Coding FriendCoding Friendbeta

Configuration File

Customize Coding Friend behavior with .coding-friend/config.json.

Overview

Configure Coding Friend using .coding-friend/config.json. Two locations:

  • Global: ~/.coding-friend/config.json — applies to all projects
  • Local: .coding-friend/config.json — project-specific, overrides global

Edit interactively with cf config, or manually edit the JSON files.

Full Config

{
  "language": "en",
  "docsDir": "docs",
  "privacyBlock": true,
  "scoutBlock": true,
  "commit": {
    "verify": true
  },
  "learn": {
    "language": "en",
    "outputDir": "~/.coding-friend/learn",
    "categories": [
      {
        "name": "concepts",
        "description": "Design patterns, algorithms, architecture principles"
      },
      {
        "name": "patterns",
        "description": "Repository pattern, observer pattern"
      },
      {
        "name": "languages",
        "description": "Language-specific features, syntax, idioms"
      },
      { "name": "tools", "description": "Libraries, frameworks, CLI tools" },
      { "name": "debugging", "description": "Debugging techniques, bug fixes" }
    ],
    "autoCommit": false,
    "readmeIndex": false
  },
  "statusline": {
    "components": ["version", "folder", "model", "branch", "context", "usage"],
    "accountAliases": {
      "me@work.com": "Work"
    }
  },
  "autoApprove": false,
  "autoApproveCodex": false,
  "autoApproveAllowExtra": [],
  "autoApproveIgnore": [],
  "disableGUIPlan": true,
  "guiPlanFormat": "html",
  "memory": {
    "tier": "auto",
    "embedding": {
      "provider": "transformers",
      "model": "Xenova/all-MiniLM-L6-v2",
      "ollamaUrl": "http://localhost:11434"
    },
    "autoCapture": false,
    "autoStart": false
  },
  "review": {
    "withCodex": false
  }
}

Properties

language

  • Type: string
  • Default: "en"
  • Language for doc-generating skills (/cf-ask, /cf-remember, /cf-plan, /cf-research)

docsDir

  • Type: string
  • Default: "docs"
  • Top-level folder name where all skill outputs are stored
  • Must be a single folder name (no slashes) — all subfolders (plans/, memory/, research/, sessions/) sit inside it

privacyBlock

  • Type: boolean
  • Default: true
  • Enables the privacy-block hook that prevents reading sensitive files (.env, credentials, etc.)
  • Set to false to disable

scoutBlock

  • Type: boolean
  • Default: true
  • Enables the scout-block hook that prevents the agent from reading too many files at once
  • Set to false to disable

commit

Settings for the /cf-commit workflow.

commit.verify

  • Type: boolean
  • Default: true
  • Run tests before allowing commit
  • Set to false to skip pre-commit verification

learn

Settings for the /cf-learn skill.

learn.language

  • Type: string
  • Default: "en"
  • Language for /cf-learn notes
  • Falls back to top-level language, then "en"

learn.outputDir

  • Type: string
  • Default: "~/.coding-friend/learn"
  • Where learning docs are saved — always stored globally (never per-project)
  • Supports absolute paths (/path/to/dir) or ~/ paths

learn.disabled

  • Type: boolean
  • Default: false
  • Set to true to disable CF Learn file writing and MCP serving entirely

learn.categories

  • Type: array of { name, description }
  • Default: 5 categories (concepts, patterns, languages, tools, debugging)
  • Subdirectories and their descriptions for organizing learning docs

learn.autoCommit

  • Type: boolean
  • Default: false
  • Auto-commit after each /cf-learn (useful for external output dirs)

learn.readmeIndex

  • Type: boolean | "per-category"
  • Default: false
  • false — no index
  • true — single README containing all indexes
  • "per-category" — one README per category folder

statusline

Settings for the statusline display (global config only).

statusline.components

  • Type: array of component IDs
  • Default: all components shown
  • Available components: "version", "folder", "model", "branch", "context", "usage"
  • Only listed components are shown — omit a component to hide it

statusline.accountAliases

  • Type: Record<string, string> (email → alias)
  • Default: {} (no aliases)
  • Custom display names for accounts, keyed by email address
  • When set, the alias replaces the full name/email display in the statusline (e.g., 👤 Work instead of 👤 Name (email))
  • Each account can have its own alias — switching accounts preserves all aliases
  • Set interactively via cf statusline, cf config, or cf init

autoApprove

  • Type: boolean
  • Default: false
  • Enables the auto-approve hook that reduces permission prompts
  • Uses a 3-step classification pipeline (rules → working-dir → LLM) to auto-approve safe operations

