New: 98,000+ skills indexed from GitHub repos with 100+ stars · every skill safety-checkedBrowse top skillsTop skills
Esc to closeBrowse all skills
Submit a skillBrowse skills
Submit a skillBrowse skills

Agent skill

Junshi

by aruis in aruis/codex-cookbook

Use only when the user explicitly enters 军师 mode, such as saying “召唤军师”, “军师跟进”, or “让军师看一下”, and keep using it for later turns in the same thread until the user explicitly exits 军师 mode or 军师…

MITAuto-check passed

Install Junshi

skills CLI
$ npx skills add aruis/codex-cookbook --skill junshi -a claude-code

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

GitHub CLI
$ gh skill install aruis/codex-cookbook junshi --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/aruis/codex-cookbook.git skills-src && mkdir -p .claude/skills && cp -r skills-src/skills/junshi .claude/skills/junshi && 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
junshi
GitHub stars
165
Token cost
~1.3k tokens
SKILL.md length
669 words
Files
6 (incl. references)
Skills in repo
4
Repo updated
First seen
Licence
MIT

At a glance

Use only when the user explicitly enters 军师 mode, such as saying “召唤军师”, “军师跟进”, or “让军师看一下”, and keep using it for later turns in the same thread until the user explicitly exits 军师 mode or 军师…

  • Works in 2 steps: Is this actually a system-level or… → If it is broader than a local task, is…
  • Explicitly enters 军师 mode
  • SKILL.md covers Activation, Core Judgment, Delegation Authorization and Operating Model, plus 5 more sections
  • Calls git

What it does

Junshi is an agent skill from aruis/codex-cookbook. Use only when the user explicitly enters 军师 mode, such as saying “召唤军师”, “军师跟进”, or “让军师看一下”, and keep using it for later turns in the same thread until the user explicitly exits 军师 mode or 军师 explicitly falls back to direct handling.

Its SKILL.md is about 1.3k tokens, which your agent loads only when the skill is triggered. The skill folder holds 7 other files, including reference files (for example `agents/openai.yaml`, `references/artifact-templates.md` and `references/decision-rules.md`).

The repository describes itself as: Codex prompt cookbook and collaboration templates. The licence is MIT.

When your agent uses it

  • Explicitly enters 军师 mode
  • Such as saying 召唤军师
  • Keep using it for later turns in the same thread until the user explicitly exits 军师 mode
  • 军师 explicitly falls back to direct handling

Example prompts

  • “让军师看一下”
  • “/junshi”

Workflow steps

2 steps, taken from the first numbered list in SKILL.md.

  1. Is this actually a system-level or cross-module problem, or just a local task?
  2. If it is broader than a local task, is it worth pursuing with mid/long-term design?

What it can do on your machine

Read from SKILL.md and the folder at commit 0d1e8b3. 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

Junshi loads about 1.3k tokens when it runs, and up to ~3.7k if it reads all its reference files. Until then it costs about 60 tokens; SKILL.md has 669 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
~1.3k
With references · SKILL.md plus every file in references/, read only if the agent opens them
~3.7k

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 aruis/codex-cookbook at commit 0d1e8b3, republished under its MIT licence (© aruis). 669 words, ~1,289 tokens.

Download SKILL.mdSave it as .claude/skills/junshi/SKILL.md (or your agent's skills folder). This skill also uses 5 other files; get the full folder from GitHub.
name
junshi
description
Use only when the user explicitly enters 军师 mode, such as saying “召唤军师”, “军师跟进”, or “让军师看一下”, and keep using it for later turns in the same thread until the user explicitly exits 军师 mode or 军师 explicitly falls back to direct handling.

军师

Use this skill for medium/large software project threads where a lead agent should decide whether to analyze, execute, coordinate, or converge.

This is a lead-agent skill, not a ceremony generator. Keep the thread lightweight unless the work really needs orchestration.

Activation

军师 mode requires explicit user activation.

Enter only when the user clearly says phrases such as:

  • 召唤军师
  • 军师跟进
  • 让军师看一下
  • or otherwise explicitly asks to enter 军师 / lead-agent mode

Do not infer or auto-enter 军师 mode from task complexity alone.

If the user only says 召唤军师 without a concrete task:

  • switch into 军师 mode
  • reply briefly
  • ask for the task directly

Once explicitly activated in the current thread, keep using 军师 mode until:

  • the user explicitly exits it, or
  • 军师 explicitly falls back to direct handling

Core Judgment

Before expanding work, 军师 should answer two questions:

  1. Is this actually a system-level or cross-module problem, or just a local task?
  2. If it is broader than a local task, is it worth pursuing with mid/long-term design?

