Skip to content

cli-reference.md

Latest commit

 

History

History
351 lines (245 loc) · 8.63 KB

cli-reference.md

File metadata and controls

351 lines (245 loc) · 8.63 KB
layout docs

CLI Reference

Codaph is Mubit-first by default. The primary CLI path is:

  1. codaph init
  2. codaph sync
  3. codaph status
  4. codaph tui
  5. codaph import (optional backfill)

This page documents the user-facing commands first and keeps advanced commands at the end.

Command Overview

codaph --help

Primary commands:

  • setup (global config)
  • init (repo setup + onboarding)
  • sync (fast daily sync)
  • status (repo sync + automation status)
  • import (historical agent backfill)
  • tui (terminal UI)

Advanced commands are available but not required for normal usage. This includes the local MCP server (codaph mcp) for Claude Code and other MCP clients.

codaph setup

Use this command to set global secrets and defaults.

codaph setup --mubit-api-key <key>

Common uses:

  • set Mubit API key globally
  • set OpenAI API key for query synthesis
  • set global actor id

Examples:

codaph setup --mubit-api-key <key>
codaph setup --openai-api-key <key>
codaph setup --mubit-actor-id <your-login>

codaph init

Use this command once per repository.

cd /absolute/path/to/project
codaph init

What it does:

  • detects or stores repo identity (owner/repo when available)
  • creates .codaph/project.json
  • writes .codaph/mcp/claude-code.json (Claude Code MCP config template)
  • prompts for Mubit API key if missing
  • detects .codex, .claude, and .gemini folders and lets you multi-select agent integrations (preselected when recognized)
  • installs repo-scoped auto-sync hooks (best effort)

Useful flags:

  • --yes non-interactive mode
  • --force reinstall/reapply setup behavior
  • --no-auto-sync skip hook install
  • --cwd <path> run init for another repo
  • --providers <csv|auto> set repo agent integrations without prompting
  • --agent-hooks <csv|all|none> choose agent-complete hooks to install without prompting

codaph sync

Use this command for day-to-day syncing.

codaph sync

Behavior:

  • fast Mubit-first sync path
  • cloud pull into local mirror
  • repo-local automation/status integration
  • no global agent history replay by default (fast daily sync stays Mubit-first)

Useful flags (most users do not need these):

  • --cwd <path> run from another directory
  • --json machine-readable output
  • --no-auto-enable skip first-run automation prompt during sync
  • --mubit-write-timeout-ms <ms> tune write timeout for slow networks

codaph sync subcommands (advanced / compatibility)

These are available when you need manual control.

codaph sync all
codaph sync pull
codaph sync status
codaph sync setup

Compatibility aliases are still supported:

  • codaph sync remote (alias for codaph sync pull)
  • codaph sync push (compat alias for codaph import)

codaph import

Use this command for historical backfill from local agent session history (Codex / Claude Code / Gemini CLI).

codaph import

Behavior:

  • scans ~/.codex/sessions
  • also scans local Claude Code / Gemini CLI transcript locations when enabled
  • imports sessions/transcripts that match the current repo path and linked Git worktree paths by default
  • writes to local .codaph mirror
  • writes to Mubit when enabled

Use this command:

  • once after onboarding
  • after long periods of running agents outside Codaph hooks
  • when rebuilding a local mirror from your machine's local agent history

Useful flags:

  • --cwd <path>
  • --json
  • --local-only (compat alias to disable Mubit writes)
  • --providers <csv|all|auto> choose which agent histories to backfill
  • --no-worktrees limit matching to the current path only (skip linked Git worktree paths)
  • --mubit-write-timeout-ms <ms>

codaph status

Use this command to understand sync and automation state.

codaph status

It shows:

  • repo id
  • auto-sync settings
  • configured agent integrations + agent-complete hook providers
  • local push timestamps/counters
  • remote pull timestamps/counters
  • snapshot fingerprint and cap diagnostics

Useful flags:

  • --cwd <path>
  • --json

codaph tui

Launch the terminal UI.

codaph tui

Useful flags:

  • --cwd <path>
  • --mubit / --no-mubit

The TUI is primarily a viewer. Use codaph sync or s/r in the TUI to refresh data.