autoApproveCodex

  • Type: boolean
  • Default: false
  • Enables deterministic-only auto-approve for Codex CLI
  • Unknown actions defer to Codex native approval; the Claude Sonnet classifier is not used from Codex hooks
  • Set with cf permission --agent codex --enable-auto-approve or cf permission --agent codex --disable-auto-approve

autoApproveAllowExtra

  • Type: string[]
  • Default: []
  • Additional Bash command prefixes to auto-approve (merged with built-in allow list)
  • Entries are matched as command prefixes: "cargo test" matches cargo test, cargo test --lib, etc.
  • Cannot override DENY patterns — adding "rm -rf" does not bypass the built-in deny
  • Both global and local configs are merged (union, deduped)
  • See Auto-Approve reference for details

autoApproveIgnore

  • Type: string[]
  • Default: []
  • Bash command prefixes the hook should not classify — returns no decision so Claude Code's native permissions.allow patterns handle them
  • Only applies to commands that would be prompted ("ask") or sent to the LLM classifier ("unknown"). Security DENY patterns are always enforced
  • For compound commands (pipes, chains), the first segment is checked
  • Both global and local configs are merged (union, deduped)
  • Use this when you have flexible glob patterns in .claude/settings.json (e.g. "Bash(cargo test * | *)") and want Claude Code to decide instead of the hook
  • See Auto-Approve reference for the difference with autoApproveAllowExtra

disableGUIPlan

  • Type: boolean
  • Default: true
  • When true (the default), /cf-plan does not generate the human-readable overview doc (it costs extra tokens)
  • Set to false to generate the overview alongside every agent plan
  • Per-run override: pass --gui (alias --human) to generate the overview for a single run without changing config

guiPlanFormat

  • Type: "html" | "md"
  • Default: "html"
  • Format for the /cf-plan human overview doc
  • "html" — styled, self-contained HTML file with embedded Mermaid diagrams
  • "md" — plain Markdown file with mermaid fences

memory

Settings for the memory system.

memory.tier

  • Type: "auto" | "full" | "lite" | "markdown"
  • Default: "auto"
  • Controls which search backend the MCP server uses — this does not start or stop the daemon
  • "auto" — detect best available at startup: SQLite installed → Tier 1, daemon running → Tier 2, otherwise → Tier 3
  • "full" — require Tier 1 (SQLite + hybrid search). Fails if deps not installed via cf memory init
  • "lite" — require Tier 2 (MiniSearch daemon). Fails if daemon not running via cf memory start-daemon
  • "markdown" — force Tier 3 (grep over markdown files). Always works, no setup needed
  • Use this to force a specific tier for testing or to skip auto-detection (e.g., "markdown" to use Tier 3 even when SQLite is installed)

memory.embedding

Settings for the embedding model used in Tier 1 semantic search. See Embedding Models for details.

memory.embedding.provider

  • Type: "transformers" | "ollama"
  • Default: "transformers"
  • "transformers" — local Transformers.js model, no external service needed
  • "ollama" — local Ollama server, supports more models

memory.embedding.model

  • Type: string
  • Default: "Xenova/all-MiniLM-L6-v2" (transformers) or "all-minilm:l6-v2" (ollama)
  • The embedding model to use for vector generation
  • Changing to a model with different dimensions requires cf memory rebuild

memory.embedding.ollamaUrl

  • Type: string
  • Default: "http://localhost:11434"
  • URL of the Ollama server (only used when provider is "ollama")

memory.autoCapture

  • Type: boolean
  • Default: false
  • Enable automatic memory capture via PreCompact hook
  • When enabled, session summaries are saved as episode memories before context compaction

memory.autoStart

  • Type: boolean
  • Default: false
  • Auto-start the daemon when the MCP server connects

review

Settings for the /cf-review skill.

review.withCodex

  • Type: boolean
  • Default: false
  • Run a Codex second-opinion review alongside Claude's review by default — equivalent to always passing --with-codex
  • When true, every /cf-review (including the auto-invoked ones from /cf-plan, /cf-fix, /cf-optimize) runs a Codex review in parallel and merges both reviews into one report. Codex auto-selects its scope (committed branch changes via --base, unpushed commits on the base branch, or --uncommitted) so it covers a phase even after it is committed
  • Requires the Codex CLI installed and logged in; degrades gracefully to a Claude-only review with a warning if Codex is unavailable

Layered Config

Local settings override global settings at the top-level key level (not deep merge).

  • If global has { "language": "en", "docsDir": "docs" } and local has { "language": "vi" }, the result is { "language": "vi", "docsDir": "docs" }
  • If local has a learn object, it replaces the entire global learn object

Custom Skill Guides

Separate from config.json, you can extend built-in skills with project-specific guidance. See Custom Skill Guides for details.

Previous
cf update
Next
Custom Skill Guides