Agent skill

Auto Devflow

by HuangPuStar in HuangPuStar/FenixAgent

A skill your agent uses when starting an issue, bugfix, feature, or refactor that should be driven by multiple coordinated subagents: explore → plan → code → review, with the main agent acting as…

Apache-2.0Auto-check passedDevelopment

Install Auto Devflow

skills CLI
$ npx skills add HuangPuStar/FenixAgent --skill auto-devflow -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install HuangPuStar/FenixAgent auto-devflow --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/HuangPuStar/FenixAgent.git skills-src && mkdir -p .claude/skills && cp -r skills-src/.claude/skills/auto-devflow .claude/skills/auto-devflow && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
auto-devflow
GitHub stars
552
Token cost
~3k tokens
SKILL.md length
1,139 words
Files
1
Skills in repo
6
Repo updated
First seen
Licence
Apache-2.0

At a glance

A skill your agent uses when starting an issue, bugfix, feature, or refactor that should be driven by multiple coordinated subagents: explore → plan → code → review, with the main agent acting as…

  • Works in 7 steps: Intake → Explore → Plan → …
  • Starting an issue
  • SKILL.md covers Non-Negotiables, Operating Modes, Handoff Directory and Flow, plus 4 more sections
  • Calls git

What it does

Auto Devflow is an agent skill from HuangPuStar/FenixAgent. Use when starting an issue, bugfix, feature, or refactor that should be driven by multiple coordinated subagents: explore → plan → code → review, with the main agent acting as controller and subagents exchanging handoff files.

Its SKILL.md is about 3k tokens, which your agent loads only when the skill is triggered. It is a single SKILL.md file with no bundled scripts.

It sits in Development, covering Subagents, Refactoring and Code review. The licence is Apache-2.0.

When your agent uses it

  • Starting an issue
  • Refactor that should be driven by multiple coordinated subagents: explore → plan → code → review
  • With the main agent acting as controller and subagents exchanging handoff files

Example prompts

  • “/auto-devflow”

Workflow steps

7 steps, taken from the step headings in SKILL.md.

  1. Intake
  2. Explore
  3. Plan
  4. Code
  5. Review
  6. Verify
  7. Report

What it can do on your machine

Read from SKILL.md and the folder at commit da5eb54. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Shell commands in SKILL.md call:

    • git

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md. Its commands use git, which can reach the network depending on how they are called.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Auto Devflow loads about 3k tokens when it runs. Until then it costs about 60 tokens; SKILL.md has 1,139 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~60
When it runs · the whole SKILL.md, loaded when a task matches
~3k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from HuangPuStar/FenixAgent at commit da5eb54, republished under its Apache-2.0 licence (© HuangPuStar). 1,139 words, ~2,963 tokens.

Download SKILL.mdSave it as .claude/skills/auto-devflow/SKILL.md (or your agent's skills folder).
name
auto-devflow
description
Use when starting an issue, bugfix, feature, or refactor that should be driven by multiple coordinated subagents: explore → plan → code → review, with the main agent acting as controller and subagents exchanging handoff files.

Start Devflow

Run a controlled devflow: the main agent stays the controller, each subagent owns one phase, and phase outputs are written to temporary handoff files before the next phase starts.

Use this when the user says things like "开始解决这个 issue", "start this issue", "fix this bug end-to-end", "实现这个功能", or asks for a multi-agent workflow rather than a single-agent edit.

Non-Negotiables

  • The main agent owns scope, approvals, TodoWrite, git safety, and final reporting.
  • Subagents do focused work only. Do not let a subagent silently broaden scope.
  • Use handoff files as the shared memory. Do not rely on hidden subagent context.
  • Run writable phases sequentially. Never run two coding subagents against the same checkout at the same time.
  • Ask the user before destructive operations, branch changes, commits, or scope expansion.

Operating Modes

Pick a mode during Intake and record it in 00-context.md.

Relaxed Mode

Default for simple, low-risk tasks.

Use relaxed mode when:

  • the request is small and localized
  • the likely change touches a few files with obvious ownership
  • failure impact is low and rollback is straightforward
  • the agent has not already made repeated mistakes on this task

Relaxed flow:

text
Intake → Explore → Plan → Code → Review → Verify → Report

Plan review is optional. The main agent may self-check 02-plan.md instead of dispatching a separate reviewer.

