update-claude

Diagnose the existing

.claude/

system, present the differences from the ideal state (Japanese operation, category-based Agents, Rules maintenance, delegation rules, model allocation, SessionStart hooks, implement-issue-tree prerequisites), and only supplement the differences after user approval. Confirm before overwriting existing Agents, Rules, and CLAUDE.md.

Use

init-claude

for repositories where

.claude/

does not exist.

Usage

update-claude [path-to-target-repository]

If the path is omitted, the current directory is targeted.

Prerequisites

The
```
.claude/
```
directory exists in the target repository
The
```
gh
```
CLI is installed and authenticated (check with
```
gh auth status
```
)
```
npx
```
is available (used for skill supplementation via
```
npx skills add
```
)

Flow

Step 1: Investigate the target repository and existing .claude/

bash

# Check for existence of .claude/
ls <target-repo>/.claude/ 2>/dev/null || echo ".claude/ does not exist"

# List existing Agents
find <target-repo>/.claude/agents -name "*.md" 2>/dev/null | sort

# List existing Rules
find <target-repo>/.claude/rules -name "*.md" 2>/dev/null | sort

# List existing Skills
ls <target-repo>/.claude/skills/ 2>/dev/null | sort

# Check hooks
cat <target-repo>/.claude/settings.json 2>/dev/null || echo "No settings.json"

# Check CLAUDE.md
cat <target-repo>/CLAUDE.md 2>/dev/null | head -80 || echo "No CLAUDE.md"

# Check skills-lock
cat <target-repo>/skills-lock.json 2>/dev/null | head -30 || echo "No skills-lock.json"

# Check implement-issue-tree workflow
ls <target-repo>/.claude/workflows/implement-issue-tree.js 2>/dev/null || echo "No workflow js"

.claude/

does not exist, guide the user to use

init-claude

and interrupt processing.

Step 2: Diagnose differences from the ideal state

Diagnose the current state from the following perspectives and create a gap list.

2-1. Agent Diagnosis

Diagnosis Item	Check Content
Category Division	Is there category division into research / implement / testing / quality / docs?
Tech Layer-specific Builders	Are there builder Agents corresponding to the repo's technical layers?
Model Allocation	Is the allocation followed: implementation/research=sonnet / mechanical tasks=haiku / cross-cutting decisions=opus or fable (fable is the top tier above Opus)?
Least Privilege	Is the Agent's `tools` list kept to the minimum necessary?
Delegation Agents	Are there Agents equivalent to skill-author / agent-author / rules-author / docs-writer?

2-2. Rules Diagnosis

Diagnosis Item	Check Content
delegation.md	Does it contain delegation principles for research mode and path-based switching table?
delegation-impl.md	Does it contain delegation mappings for creation/edit mode?
Coding Standards	Are there coding standards corresponding to the repo's main language?
security.md	Does it include OWASP Top 10 and prevention of sensitive information inclusion?
japanese-style.md	Does it describe Japanese output style?
conventional-commits.md	Does it include detailed Conventional Commits rules?
code-comment-style.md	Does it include comment rules (policy for embedding role, invocation context, responsibility boundaries)?
out-of-scope-tracking.md	Does it include rules for tracking out-of-scope items (Issue confirmation → user approval → ticketing flow)?

2-3. Skills Diagnosis

List skills available via

npx skills add Fandhe-AI/agent-cli-skills

that are not included in

skills-lock.json

Mandatory Skills (to be supplemented if missing):

```
create-commit
```
,
```
create-pr
```
,
```
create-issue
```
```
implement-issue
```
,
```
implement-issue-tree
```
```
implement-review
```
,
```
implement-review-pr
```
```
update-docs
```
```
comment-code
```
(skill to add/reinforce comments according to
```
code-comment-style.md
```
rules)

2-4. hooks Diagnosis

Diagnosis Item	Check Content
SessionStart	Does it include reminders for Japanese language use, delegation, Conventional Commits compliance, and prohibition of --no-verify?
PostToolUse	Is there an auto-formatting hook corresponding to the language?

