[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-80908":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":15,"stars7d":15,"stars30d":16,"stars90d":15,"forks30d":15,"starsTrendScore":15,"compositeScore":17,"rankGlobal":10,"rankLanguage":10,"license":18,"archived":19,"fork":19,"defaultBranch":20,"hasWiki":21,"hasPages":19,"topics":22,"createdAt":10,"pushedAt":10,"updatedAt":31,"readmeContent":32,"aiSummary":33,"trendingCount":15,"starSnapshotCount":15,"syncStatus":34,"lastSyncTime":35,"discoverSource":36},80908,"agent-workflows","franklioxygen\u002Fagent-workflows","franklioxygen","Engineering workflows for AI coding agents or flesh engineers. It helps absorb silent base-model quality drift.","",null,"Python",70,7,33,0,37,46.41,"MIT License",false,"master",true,[23,24,25,26,27,28,29,30],"agent-skills","ai-agents","ai-skills","ai-workflows","development-guidelines","prompt-engineering","prompt-workflow","workflow","2026-06-11 04:07:22","\u003C!--\nFunction Name: README\nDescription: Public root README for the agent-workflows library.\n-->\n\n# Agent Workflows\n\nLanguage: **English** | [简体中文](zh-cn\u002FREADME.md)\n\nReusable engineering workflows for AI coding agents.\n\n`agent-workflows` helps agents choose the right process for project initialization, feature work, bug fixes, code review, incident response, refactoring, and tech debt cleanup. The library separates workflow-specific guidance from shared safety, preflight, and validation conventions so the docs stay reusable and easier to maintain, while adding checkpoints that help absorb silent base-model quality drift.\n\n\u003Cp align=\"center\">\n    \u003Cimg width=\"400\" height=\"400\" alt=\"EC6BFAEE-1138-4B0E-B6AB-23A17A72F38A\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Fb243f0ae-4758-48c4-bdaa-7f133a9cfd5b\" \u002F>\n\u003C\u002Fp>\n\n\nNew to the library? Start with [how-to-use-agent-workflows.md](how-to-use-agent-workflows.md).\n\n## Quick Start\n\nChoose one workflow and follow it directly:\n\n- New project or greenfield codebase: [project-initialization-agent-workflow.md](project-initialization-agent-workflow.md)\n- New feature or product behavior change: [feature-development-agent-workflow.md](feature-development-agent-workflow.md)\n- Existing behavior is broken: [bug-fix-agent-workflow.md](bug-fix-agent-workflow.md)\n- Review a PR, branch, or diff: [code-review-agent-workflow.md](code-review-agent-workflow.md)\n- Production incident or post-incident debugging: [incident-debugging-agent-workflow.md](incident-debugging-agent-workflow.md)\n- Behavior-preserving structural improvement: [refactoring-agent-workflow.md](refactoring-agent-workflow.md)\n- Cleanup, dependency upgrades, or debt survey: [tech-debt-cleanup-agent-workflow.md](tech-debt-cleanup-agent-workflow.md)\n\nManual usage example:\n\n```text\nUse the bug-fix workflow in bug-fix-agent-workflow.md for this issue:\n\n\u003Cbug report>\n```\n\nAutomation usage example:\n\n```text\nUse $workflow-automation to select the right workflow and execute it for this task:\n\n\u003Ctask description>\n```\n\n## Why Use Workflows Instead of a Single Prompt?\n\nThe difference is usually not raw model capability. It is process discipline: workflows force triage, validation, and handoff steps that ad hoc prompting often skips.\n\nWorkflows also help absorb silent base-model quality drift. If the underlying model becomes less careful, less reliable, or less consistent without an obvious product change, the workflow still adds checkpoints that reduce the chance of a major hidden drop in output quality.\n\nThe chart below comes from a 6-task exploratory Claude Code study. The full statistical protocol (blinded scoring, bootstrap confidence intervals, paired permutation tests with Holm–Bonferroni correction, inter-rater agreement) is defined in [evaluation\u002FREADME.md](evaluation\u002FREADME.md).