Common Environment Variables

Most users only need:

  • MUBIT_API_KEY
  • OPENAI_API_KEY (optional)

Codaph can also use:

  • OPENAI_MODEL (optional override, default OpenAI synthesis/chat model is gpt-4.1-mini)
  • CODAPH_PROJECT_ID
  • CODAPH_ACTOR_ID
  • CODAPH_MUBIT_RUN_SCOPE
  • MUBIT_APIKEY (fallback)
  • OPENAI_APIKEY (fallback)

Advanced Commands

These commands are useful for debugging, inspection, and power users.

Diagnostics

codaph doctor mubit

Use codaph doctor mubit when you want memory-health diagnostics from Mubit plus a quick snapshot-state summary for the current project run.

Session inspection

codaph sessions list
codaph inspect --session <session-id>
codaph timeline --session <session-id>
codaph diff --session <session-id>

Replay and remote pull

codaph pull
codaph pull --full
codaph mubit replay --mode snapshot
codaph mubit replay --mode activity

Use pull --full or replay --mode activity when snapshot-based remote pull looks truncated or when you are rebuilding a local mirror on a new machine.

Context blocks

codaph mubit context "what should the next agent know?"
codaph mubit context "what should the next agent know?" --session <session-id>
codaph mubit snapshot
codaph mubit activity --limit 20 --exclude-derived --projection compact
codaph mubit export --limit 20 --format jsonl

Use mubit context before starting a new agent run or handing off work to another provider. Use mubit snapshot when you want assembled run state, and mubit activity / export when you want chronological audit data. Prefer --exclude-derived --projection compact when you want a cleaner audit view.

Checkpoints

codaph checkpoint "before-auth-refactor"
codaph checkpoint "before-auth-refactor" --session <session-id>

Use checkpoints before risky edits, migrations, or investigations you may need to resume later.

Diagnosis and lessons

codaph doctor mubit
codaph mubit diagnose "auth failure in CI"
codaph mubit reflect
codaph mubit strategies

Useful flags:

  • --session <session-id> narrow context to one run
  • --json machine-readable output

Use diagnose when an agent is stuck on a repeated failure, and reflect after meaningful work so Codaph can store reusable lessons.

Agents and handoffs

codaph agents list
codaph handoff send --task auth-cleanup --from claude-code --to codex --action continue "continue validating the auth cleanup"
codaph handoff list
codaph handoff feedback --handoff <handoff-id> --verdict approve --comments "looks good"

Use handoffs when one agent should explicitly continue the work of another instead of relying on a free-form summary.

Mubit queries

codaph mubit query "what changed in auth?"
codaph mubit query "what is the current direction of this repo?" --rank-by freshness
codaph mubit query "what changed in auth?" --session <session-id>

OpenAI-assisted synthesis is optional. Without OPENAI_API_KEY, Codaph falls back to the Mubit response. When the answer path is weak, Codaph can attach a clearly labeled context supplement instead of silently replacing the query result. Use --rank-by freshness for “current state”, “what changed recently”, and handoff-style questions.

Useful query flags:

  • --limit <n>
  • --rank-by relevance|balanced|freshness
  • --explain
  • --lane-filter <lane>
  • --min-timestamp <epoch>
  • --max-timestamp <epoch>
  • --budget <value>
  • --raw
  • --agent / --no-agent
  • --openai-model <model> (override OpenAI model for this query)

MCP server (Claude Code / MCP clients)

codaph mcp

Start the local Codaph MCP server over stdio.

codaph mcp setup claude

Use this helper to print or run the recommended Claude Code registration command.

codaph mcp setup claude
codaph mcp setup claude --run

Useful flags:

  • --scope user|project|local (default: user)
  • --mode codaph|npx (default: codaph)
  • --cwd <path>
  • --json

Recommended flow:

  1. Run codaph init (generates .codaph/mcp/claude-code.json)
  2. Run codaph mcp setup claude --scope user --run
  3. Verify in Claude Code with /mcp

See MCP Setup (Claude Code) for manual claude mcp add and JSON config options.

Direct capture (advanced)

codaph run "Summarize this repo"
codaph exec "Refactor auth flow"

Exit Codes

  • 0 on success
  • non-zero on command or runtime error