Only then decide whether to:

  • handle it directly
  • send one 谋士
  • send one 武将
  • bring in 法正
  • coordinate multiple people

Default preference:

  • local or small tasks: direct handling or one 武将
  • high-risk judgment: one 谋士
  • review, test closure, submit-readiness, or commit-side handling: 法正
  • multi-role coordination only when one-person delegation is no longer enough

Do not turn routine work into multi-agent process just because 军师 mode is active.

Delegation Authorization

Once 军师 is active, assume 军师 may delegate.

If the environment still requires explicit authorization for subagents, ask one short confirmation question, then proceed once the user says yes.

Treat that authorization as valid for the rest of the current thread unless the user narrows or revokes it.

Operating Model

There is one lead role, two rosters, and one single-seat convergence role:

  • 军师: lead role
  • 谋士池: 贾诩 / 庞统 / 郭嘉
  • 武将池: 关羽 / 赵云 / 马超
  • 法正: single-seat review, test, and commit-handling role

Rules:

  • select by pool unless the user explicitly names an identity
  • if the same identity is repeatedly assigned in one continuous task, prefer reusing it when continuity helps
  • names indicate identity only, not fixed specialties
  • 军师 remains the single owner of direction, escalation, convergence, and commit judgment

Default skill mapping:

  • 谋士池 -> $junshi-moushi
  • 武将池 -> $junshi-wujiang
  • 法正 -> $junshi-fazheng

When delegating, keep the assignment short: task, scope boundary, expected output.

Show full SKILL.md (286 more words)Show less

Working Style

军师 should think, not pattern-match.

Use a compact internal frame:

  • 目标: what problem is being solved
  • 模式: diagnose first, execute first, reproduce first, or plan first
  • 约束: scope, non-goals, acceptable tradeoffs
  • 输出: decision, implementation, review, or convergence
  • 验证: minimal proof that the work actually holds

Do not recite the full frame unless alignment is actually missing.

Guardrails

When driving a thread:

  • solve the current core problem first
  • read context before acting
  • separate fact from inference when the distinction matters
  • ask only the 1 to 3 most decision-relevant questions when user input is required
  • prefer explicit verification over verbal confidence
  • avoid busy-waiting on delegated work when the next useful move is to report status or wait for user direction
  • default to one appropriate delegated role first; parallelize only with stable non-conflicting boundaries
  • for review-like requests, prefer delegated review or validation unless the task is trivial or the user explicitly wants a personal review

Commit Judgment

军师 owns the judgment of whether the work is ready to converge toward a commit point.

That does not mean perfect completion, and it does not automatically mean git commit.

Care most about:

  • whether execution stayed within the approved boundary
  • whether known leftovers are explicit
  • whether higher-risk work has gone through an appropriate review or validation pass

Non-Goals

Do not use this skill to:

  • force small tasks into orchestration
  • over-specify implementation details too early
  • parallelize without stable boundaries
  • let documents or process overwhelm the task itself
  • keep abstracting once the reuse value is unclear

Reference Guide

Use companion docs only when needed:

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

Files

SKILL.md and 5 other files (references) in skills/junshi of aruis/codex-cookbook.

  • SKILL.md
  • agents/openai.yaml
  • references/artifact-templates.md
  • references/decision-rules.md
  • references/dispatch-prompts.md
  • references/working-shape.md

Open the folder on GitHubat commit 0d1e8b3

Compare with similar skills

Junshi 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.

Junshi compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Junshi this skillaruis/codex-cookbook165—~1.3kAutomated safety check: PassMIT
No Explicit Anythedaviddias/Front-End-Checklist74k—~565Automated safety check: PassMIT
Explicit Identityparcadei/Continuous-Claude-v33.9k1 repos~323Automated safety check: PassMIT
Explicit Checkerbitwize-music-studio/claude-ai-music-skills544—~1.3kAutomated safety check: PassCC0-1.0
Explicit Instruction Sequence BuilderGarethManning/education-agent-skills840—~5.2kAutomated safety check: PassCustom licence
Research Junshijunshi-research/research-junshi126—~2.1kAutomated safety check: PassApache-2.0

Similar skills

  • No Explicit Any

    thedaviddias/Front-End-Checklist

    A skill your agent uses when reviewing TypeScript files for type safety regressions, during code review of functions that handle external data, or when the codebase has ESLint warnings for…

    74k GitHub stars~565 tokensUpdated 4 days ago
    DevelopmentAuto-check passed

  • Explicit Identity

    parcadei/Continuous-Claude-v3

    Explicit Identity Across Boundaries

    3.9k GitHub starsUsed in 1 repo~323 tokens
    Auto-check passed

  • Explicit Checker

    bitwize-music-studio/claude-ai-music-skills

    Scans lyrics for explicit content and verifies that explicit flags match actual content.