Cautious Mode

Use cautious mode when the user says the task is hard, risky, subtle, or when the agent has already made too many mistakes. Also use it for public API changes, data/schema changes, security-sensitive code, concurrency, async lifecycle, rendering/event-loop bugs, or broad refactors.

Cautious flow:

text
Intake → Explore → Plan → Plan Review → User/Controller Gate → Code → Review → Verify → Report

Cautious mode adds a plan review before coding. Dispatch a read-only reviewer (plan or code-reviewer) to critique 02-plan.md and write 02-plan-review.md. Plan Review is a mandatory gate in cautious mode——coding must not start until every critical issue is fixed or explicitly accepted by the main agent/user. Historical evidence: Plan Review intercepted 9+ blocking issues across 3 major refactors (StageContext、ProviderAdapter、compact_v2) in 2026-07-16 alone.

Completion criterion: 00-context.md states Mode: relaxed or Mode: cautious, why that mode was chosen, and which gates are required.

Handoff Directory

Create one directory per devflow:

text
.peri/plans/<title>/
  00-context.md
  01-explore.md
  02-plan.md
  02-plan-review.md   # cautious mode only
  03-code.md
  04-review.md
  05-verification.md

<title> 用任务标题的 kebab-case 形式(如 fix-auth-timeout、add-pagination),有 issue id 时附加 id。

Each handoff file must contain:

markdown
# <Phase> Handoff

## Goal
<one sentence>

## Scope
<in scope / out of scope>

## Findings or Changes
<phase-specific details>

## Files
<relevant file paths>

## Commands
<commands run and outcomes, or "not run">

## Open Questions
<none, or exact blockers>

## Next Prompt
<self-contained prompt for the next subagent>

Completion criterion: a later subagent can start from the previous handoff file plus the original user request without guessing.

Flow

1. Intake

Clarify only if needed. If the user gave an issue id, URL, stack trace, or spec, capture it exactly in 00-context.md.

Write 00-context.md with:

  • original request
  • success criteria
  • operating mode: relaxed or cautious, with reason
  • constraints from the user and project memory
  • current git branch/status summary
  • explicit approval state for commits or branch changes

Stop and ask if success criteria or target issue is ambiguous.

2. Explore

Dispatch an explore subagent in read-only mode.

Prompt it to:

  • inspect relevant files and tests
  • identify execution paths and ownership boundaries
  • find existing conventions
  • write 01-explore.md
  • not edit code

Completion criterion: 01-explore.md names likely files to touch, relevant tests, known traps, and at least one plausible implementation seam.

3. Plan

Dispatch a plan subagent in read-only mode, using 00-context.md and 01-explore.md.

Prompt it to:

  • produce a small implementation plan
  • identify risks and rollback points
  • define verification commands
  • write 02-plan.md
  • not edit code

Completion criterion: 02-plan.md has ordered steps, exact files, verification commands, and a clear stop/ask condition.

3.5. Plan Review

In relaxed mode, the main agent self-checks the plan for missing files, missing verification, scope creep, and unsafe operations. Write the self-check result into 02-plan.md or 00-context.md.

In cautious mode, dispatch a read-only plan reviewer before coding.

Prompt it to:

  • read 00-context.md, 01-explore.md, and 02-plan.md
  • challenge assumptions and hidden risks
  • check that planned files and tests match the exploration findings
  • identify scope creep, missing rollback, missing verification, and unsafe steps
  • write 02-plan-review.md
  • not edit code

Completion criterion: relaxed mode has a recorded self-check; cautious mode has 02-plan-review.md with APPROVED or a concrete issue list. If the reviewer finds critical issues, update 02-plan.md and repeat plan review before coding.

If the plan changes user-visible behavior, public APIs, data format, migrations, or architecture boundaries, present the plan summary to the user and wait for approval before coding.

Show full SKILL.md (462 more words)Show less
4. Code

Dispatch exactly one coder subagent for the current coding slice, using 00-context.md, 01-explore.md, and 02-plan.md.

Prompt it to:

  • implement only the approved slice
  • follow existing style
  • run targeted checks if practical
  • write 03-code.md
  • report any blockers instead of guessing

Completion criterion: 03-code.md lists changed files, commands run, residual risks, and whether the plan was followed exactly.

