[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-96006":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":14,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":15,"stars7d":15,"stars30d":16,"stars90d":15,"forks30d":15,"starsTrendScore":15,"compositeScore":17,"rankGlobal":9,"rankLanguage":9,"license":18,"archived":19,"fork":19,"defaultBranch":20,"hasWiki":21,"hasPages":19,"topics":22,"createdAt":9,"pushedAt":9,"updatedAt":23,"readmeContent":24,"aiSummary":25,"trendingCount":15,"starSnapshotCount":15,"syncStatus":26,"lastSyncTime":27,"discoverSource":28},96006,"agentflow","agfnow\u002Fagentflow","agfnow","File-logged conversation protocol and on-demand development pipeline",null,"JavaScript",121,29,3,1,0,14,45.83,"Apache License 2.0",false,"main",true,[],"2026-09-20 04:01:32","# agentflow\n\n**English** · [繁體中文](README.zh-tw.md)\n\nA file-logged way of working for AI coding agents (Claude Code, Codex, and compatible hosts).\n\n- **What it does:** the agent converses with you through a plain text record file (`devlog.md`) instead of the terminal. Every ask, answer, decision, and commit lands in the file, so the whole project history is auditable and any dead session recovers from the record.\n\n- **Two layers:** a base conversation protocol (rounds, STATUS, Git discipline, delegation rules) for any task, and an on-demand development pipeline (requirements → specification → implementation → acceptance) that loads only when a wish changes product behavior.\n\n## Install\n\n- **As a skill (Claude Code, Codex, and other hosts):**\n\n\t```\n\tnpx skills add agfnow\u002Fagentflow\n\t```\n\n- **As a Claude Code plugin (auto-updates on new releases):**\n\n\t```\n\t\u002Fplugin marketplace add agfnow\u002Fagentflow\n\t```\n\n## Verify installation\n\nRequirements: Node.js 18 or newer and Git.\n\nSet `AGENTFLOW_SKILL_DIR` to the actual directory where `npx skills add` installed `agentflow`, then run the installed setup check:\n\n\t```sh\n\tAGENTFLOW_SKILL_DIR=\"\u003Cactual installed agentflow skill directory>\"\n\tnode \"$AGENTFLOW_SKILL_DIR\u002Fscripts\u002Fsetup.js\"\n\t```\n\nThe first `godev` use installs the project's hooks. If an optional worker is unavailable, that is an availability result, not proof that installation failed.\n\n## Use\n\n- Type `godev` (or `\u002Fdevlog`) in a session to activate the protocol; type `ag` to force the full development pipeline on a wish.\n\n- The full user guide ships with the skill: see [`skills\u002Fagentflow\u002Fdocs\u002FAG_GUIDE.md`](skills\u002Fagentflow\u002Fdocs\u002FAG_GUIDE.md) (English) and [`skills\u002Fagentflow\u002Fdocs\u002FAG_GUIDE.zh-tw.md`](skills\u002Fagentflow\u002Fdocs\u002FAG_GUIDE.zh-tw.md) (繁體中文), including which AI models are strong enough to run it.\n\n## How model routing works\n\n- Each project keeps version-7 configuration in the adjacent `ag.json`. Its public roots and setting paths use kebab-case, including `schema-version`, `pipeline-roles`, `external-workers`, and switches such as `target-doc` and `allow-ag`. Project-defined worker profiles map the `best`, `better`, `basic`, and `cheap` tiers to literal commands and models. Unknown keys warn and are ignored; malformed JSON or invalid recognized values are rejected.\n","agentflow 是一个面向 AI 编程代理（如 Claude Code、Codex）的文件化对话协议与按需开发流水线工具。它通过纯文本日志文件（devlog.md）记录全部交互过程（提问、响应、决策、Git 提交），实现可审计的历史追踪与会话断点恢复；内置两层设计：通用对话协议（含轮次管理、状态标记、Git 规范、任务委派规则）和按需触发的开发流水线（需求→规格→实现→验收）。适用于需要可复现、可协作、可回溯的 AI 辅助软件开发场景，尤其适合中长期项目迭代与团队协同调试。",2,"2026-09-08 02:30:07","CREATED_QUERY"]