    544 GitHub stars~1.3k tokensUpdated 3 days ago
    Auto-check passed

  • Explicit Instruction Sequence Builder

    GarethManning/education-agent-skills

    Build a complete explicit instruction sequence from teacher modelling through guided practice to independent work.

    840 GitHub stars~5.2k tokensUpdated 1 mo ago
    EducationAuto-check passed

  • Research Junshi

    junshi-research/research-junshi

    Personalized academic literature digests and ranked research ideas grounded in the researcher's papers and persistent memory.

    126 GitHub stars~2.1k tokensUpdated 24 days ago
    Agent WorkflowsAuto-check passed

  • Masm Explicit Stack Interface

    0xMiden/protocol

    A skill your agent uses when defining the interface for a new MASM procedure — keep its inputs explicit on the stack and its outputs limited to values it produces.

    133 GitHub stars~716 tokensUpdated yesterday
    DevelopmentAuto-check passed

More from aruis/codex-cookbook

  • Junshi Fazheng

    aruis/codex-cookbook

    Use only when junshi is already active in the current thread and 法正 has been explicitly delegated by junshi; never trigger from a direct end-user request alone.

    165 GitHub stars~623 tokensUpdated 5 mo ago
    Auto-check passed

  • Junshi Moushi

    aruis/codex-cookbook

    Use only when junshi is already active in the current thread and this role has been explicitly delegated by junshi; never trigger from a direct end-user request alone.

    165 GitHub stars~469 tokensUpdated 5 mo ago
    Auto-check passed

  • Junshi Wujiang

    aruis/codex-cookbook

    Use only when junshi is already active in the current thread and this role has been explicitly delegated by junshi; never trigger from a direct end-user request alone.

    165 GitHub stars~422 tokensUpdated 5 mo ago
    Auto-check passed

Questions about Junshi

What does Junshi do?

Use only when the user explicitly enters 军师 mode, such as saying “召唤军师”, “军师跟进”, or “让军师看一下”, and keep using it for later turns in the same thread until the user explicitly exits 军师 mode or 军师…. Junshi is an agent skill from aruis/codex-cookbook. Use only when the user explicitly enters 军师 mode, such as saying “召唤军师”, “军师跟进”, or “让军师看一下”, and keep using it for later turns in the same thread until the user explicitly exits 军师 mode or 军师 explicitly falls back to direct handling.

When should I use Junshi?

Junshi fits situations like: explicitly enters 军师 mode; such as saying 召唤军师; keep using it for later turns in the same thread until the user explicitly exits 军师 mode; 军师 explicitly falls back to direct handling.

How do I install Junshi in Claude Code?

Run `npx skills add aruis/codex-cookbook --skill junshi -a claude-code`. Or copy the skill folder (skills/junshi in aruis/codex-cookbook) into .claude/skills/junshi in your project. Claude Code loads it when a task matches its description.

How do I install Junshi in Codex?

Run `npx skills add aruis/codex-cookbook --skill junshi -a codex`. Or copy the skill folder (skills/junshi in aruis/codex-cookbook) into .agents/skills/junshi in your project. Codex loads it when a task matches its description.

Can I use Junshi 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 aruis/codex-cookbook --skill junshi -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/junshi, .gemini/skills/junshi, .github/skills/junshi and .opencode/skills/junshi in your project.

What does Junshi need to run?

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

Does Junshi 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 Junshi 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 Junshi use?

Junshi is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Junshi use?

About 1.3k tokens (SKILL.md is roughly 5.2k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full. Its references folder adds about 2.4k tokens, read only when the agent opens those files.

What are the alternatives to Junshi?

Skills that share tags, products or a category with Junshi: No Explicit Any (thedaviddias/Front-End-Checklist, 74k stars), Explicit Identity (parcadei/Continuous-Claude-v3, 3.9k stars), Explicit Checker (bitwize-music-studio/claude-ai-music-skills, 544 stars) and Explicit Instruction Sequence Builder (GarethManning/education-agent-skills, 840 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Junshi?

aruis (a GitHub user) maintains it in aruis/codex-cookbook, which has 165 GitHub stars. The repository holds 4 skills in this directory. The repository was last updated on April 25, 2026.

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