2-5. CLAUDE.md Diagnosis

Diagnosis Item	Check Content
Delegation Policy	Does it include a path-based switching table and model allocation table?
Sub-agents List	Does it include a table of category-based Agents?
Rules List	Does it include a table of rule files?
Current Skills	Does it include a list of installed skills?

2-6. implement-issue-tree Prerequisite Diagnosis

bash

gh auth status
# sub_issues are only valid with issue numbers in `issues/{n}/sub_issues` (there is no sub_issues endpoint directly under the repository). If existing issues exist, specify the number to verify connectivity
# (if no issue has been created, only gh auth status will be used for prerequisite confirmation)
gh api "repos/<owner>/<repo>/issues/<existing-issue-number>/sub_issues" 2>&1 | head -5
ls <target-repo>/.claude/workflows/implement-issue-tree.js 2>/dev/null

gh auth authentication status
sub_issues API response (specify existing issue number. 404 = possibility that GitHub Apps is not enabled. Note: Do not use the repository-level endpoint
```
repos/{owner}/{repo}/sub_issues
```
as it does not exist)
Existence of workflow js

Step 3: Present gap list to user for approval

Present the gap list in the following format.

## .claude/ Diagnosis Results

### Items requiring supplementation/enhancement

#### Agents
- [ ] <category>/<Agent-name>.md: <reason for deficiency>
- ...

#### Rules
- [ ] delegation.md: Does not exist
- [ ] coding-<lang>.md: No standards for main language (<lang>)
- ...

#### Skills (supplementable via npx skills add)
- [ ] <skill-name>: Not included in skills-lock.json
- ...

#### hooks
- [ ] SessionStart: No SessionStart hook in settings.json
- [ ] PostToolUse: No auto-formatting hook for <lang>
- ...

#### CLAUDE.md
- [ ] Delegation Policy Table: No path-based switching table
- ...

#### implement-issue-tree Prerequisites
- [ ] workflow js: .claude/workflows/implement-issue-tree.js does not exist
- ...

### Existing Assets (items requiring overwrite confirmation)

The following already exist. Confirm individually if changes are needed.
- <file-path>: <summary of current state>

Do not start supplementation/changes without user approval.

Approval granularity:

"Supplement all" → Proceed to Step 4
"Supplement specific items" → Confirm targets then proceed to Step 4
"Overwrite existing files" → Confirm individually before overwriting

Step 4: Supplement differences

Only supplement approved items. Only modify existing files that have received overwrite confirmation.

4-1. Add missing Agents

Add to

.claude/agents/<category>/<name>.md

. If the target repo has a

dotclaude-via-temp

rule (via

_/dotclaude/

), follow that rule. If not, you can write directly to

.claude/

yaml

---
name: <name>
description: "<description of role (including trigger words)>"
model: <haiku|sonnet|opus|fable>
tools: [minimum necessary tools]
---

Follow Claude Code's subagent specification for frontmatter keys, using

name

(

subagent_type

is a parameter name for Agent tool calls, not a definition key).

4-2. Add missing Rules

Add to

.claude/rules/

. Customize

delegation.md

and

delegation-impl.md

according to the target repo's path structure, referencing examples from Fandhe-AI/agent-cli-skills.

code-comment-style.md

and

out-of-scope-tracking.md

are missing, generate them by referring to the template outline described in "3-3. Generate rules/" of the init-claude skill. Adjust according to the target repo's language and structure (documentation comment format, directory structure, etc.) when generating.

4-3. Supplement missing Skills

bash

cd <target-repo>
npx skills add Fandhe-AI/agent-cli-skills

Confirm that

skills-lock.json

is updated.