\n\n![Measured Outcome Comparison](assets\u002Festimated-outcome-comparison.svg)\n\nScoring note: higher is better for every score.\n\nMetric key:\n\n- `TPR`: task pass rate — average task-level pass rate across repeated runs.\n- `RP@k`: reliable pass at k — share of tasks where every repeated run passes.\n- `CPR`: clean pass rate — share of runs that pass, validate, introduce no regression, and need no rework.\n- `RFR`: regression-free rate — share of runs with no unrelated regression and, when declared, passing locked evaluator checks.\n- `NRR`: no-rework rate — share of runs that need no repair pass.\n\nInterpretation:\n\n- `Task Pass Rate` improves because workflows help execution discipline and validation, not because they change the underlying model's raw capability.\n- `Reliable Pass@k` improves more noticeably because workflows reduce variance by making the agent follow a stable sequence of triage, implementation, and validation steps.\n- `Clean Pass Rate` is the strictest headline quality metric because a run must pass acceptance criteria, pass validation, avoid unrelated regressions, and need no repair pass.\n- `Regression-Free Rate` and `No-Rework Rate` improve because workflows reduce mistakes by enforcing baseline capture, hidden regression\u002Fevaluator checks, and post-change revalidation.\n- Workflows are also more resilient to silent base-model regressions, because process checkpoints catch quality drops that a one-shot prompt may otherwise let through unchecked.\n\n## Available Workflows\n\n- [project-initialization-agent-workflow.md](project-initialization-agent-workflow.md): Bootstrap a new project from requirements through scaffolding, validation, and handoff.\n- [feature-development-agent-workflow.md](feature-development-agent-workflow.md): Design, implement, review, and hand off medium-to-large feature work.\n- [bug-fix-agent-workflow.md](bug-fix-agent-workflow.md): Reproduce, diagnose, fix, and validate a bug.\n- [code-review-agent-workflow.md](code-review-agent-workflow.md): Review code changes with structured findings and optional post-fix re-review.\n- [incident-debugging-agent-workflow.md](incident-debugging-agent-workflow.md): Mitigate production impact first, then diagnose root cause and track follow-up work.\n- [refactoring-agent-workflow.md](refactoring-agent-workflow.md): Improve structure without changing behavior, with baseline and revalidation steps.\n- [tech-debt-cleanup-agent-workflow.md](tech-debt-cleanup-agent-workflow.md): Survey, scope, and execute cleanup work incrementally.\n\n## Shared Building Blocks\n\n- [shared\u002Frepository-preflight.md](shared\u002Frepository-preflight.md): Repository-aware preflight prompts for coding, review, and incident workflows.\n- [shared\u002Fsafety-rules.md](shared\u002Fsafety-rules.md): Reusable safety-rule blocks for different workflow types.\n- [shared\u002Fworkflow-conventions.md](shared\u002Fworkflow-conventions.md): Shared conventions for scope control, escalation, baselines, validation, and reporting.\n\n## Bundled Skills\n\nThis repository includes Codex skills for using and maintaining the workflow library:\n\n- [skills\u002Fworkflow-automation\u002F](skills\u002Fworkflow-automation\u002F): Routes tasks to the correct workflow and loads the minimum required files.\n- [skills\u002Fproject-initialization\u002F](skills\u002Fproject-initialization\u002F): Bootstraps new projects and greenfield repositories using the project initialization workflow.\n- [skills\u002Fworkflow-maintainer\u002F](skills\u002Fworkflow-maintainer\u002F): Audits workflow docs, shared references, skill metadata, links, and README inventory for drift.\n- [skills\u002Frelease-prep\u002F](skills\u002Frelease-prep\u002F): Prepares release readiness reports, validation evidence, and release-note drafts.\n- [skills\u002Fsecurity-review\u002F](skills\u002Fsecurity-review\u002F): Performs focused security reviews for auth, permissions, secrets, injection, data exposure, and dependency risk.