CLI Reference

Commands

Command	Description
`odek run [flags] <task>`	Execute a task with the agent loop (single-shot by default)
`odek run --session [flags] <task>`	Execute and save conversation as a multi-turn session
`odek run [--no-learn] [flags] <task>`	Execute with skill learning (on by default, use --no-learn to disable)
`odek continue [--id <id>] <task>`	Continue the most recent session (or by `--id`)
`odek repl [flags]`	Interactive REPL mode (persistent multi-turn session). Accepts `--model`, `--thinking`, `--sandbox`, and `--sandbox-*` flags.
`odek session list`	List sessions
`odek session show [id]`	Show session details (default: latest)
`odek session delete <id>`	Delete a session
`odek session trim <id> <n>`	Keep only the `n` most recent messages
`odek session cleanup <days>`	Delete sessions older than N days
`odek skill list`	List all available skills
`odek skill view <name>`	View a skill's full content
`odek skill delete <name>`	Delete a skill
`odek skill import <uri> [flags]`	Import a skill from file:// or https://
	`odek skill curate`
	`odek skill curate --apply`
	`odek skill curate --interactive`
	`odek skill reset-skips [name]`
	`odek serve [--addr :8080] [--open]`
	`odek subagent --goal <string> [flags]`
`odek init [--global] [--force]`	Create a config file template
`odek mcp [--sandbox]`	Start MCP server (expose tools to Claude Code) or connect to external MCP servers (via `mcp_servers` config)
`odek version`	Print version and exit

Run flags

Flag	Type	Default	Description
`--model <name>`	string	`deepseek-chat`	LLM model — profiles auto-set thinking/timeout (see Providers). Consider using `deepseek-v4-flash` for faster/cheaper tasks.
`--base-url <url>`	string	`https://api.deepseek.com/v1`	OpenAI-compatible API endpoint
`--max-iter <n>`	int	`90`	Max think→act cycles
`--thinking <level>`	string	profile default	Reasoning depth: `enabled`/`disabled`/`low`/`medium`/`high`
`--sandbox`	bool	false	Execute shell commands inside Docker container
`--deliver`	bool	false	Deliver the agent's final response to the configured Telegram `default_chat_id`. Requires `telegram.bot_token` + `telegram.default_chat_id` in config. Use with cron for scheduled agent tasks.
`--interaction-mode <mode>`	string	`engaging`	Tool-call rendering: `engaging` (emoji narration) or `verbose` (raw tool output)
`--no-color`	bool	false	Disable colored terminal output
`--prompt-caching`	bool	false	Enable Anthropic/OpenAI/DeepSeek prompt caching markers
`--no-agents`	bool	false	Skip loading AGENTS.md
`--session`	bool	false	Save conversation as a multi-turn session
`--learn`	bool	`true`	Enable skill learning mode (detects patterns, saves skills). On by default
`--no-learn`	bool	`false`	Disable skill learning mode (overrides config/default)
`--system <prompt>`	string	built-in	Override system prompt
`--ctx <files>` / `-c`	string	—	Attach comma-separated files as context blocks

File attachments

Attach files to any task prompt to provide the agent with context — no tool calls needed. File content is injected as context blocks before the prompt.

`--ctx` / `-c` (CLI)

# Single file
odek run --ctx data.csv "analyze this"

# Multiple files (comma-separated)
odek run --ctx main.go,lib.go "compare these files"

# Short flag
odek run -c config.json "validate"

# With session persistence
odek run --session --ctx schema.sql "design the migration"

`@file` inline references

Reference files directly in the task text — works in odek run, odek continue, REPL, and Web UI:

odek run "@README.md summarize this project"
odek run --session "@auth.go review the security"
odek continue "@auth.go now add rate limiting" --id 20260518-abc123

When @ref resolution fails (file not found), the reference is left as-is in the prompt.

Web UI

In odek serve:

Paperclip button next to the input to pick files
Drag-and-drop files onto the chat area
Attached file chips show filename, size, and a remove button
@ autocomplete dropdown for files and sessions
5 MB per file, 10 MB total per prompt

Implementation

Files are read client-side in the Web UI and server-side in the CLI through enrichTask() in the cmd/odek/refs.go package. The resource package handles resolution: file content is wrapped in --- filename --- / --- end filename --- markers and prepended to the task.

Sandbox file injection

When --sandbox is active, --ctx files are automatically copied into the sandbox container via docker cp, placed at /workspace/<relative-path>. Files outside the working directory use their basename. This means the agent can use tools (read_file, shell cat, patch) on the same files it sees in context — no "content visible but file doesn't exist" gap. Directories and missing files are silently skipped.