If the plan is large, split it into sequential slices. After each slice, review before starting the next.

5. Review

Dispatch a code-reviewer subagent in read-only mode, using all handoff files plus the diff.

Prompt it to check:

  • spec compliance: did the code solve the stated task and avoid extras?
  • maintainability: style, tests, error handling, security, regressions
  • handoff integrity: did 03-code.md match actual diff?
  • write 04-review.md

Completion criterion: review returns either APPROVED or an issue list with severity and exact file references.

If review finds issues, send a focused fix prompt to coder, update 03-code.md, then re-review. Repeat until approved or blocked.

6. Verify

The main agent runs or delegates final verification based on risk:

  • small change: main agent runs targeted tests/lint/build discovered in 02-plan.md
  • non-trivial change: dispatch verification subagent with original task, changed files, and approach

Write 05-verification.md with commands, outputs, and verdict.

Completion criterion: every planned verification is passed, explicitly skipped with reason, or reported as blocked.

7. Report

Reply with:

  • outcome
  • files changed
  • verification evidence
  • unresolved risks or follow-up work
  • whether commits were created (only if user explicitly requested commits)

Do not claim completion without 05-verification.md evidence.

Controller Checklist

Use TodoWrite for the full devflow:

  1. Intake / context file
  2. Explore handoff
  3. Plan handoff
  4. Plan review gate (self-check in relaxed mode, reviewer in cautious mode)
  5. Code handoff
  6. Worktree safety gate: git diff --stat in both main repo and worktree; verify edits in correct location
  7. Review loop
  8. Verification handoff
  9. Final report

Keep the main context small: read handoff summaries, not every file, unless you need to adjudicate a blocker.

Subagent Prompt Skeletons

Explorer
text
Goal: explore the codebase for <task>. Read .peri/plans/<title>/00-context.md first.

Write .peri/plans/<title>/01-explore.md using the required handoff template.
Do not edit files. Focus on relevant paths, conventions, tests, and risks.
Completion: the next planner can create an implementation plan without additional search.
Planner
text
Goal: plan <task>. Read 00-context.md and 01-explore.md first.

Write .peri/plans/<title>/02-plan.md using the required handoff template.
Do not edit files. Include ordered steps, exact files, verification commands, risks, and stop/ask conditions.
Completion: a coder can implement from the plan without guessing.
Plan Reviewer
text
Goal: review the plan for <task>. Read 00-context.md, 01-explore.md, and 02-plan.md first.

Write .peri/plans/<title>/02-plan-review.md. Return APPROVED or list issues with severity, file references, and fix guidance.
Challenge assumptions, hidden risks, missing verification, rollback gaps, unsafe operations, and scope creep.
Do not edit files.
Coder
text
Goal: implement the approved plan for <task>. Read 00-context.md, 01-explore.md, and 02-plan.md first.

CRITICAL — Worktree safety:
- ALL file paths in edits MUST use absolute paths. Append cwd prefix to every relative path.
- Before any edit, READ the file to confirm it exists at the absolute worktree path.
- After all edits, run: cd <absolute_worktree_path> && git diff --stat HEAD
- If you edited a file NOT in the worktree, STOP and report immediately.

CRITICAL — Scope boundary:
- Only modify files explicitly listed in the plan. DO NOT touch adjacent files.
- If the plan says change Cargo.toml versions and import paths, DO NOT refactor other code.
- If you see a related improvement not in the plan, record it in 03-code.md Open Questions — DO NOT implement it.

Edit only files required by the plan. Follow repository style. Run targeted checks if practical.
Write .peri/plans/<title>/03-code.md using the required handoff template.
If blocked or the plan is wrong, stop and report instead of broadening scope.
Reviewer
text
Goal: review the implementation for <task>. Read all handoff files and inspect the current diff.

Write .peri/plans/<title>/04-review.md. Return APPROVED or list issues with severity, file references, and fix guidance.
Check spec compliance first, then code quality. Do not edit files.
Verifier
text
Goal: verify <task> is complete. Read all handoff files and inspect changed files.

