configuration directoriesmediumAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
python "$(cat .claude/workframe-state/plugin-root.txt)/scripts/maintenance_workorder.py" \ How it starts
The opening of the file, as written. The whole thing — 236 lines — stays where its author put it; the contents beside it link to each section on GitHub.
自迭代技能
5 阶段流程
阶段 1:数据收集 + 已应用提案闭环验证
(a) 数据源收集(主要在 .claude/workframe-state/(另含 .claude/agent-memory/ 与 projects/ 下若干路径,见下表) 下,由 hook / deterministic scripts / system skills 维护):
.claude/workframe-state/skill-metrics.yaml— 技能/规则使用汇总(由recompute_skill_metrics.py从 events.jsonl 重算).claude/workframe-state/events.jsonl— 原始事件流(审计/追因时读,日常决策读 metrics 即可).claude/workframe-state/activity-state.json— 活跃度 + dormant 状态 +pending_maintenance(status=open);若dormant=true或wake_up_pending=true则本次自迭代直接退出(除非由/core:maintenance-review显式触发)。pending_maintenance里的 kind/details 是本次识别模式的重要线索,应与 notes/events 证据一起纳入阶段 2 分析。.claude/agent-memory/*/notes.md— 各角色微反思.claude/agent-memory/shared/MEMORY.md和shared/notes.md— 跨角色共识projects/changelog.md— 历史操作日志projects/issues/— 历史问题记录(若有结构化文件)
(b) 扫描 projects/proposals/applied/*.yaml 中 verified: null 的条目做闭合验证:
- 对每条读取
verify_by(日期)和verify_signal(需观察到的信号表达式) - 若今天 ≥
verify_by:- 读取 skill-metrics.yaml / events.jsonl,判断
verify_signal是否已达成 signal_met=true→ append events.jsonl:{"ts":"<ISO-8601>","type":"proposal_verified","proposal_id":"<id>","signal_met":true};将提案文件中verified: truesignal_met=false→ append events.jsonl:{"ts":"<ISO-8601>","type":"proposal_verified","proposal_id":"<id>","signal_met":false},再 append{"ts":"<ISO-8601>","type":"proposal_failed","proposal_id":"<id>"}(供 audit / 下一轮 self-iteration 反思用;proposal_failed不计入check-iteration-trigger.py的 problem 加权分;recompute_skill_metrics.py实际统计的是proposal_verified.signal_met=false累加到proposal_failures_count,不直接读proposal_failed);将提案文件中verified: false
- 读取 skill-metrics.yaml / events.jsonl,判断
阶段 2:模式识别 + 置信度评分
从阶段 1 收集的数据中识别候选模式:
- 重复问题:notes / changelog / events 中有明确证据显示同类问题重复出现。
occurrences是置信度计算的证据输入,不是硬门槛;低于 3 次仍可计算 confidence,但通常低于提案阈值。 - 低效流程:仅当 notes / changelog / issues 中存在明确的耗时或阻塞记录时方可使用;系统无耗时事件,不得凭感觉声称"平均耗时高于预期"。
- 未覆盖场景:仅当 notes / changelog 中有明确的"用户重复手工处理"记录时方可使用;不得无证据臆造"手工处理"场景。
- 技能低成功率:近 30 天某 skill
success/invocations < 0.6(来自 skill-metrics.yaml)。仅供人工判读,不作自动触发信号(success为 agent 自评,实测从未产出 false,作触发条件永不满足)——用它时须结合 notes / user_correction 等独立证据,不得仅凭该比值提案。 - 提案失败回路:有
proposal_failed事件的旧提案 → 反思当初假设,识别失败原因。
What ships with it
3 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.
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 · 236 lines · 23 tokens per session scan B db66a70e96db
self-iteration is a skill published in the GitHub repository ryanzhao1011/workframe (4 stars, last pushed 17d ago), licensed MIT. It adds 23 tokens to every session and 4,596 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it B with 1 finding (reads agent configuration directories). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
discover
Runtime catalog — list all installed skills, agents, and active hooks. Use when you want to know what's available.
html-prototype
Generate HTML prototypes (single-file, browser-openable) for product mockups using the user's design system — fonts, color scales, radii, shadows, components, all driven by a tokens.css file. Use when the user asks to make a 原型 / mockup / HTML 设计图 / 设计图给前端 for a PRD module, OR redesign an existing prototype. Output…
feature-acceptance
研发提测后接管浏览器,对照 PRD(必给)+ HTML 原型(自动找)逐项验收线上实现,产出差异清单(实现 / 部分实现 / 未实现 / 与原型或 PRD 不一致)+ 截图证据 + 阻塞问题排序。Use whenever PM 说「验收 XX」「跑一下 XX」「研发提测了」「对一下需求和线上」「点一下 XX 看符不符合需求」「看哪些实现了哪些没做」「上线前 check」「灰度验收」「acceptance」「QA check」,or 给出 PRD 路径 + 线上 URL 让 Claude 走流程。也适用于 PM 想自己点某功能、判断和需求差异时陪她一起跑。区别于 prd skill:本 skill 验证线上现状,不修改 PRD…
prd
Write, rewrite, polish, restructure, or review product requirements documents (PRDs). Use whenever a task touches a 需求文档 / PRD —— 新写、重构、统一口径、补页面结构 / 字段 / 规则 / 边界 / 异常处理、加指标口径定义、按数据看板或某类骨架组织章节,或把讨论 / 会议记录整理成正式产品文档。Trigger even when the user does NOT say "写需求" —— e.g. "把这个加到…
gbg
Convert any plan, PRD, feature idea, or brief into a sequenced goal-by-goal execution document with reviewer-gated commits — for everyday feature development, MVPs, migrations, refactors, and hardening alike. Generates a project brief, per-goal scope/tasks/acceptance criteria, an LLM-reviewer prompt (Codex/etc.), a…
ui-walkthrough
把一份 UI 设计 HTML 和研发上线的页面,同视口、同状态截图比对 + 读 computed style 实锤,产出 UI 还原度差异清单(结构 / 组件 / 颜色 / 字体 / 圆角 / 图标 / 间距 七维,每条标明设计值 vs 线上值)。Use whenever PM 说「UI 走查」「走查一下 UI」「对一下 UI」「UI 还原度」「还原度走查」「线上跟设计稿对一下」「前端还原得对不对」「页面跟这份 HTML 对一下」「UI 验收」「设计还原走查」,或给出 设计 HTML 路径 + 线上 URL 让 Claude 比视觉差异。区别于 feature-acceptance skill:那个验业务逻辑 / 字段 / 操作…