4-4. Supplement hooks

Add/update SessionStart hook in

settings.json

json

{
  "hooks": {
    "SessionStart": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "echo '<repo-name>: Communicate in Japanese / Delegate tasks to subagents to reduce main model consumption / Strictly follow Conventional Commits (--no-verify prohibited) / Implement implement-issue only after plan approval'"
          }
        ]
      }
    ]
  }
}

Security Notes:

Do not embed API keys, tokens, or passwords in the
```
command
```
value
Do not directly expand user input into
```
command
```

Propose PostToolUse auto-formatting hooks after verifying the existence of language-specific tools, and only add them if the user wishes.

4-5. Update CLAUDE.md

Supplement missing sections (delegation policy table, Sub-agents list, Rules list, model allocation table, Current Skills). Only modify approved items when overwriting existing sections.

4-6. Prepare implement-issue-tree prerequisites

Guidance if workflow js does not exist:

If not called as a named workflow (

{name: "implement-issue-tree"}

), placement in

.claude/workflows/

is not necessary at all—you can directly specify

.claude/skills/implement-issue-tree/script/implement-issue-tree.js

in the Workflow tool's

scriptPath

If placing as a named workflow, use a relative symlink instead of

cp

. If placed via

cp

, the symlink will be replaced with an actual file, and updates via

npx skills add

will not reach the named workflow.

bash

# Place in .claude/workflows/ (only if using as a named workflow)
mkdir -p <target-repo>/.claude/workflows/

# Leave existing symlinks as-is (do not overwrite actual files)
if [ ! -e "<target-repo>/.claude/workflows/implement-issue-tree.js" ]; then
  # Create symlink with relative path from .claude/workflows/
  ln -s ../skills/implement-issue-tree/script/implement-issue-tree.js \
        <target-repo>/.claude/workflows/implement-issue-tree.js
else
  echo "Already exists (check if symlink or actual file: ls -la <target-repo>/.claude/workflows/implement-issue-tree.js)"
fi

If the sub_issues API cannot be used, guide the user to enable GitHub Apps.

Step 5: Report supplementation results

Report Items:

List of supplemented files and changes made
Skipped items and reasons
Satisfaction status of implement-issue-tree prerequisites
Guidance on next actions for the user (items requiring manual setup, etc.)

Verification

bash

# Check list of files after supplementation
find <target-repo>/.claude -type f | sort

# Check existence of CLAUDE.md
ls <target-repo>/CLAUDE.md

# Check update to skills-lock.json
cat <target-repo>/skills-lock.json 2>/dev/null | head -20

# Check implement-issue-tree prerequisites
ls <target-repo>/.claude/workflows/implement-issue-tree.js 2>/dev/null || echo "workflow js: Not placed"
gh auth status

Notes

For repositories where
```
.claude/
```
does not exist, guide the user to use
```
init-claude
```
and interrupt processing
Only overwrite existing files after individual user approval
Do not start changes without user approval for diagnosis
Do not hardcode tokens or secrets in
```
settings.json
```
```
command
```
Do not include commands with
```
--no-verify
```
in hooks
If
```
npx skills add
```
fails, display the error message and guide the user through manual steps
Follow the
```
dotclaude-via-temp
```
rule if it exists in the target repo (if not, direct writing is allowed)
Follow the principle of least privilege when listing Agent
```
tools
```
—only include necessary items
If security issues (sensitive information inclusion, injection risks) are found, interrupt supplementation and warn the user

update-claude

NPX Install

Tags

SKILL.md Content (Chinese)

update-claude

Usage

Prerequisites

Flow

Step 1: Investigate the target repository and existing .claude/

Step 2: Diagnose differences from the ideal state

2-1. Agent Diagnosis

2-2. Rules Diagnosis

2-3. Skills Diagnosis

2-4. hooks Diagnosis

2-5. CLAUDE.md Diagnosis

2-6. implement-issue-tree Prerequisite Diagnosis

Step 3: Present gap list to user for approval

Step 4: Supplement differences

4-1. Add missing Agents

4-2. Add missing Rules

4-3. Supplement missing Skills

4-4. Supplement hooks

4-5. Update CLAUDE.md

4-6. Prepare implement-issue-tree prerequisites

Step 5: Report supplementation results

Verification

Notes