Run appropriate checks or explain why they cannot run. Write .peri/plans/<title>/05-verification.md with evidence and verdict.
Do not edit files.

Red Flags

Stop and ask the user when:

  • the issue/spec is missing or contradictory
  • the chosen operating mode is unclear after intake
  • cautious-mode plan review finds unresolved critical issues
  • the plan requires deleting data, changing schemas, or altering public behavior not requested
  • the repo has unrelated dirty changes that overlap planned files
  • a subagent reports BLOCKED, NEEDS_USER_DECISION, or an unsafe workaround
  • worktree safety gate fails: main repo has unexpected diffs after coder phase
  • verification fails twice for the same reason

Relationship to Other Skills

  • Use diagnose or systematic-debugging inside the explore phase for hard bugs.
  • Use writing-plans when the output should be a standalone long-form implementation plan.
  • Use subagent-driven-development when you already have a detailed plan and just need task-by-task execution.
  • Use verification-before-completion before claiming success.

© HuangPuStar, Apache-2.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

Just SKILL.md in .claude/skills/auto-devflow of HuangPuStar/FenixAgent.

Open the folder on GitHubat commit da5eb54

Compare with similar skills

Auto Devflow next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Auto Devflow compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Auto Devflow this skillHuangPuStar/FenixAgent552—~3kAutomated safety check: PassApache-2.0
Codex CLIsundial-org/awesome-openclaw-skills663—~2kAutomated safety check: PassNone
Review And Simplify ChangesDimillian/Skills4k—~2kAutomated safety check: PassMIT
Local Codexdaymade/claude-code-skills1.4k—~1.6kAutomated safety check: PassMIT
GitHub Review Iterationprisma/orm48k—~2.2kAutomated safety check: PassApache-2.0
Cherry Studio PR ReviewCherryHQ/cherry-studio52k—~3.9kAutomated safety check: PassAGPL-3.0

Similar skills

  • Codex CLI

    sundial-org/awesome-openclaw-skills

    Use OpenAI Codex CLI for coding tasks. An agent skill from sundial-org/awesome-openclaw-skills.

    663 GitHub stars~2k tokensUpdated 7 mo ago
    DevelopmentAuto-check passed
  • Review a git diff or explicit file scope for reuse, code quality, efficiency, clarity, and standards issues, then optionally apply safe Codex-driven fixes.

    4k GitHub stars~2k tokensUpdated 6 mo ago
    DevelopmentAuto-check passed
  • Local Codex

    daymade/claude-code-skills

    Launches and manages the local OpenAI Codex CLI as a non-interactive coding sub-agent via ChatGPT Pro OAuth, never API keys.

    1.4k GitHub stars~1.6k tokensUpdated today
    Backend & APIsAuto-check passed
  • Official

    Runs a loop on a GitHub pull request: fetch review state, triage comments into actions, implement them and resolve threads, repeating until nothing actionable is left.

    48k GitHub stars~2.2k tokensUpdated today
    DevelopmentAuto-check passed
  • Cherry Studio PR Review

    CherryHQ/cherry-studio

    Reviews Cherry Studio branches, pull requests, commits, files and docs against the project's own architecture, naming, API-boundary and UI rules, report-only by default.

    52k GitHub stars~3.9k tokensUpdated today
    DevelopmentAuto-check passed
  • PR Review

    jaemk/self_update

    Targeted, read-only review of a PR or checked-out branch. An agent skill from jaemk/self_update.

    961 GitHub stars~1.5k tokensUpdated 1 mo ago
    DevelopmentAuto-check: notes

More from HuangPuStar/FenixAgent

  • Zentao Manager

    HuangPuStar/FenixAgent

    通过 zentao 命令行工具查询和操作禅道(ZenTao)数据,覆盖项目集、产品、项目、执行、需求、Bug、任务、测试用例、测试单、产品计划、版本、发布、反馈、工单、应用、用户、附件等模块的增删改查及状态流转。当用户提到禅道、zentao、查询项目进展、获取 Bug 列表、创建任务、更新需求状态等项目管理操作时使用本技能。

    552 GitHub starsUsed in 1 repo~2.5k tokens
    Auto-check passed
  • Agent Platform API

    HuangPuStar/FenixAgent

    RCS Platform API 完整参考。Agent 通过 curl + jq 调用 REST API 操作平台资源:环境、工作流、配置、任务、知识库、组织等。

