easy-claude-help

[Utilities] Configuration guide for the easy-claude framework — explain available settings, guide user/AI through configuring .claude/.ck.json.

When to Use

User asks "how do I configure Claude?", "how do I disable workflow confirmations?", "how do I turn off hooks?"
User wants to know what settings are available in easy-claude
User says "configure workflow mode", "set power user mode", "disable workflow detection"
AI needs to help user update their .claude/.ck.json configuration

Quick Reference

Config File Location

~/.claude/.ck.json        ← global (applies to all projects)
.claude/.ck.json          ← project (shared, committed to git)
.claude/.ck.local.json    ← personal override (gitignored, per-developer)

Each layer overrides the previous via deep merge — you only need to set values you want to override. Use .ck.local.json for personal preferences that shouldn't be committed (e.g., your own codingLevel or workflow.confirmationMode).

Key Settings

1. Workflow Confirmation Mode

Controls whether workflow detection requires user confirmation before activating.

Location: .claude/.ck.json → workflow.confirmationMode

Value	Behavior
`"always"`	(default) Always asks via AskUserQuestion before activating any workflow. Collaborative mode — best for most users.
`"never"`	Auto-executes detected workflow without asking. Use `quick:` prefix behavior globally. Best for power users who trust the detection.
`"off"`	Disables workflow detection entirely. Plain Claude behavior — no catalog injection, no confirmation gate, no overhead.

Example — disable confirmation gate:

{
    "workflow": {
        "confirmationMode": "never"
    }
}

Example — full opt-out (plain Claude):

{
    "workflow": {
        "confirmationMode": "off"
    }
}

Tip: Even in "always" mode, you can prefix any prompt with quick: to skip confirmation for that prompt only (e.g., quick: review the auth code).

2. Coding Level

Controls the response style and verbosity level. Affects how Claude explains things.

Location: .claude/.ck.json → codingLevel

Value	Style
`-1`	(default) Disabled — no style injection
`0`	ELI5 — explain like I'm 5
`1`	Junior developer
`2`	Mid-level developer
`3`	Senior developer
`4`	Tech lead
`5`	Expert/God mode — minimal explanation

Example:

{
    "codingLevel": 4
}

3. Language Settings

Configure Claude's thinking and response language.

Location: .claude/.ck.json → locale

{
    "locale": {
        "thinkingLanguage": "en",
        "responseLanguage": "fr"
    }
}

4. Assertions

Custom instructions injected into every session as reminders.

Location: .claude/.ck.json → assertions

{
    "assertions": ["Always use TypeScript strict mode", "Prefer functional components over class components"]
}

5. Plan & Naming

Control plan file naming format and date pattern.

Location: .claude/.ck.json → plan

{
    "plan": {
        "namingFormat": "{date}-{issue}-{slug}",
        "dateFormat": "YYMMDD-HHmm",
        "issuePrefix": "GH-"
    }
}

6. Paths

Override where plans and docs are stored.

Location: .claude/.ck.json → paths

{
    "paths": {
        "docs": "docs",
        "plans": "plans"
    }
}

7. Reference Docs Staleness

Controls how old reference docs can be before the staleness gate activates.

Location: .claude/.ck.json → referenceDocs.staleDays

Value	Behavior
`60`	(default) Warn after 60 days, block prompts until scanned or dismissed
`1-365`	Custom threshold in days

Example — relax to 90 days:

{
    "referenceDocs": {
        "staleDays": 90
    }
}

How it works:

On session start, checks  in each reference doc
If any doc is older than staleDays, shows a warning listing stale docs
On next prompt, blocks until you run /scan-all, /scan-*, or type skip scan
skip scan dismisses the gate for 24 hours

Tip: Run /scan-all to refresh all 10 reference docs at once.

Quick Configuration Examples

Power User Setup (minimal overhead)

{
    "workflow": {
        "confirmationMode": "never"
    },
    "codingLevel": 5
}

Opt-out Setup (plain Claude)

{
    "workflow": {
        "confirmationMode": "off"
    }
}

Team Setup (collaborative, Vietnamese responses)

{
    "workflow": {
        "confirmationMode": "always"
    },
    "codingLevel": 3,
    "locale": {
        "responseLanguage": "vi"
    }
}

How to Apply Config

Default: always use .claude/.ck.local.json (personal, gitignored) unless the user explicitly says "update project config", "update shared config", or "update .ck.json".

AI Config Update Protocol

Default → .claude/.ck.local.json (personal override, not committed to git)
- Read current local config: Read .claude/.ck.local.json
- If file doesn't exist, create it with only the desired settings
- If file exists, merge the new settings (preserve existing keys)
Only if user explicitly requests → .claude/.ck.json (shared, committed to git)
- Use when user says "update project config", "share with team", "commit this setting"
Confirm to user what was changed and which file was updated

Config changes take effect on the next prompt (no restart needed). .ck.local.json overrides .ck.json via deep merge — safe for personal preferences without affecting teammates.

Schema Validation

.ck.json files are validated on load. Invalid values (wrong types, out-of-range numbers, unknown keys) emit stderr warnings but never block — config loading uses graceful degradation.

What gets checked:

confirmationMode must be one of "always", "never", "off"
codingLevel must be a number between -1 and 5
assertions must be an array of strings
Unknown top-level keys produce warnings (catches typos like workfow)

CLI validation command:

node .claude/hooks/lib/ck-config-schema.cjs .claude/.ck.json

Workflow Settings (Advanced)

The workflow catalog itself lives in .claude/workflows.json. You can:

Set settings.enabled: false to disable workflow injection entirely
Change settings.overridePrefix (default "quick:") for the per-prompt skip prefix

Note: Prefer .ck.json → workflow.confirmationMode over editing workflows.json directly, as .ck.json supports global+local cascading.

Code Review Graph (Structural Code Intelligence)

Optional feature that builds a knowledge graph of your codebase for graph-blast-radius analysis and smarter code reviews.

Setup: pip install tree-sitter tree-sitter-language-pack networkx then /graph-build

Skills:

Skill	Purpose
`/graph-build`	Build or update the knowledge graph
`/graph-blast-radius`	Analyze impact of current changes
`/graph-export`	Export graph to JSON
`/graph-connect-api`	Detect frontend→backend API connections
`/graph-query`	Query code relationships (callers, imports, tests)
`/graph-export-mermaid`	Export single-file graph as Mermaid diagram

Auto-features (when graph is built):

Session start: shows graph status
Review skills: auto-inject blast radius analysis
File edits: auto-update graph in background

Frontend→Backend API detection: Auto-configured by /project-config when both frontend and backend are detected. Or manually add graphConnectors to docs/project-config.json.

Docs: .claude/docs/code-graph-mechanism.md for detailed architecture.