\n- [skills\u002Ftest-strategy\u002F](skills\u002Ftest-strategy\u002F): Designs behavior-to-coverage matrices, regression plans, QA steps, and validation command sets.\n- [skills\u002Fmigration-planning\u002F](skills\u002Fmigration-planning\u002F): Plans safe schema, data, API, contract, and rollout migrations.\n- [skills\u002Fperformance-review\u002F](skills\u002Fperformance-review\u002F): Reviews changes for scalability, query, caching, memory, latency, and load risks.\n- [skills\u002Fdocs-maintenance\u002F](skills\u002Fdocs-maintenance\u002F): Maintains documentation structure, examples, links, headings, and cross-file consistency.\n\nShared support files for bundled skills live in [skills\u002F_shared\u002F](skills\u002F_shared\u002F). This is not an installable skill; it contains reusable helper scripts and shared operating rules used by the skill folders.\n\nEach installable skill includes one canonical agent metadata file:\n\n- `agents\u002Finterface.yaml`\n\nTypical setup:\n\n1. Copy the needed folder from `skills\u002F` into your Codex skills directory.\n2. Make sure the skill can find this repository, either by running it from a workspace that contains `agent-workflows\u002F` or by setting `AGENT_WORKFLOWS_ROOT`.\n3. Invoke it with a task such as:\n\n```text\nUse $workflow-automation to route and execute the right workflow for this task:\n\n\u003Ctask description>\n```\n\n## Repository Structure\n\n```text\nagent-workflows\u002F\n|- README.md\n|- how-to-use-agent-workflows.md\n|- project-initialization-agent-workflow.md\n|- feature-development-agent-workflow.md\n|- bug-fix-agent-workflow.md\n|- code-review-agent-workflow.md\n|- incident-debugging-agent-workflow.md\n|- refactoring-agent-workflow.md\n|- tech-debt-cleanup-agent-workflow.md\n|- shared\u002F\n|  |- repository-preflight.md\n|  |- safety-rules.md\n|  |- workflow-conventions.md\n|- skills\u002F\n   |- _shared\u002F\n   |- workflow-automation\u002F\n   |- project-initialization\u002F\n   |- workflow-maintainer\u002F\n   |- release-prep\u002F\n   |- security-review\u002F\n   |- test-strategy\u002F\n   |- migration-planning\u002F\n   |- performance-review\u002F\n   |- docs-maintenance\u002F\n```\n\n## When Not to Use This Library\n\n- **One-line fixes** with no ambiguity (typo, constant, import) — just make the change.\n- **Greenfield project setup without meaningful decisions** — if the project is a single script or throwaway prototype, scaffold it directly. For projects with real tech-stack, structure, or tooling decisions, use the [project initialization workflow](project-initialization-agent-workflow.md).\n- **Infrastructure-as-code or CI\u002FCD implementation changes** — the feature, bug-fix, refactoring, and cleanup workflows are oriented around application code. Code review and incident workflows can still be used to inspect infrastructure-related changes.\n- **Pure documentation changes** (README updates, runbook creation) — the overhead of a full workflow is not justified.\n- **Exploratory prototyping** — if the goal is to experiment and throw away code, skip the process.\n\nIf you are unsure, the triage gates inside each workflow will tell you to use a lighter process when the task is small enough.\n\n## Contributing\n\nIssues and pull requests are welcome.\n\nWhen contributing:\n\n- Keep workflow-specific guidance in the relevant workflow file.\n- Move repeated boilerplate into `shared\u002F` instead of copying it across multiple files.\n- Keep the automation skill aligned with the workflow library when workflow names, paths, or shared conventions change.\n\n## License\n\n[MIT](LICENSE)\n","`agent-workflows` 是一个为AI编码代理或工程师设计的可复用工程工作流库。该项目使用Python编写，提供了多种场景下的标准化流程，包括项目初始化、功能开发、错误修复、代码审查、事故响应、重构和技术债务清理等。其核心在于将具体的工作流程与共享的安全性、预检和验证规范分离，使得文档更加易于维护且具有复用性，并通过设置检查点来减轻基础模型质量无声变化带来的影响。适合于需要提升软件开发过程中一致性和可靠性的团队使用，尤其是在采用AI辅助编程时能够显著提高工作效率和代码质量。",2,"2026-06-11 04:02:47","CREATED_QUERY"]