    552 GitHub stars~565 tokensUpdated 7 days ago
    Auto-check: notes
  • Fenix Code Review

    HuangPuStar/FenixAgent

    执行 FenixAgent 项目专用代码审查。手动触发,默认审查已暂存改动,或按用户提供的文件路径、提交范围、分支 diff 范围进行审查,并生成审查报告。

    552 GitHub stars~1.7k tokensUpdated 7 days ago
    Auto-check passed
  • Show HTML Or Picture

    HuangPuStar/FenixAgent

    当需要把本地 HTML 或图片文件展示给前端时使用。优先引用当前工作目录下的 user/ 目录中的文件;如果文件在其他位置,先复制到当前工作目录的 user/ 再展示。路径由前端自动改写为 /fs/ 路由代理,Agent 只需输出 user/ 相对路径即可。最终只输出可直接渲染的相对路径片段:HTML 输出 iframe,图片输出 Markdown 图片;路径不能是绝对路径,也不能以 /…

    552 GitHub stars~475 tokensUpdated 7 days ago
    Auto-check passed
  • Skill Creator

    HuangPuStar/FenixAgent

    帮助用户创建、编辑和优化 Agent Skill 指令文件。当用户提到"创建 skill"、"写技能"、"编辑 skill"、"优化技能"、"新建技能"或需要生成 SKILL.md 文件时使用。

    552 GitHub stars~469 tokensUpdated 7 days ago
    Auto-check passed

Questions about Auto Devflow

What does Auto Devflow do?

A skill your agent uses when starting an issue, bugfix, feature, or refactor that should be driven by multiple coordinated subagents: explore → plan → code → review, with the main agent acting as…. Auto Devflow is an agent skill from HuangPuStar/FenixAgent. Use when starting an issue, bugfix, feature, or refactor that should be driven by multiple coordinated subagents: explore → plan → code → review, with the main agent acting as controller and subagents exchanging handoff files.

When should I use Auto Devflow?

Auto Devflow fits situations like: starting an issue; refactor that should be driven by multiple coordinated subagents: explore → plan → code → review; with the main agent acting as controller and subagents exchanging handoff files.

How do I install Auto Devflow in Claude Code?

Run `npx skills add HuangPuStar/FenixAgent --skill auto-devflow -a claude-code`. Or copy the skill folder (.claude/skills/auto-devflow in HuangPuStar/FenixAgent) into .claude/skills/auto-devflow in your project. Claude Code loads it when a task matches its description.

How do I install Auto Devflow in Codex?

Run `npx skills add HuangPuStar/FenixAgent --skill auto-devflow -a codex`. Or copy the skill folder (.claude/skills/auto-devflow in HuangPuStar/FenixAgent) into .agents/skills/auto-devflow in your project. Codex loads it when a task matches its description.

Can I use Auto Devflow in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add HuangPuStar/FenixAgent --skill auto-devflow -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/auto-devflow, .gemini/skills/auto-devflow, .github/skills/auto-devflow and .opencode/skills/auto-devflow in your project.

What does Auto Devflow need to run?

Going by SKILL.md and its folder, Auto Devflow needs the command-line tools its instructions call (git).

Does Auto Devflow access the network?

SKILL.md contains no URLs. Its commands use git, which can reach the network depending on how they are called. This is read from the text; nothing was executed.

Is Auto Devflow safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Auto Devflow use?

Auto Devflow is published under the Apache-2.0 licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Auto Devflow use?

About 3k tokens (SKILL.md is roughly 12k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Auto Devflow?

Skills that share tags, products or a category with Auto Devflow: Codex CLI (sundial-org/awesome-openclaw-skills, 663 stars), Review And Simplify Changes (Dimillian/Skills, 4k stars), Local Codex (daymade/claude-code-skills, 1.4k stars) and GitHub Review Iteration (prisma/orm, 48k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Auto Devflow?

HuangPuStar (a GitHub organization) maintains it in HuangPuStar/FenixAgent, which has 552 GitHub stars. The repository holds 6 skills in this directory. The repository was last updated on September 30, 2026.

Source: HuangPuStar/FenixAgent on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.