Shell tool schema

{
  "command": "rm -rf /var/log",
  "description": "Clear stale nginx logs before restarting"
}

Dangerous operations

When running without --sandbox, odek classifies every shell command by risk and prompts for high-risk operations:

Class	Default	Examples
🟢 safe	allow	`ls`, `cat`, `grep`, `go build`
🟡 local_write	allow	`rm file`, `mv`, `echo > file`
🟠 system_write	prompt	`sudo`, `apt install`, writes to `/etc/`
🔴 destructive	deny	`rm -rf /`, `dd if=/dev/zero`, `mkfs`
🔴 network_egress	prompt	`curl`, `git push`, `ssh`, `scp`
🔴 code_execution	prompt	`curl url \| bash`, `eval`, `node -e`, `go run`
🟠 install	prompt	`npm install`, `pip install`, `go install <path>`
⬛ blocked	deny	Fork bombs, `dd` to block devices

The approval prompt accepts:

A — Approve once
D — Deny (returns error to agent)
T — Trust all commands of this class for this session
? — Show full context

Configurable via dangerous section in ~/.odek/config.json or ./odek.json:

{
  "dangerous": {
    "action": "prompt",
    "non_interactive": "allow",
    "classes": {
      "destructive": "prompt",
      "network_egress": "allow"
    },
    "allowlist": ["git push origin main"],
    "denylist": ["rm -rf /"]
  }
}

See docs/SECURITY.md for details.

Skills

The skills system provides just-in-time domain knowledge to the agent. Skills are SKILL.md files with YAML frontmatter that define trigger keywords, quality metadata, and markdown body content.

How skills work

Skills are stored in ~/.odek/skills/<name>/SKILL.md (user-global) or ./.odek/skills/<name>/SKILL.md (project)
Skills with auto_load: true are injected into the system prompt on start
Lazy skills are loaded on demand when the user's input matches their trigger keywords (topic × action)
The --learn flag detects reusable patterns during a run and prompts to save as a draft skill

Skill commands

# List all skills
odek skill list

# View a skill's full content
odek skill view docker-build

# Delete a skill
odek skill delete docker-build

# Import a skill from a file or URL
odek skill import ./skills/my-skill.md
odek skill import https://example.com/skills/deploy.md

# Import with flags
odek skill import https://example.com/skills/deploy.md --basic   # skip LLM risk assessment
odek skill import https://example.com/skills/deploy.md --yes     # auto-approve (scripting)

# Run curation (quality, staleness, overlap, dedup checks)
odek skill curate

# Apply all curation suggestions automatically
odek skill curate --apply

# Review curation suggestions one-by-one
odek skill curate --interactive

# Reset skip list (re-enable suppressed suggestions)
odek skill reset-skips              # clear all
odek skill reset-skips procedure-grep  # clear specific skill

Skill file format

---
name: docker-build
description: Build and optimize Docker images
version: 1.0.0
author: odek
`odek:
  trigger:
    topic: docker container image
    action: build optimize
  auto_load: false
  quality: verified
---
## Overview

Procedure for building optimized Docker images.

## Step-by-Step

1. Write a `.dockerignore` file
2. Use multi-stage builds
3. Run `docker build -t <name> .`

## Common Pitfalls

- Forgetting `.dockerignore` leads to large build contexts
- Not pinning base image versions causes build drift

## Verification

- `docker build` exits with code 0
- `docker images` shows the new image

Curation

The odek skill curate command runs four quality passes:

Staleness — flags skills unused for 90+ days (configurable via skills.curation.staleness_days)
Trigger overlap — detects skills with 2+ shared topic keywords that may need merging
Quality audit — checks for missing sections, short bodies, long descriptions
Body dedup — detects skills with identical body content by SHA256 hash

Auto-curation runs after every session where skills are auto-saved (skills.curation.auto_curate: true by default):

Merge — overlapping draft-quality skills are automatically merged (union keywords, concatenated bodies)
Skip deletion — skills skipped ≥ skip_threshold times are auto-deleted
Stale pruning — if auto_prune: true, stale skills are deleted automatically

Run odek skill curate --apply to manually trigger the full curation pipeline with merge execution. Use odek skill reset-skips to clear the skip list and re-enable suppressed suggestions.

Sandbox flags

Flag	Default	Description
`--sandbox-image <img>`	`alpine:latest`	Docker image
`--sandbox-network <mode>`	`none`	Network: `none`/`bridge`/`host`
`--sandbox-readonly`	false	Mount working directory read-only
`--sandbox-memory <s>`	—	Memory limit (e.g. `512m`, `2g`)
`--sandbox-cpus <n>`	—	CPU limit (e.g. `0.5`, `2`)
`--sandbox-user <s>`	—	Run as user (`uid:gid`)

Init flags

Flag	Description
`--global`, `-g`	Create global config at `~/.odek/config.json`
`--force`, `-f`	Overwrite existing file without prompting

Examples

# Quick task (single-shot, no session saved)
odek run "How many Go files in this project?"

# Save as session for follow-up
odek run --session "Refactor the auth module"

# Continue a session
odek continue "Now add error handling to the refactored auth"

# Continue a specific session by ID
odek continue --id 20260518-abc123 "Add unit tests"

# List all sessions
odek session list

# Show latest session transcript
odek session show

# Show a specific session
odek session show 20260518-abc123

# Trim session to last 10 messages (preserves system prompt)
odek session trim 20260518-abc123 10

# Delete sessions older than 30 days
odek session cleanup 30

# Wipe all sessions
odek session cleanup 0

# OpenAI
odek run --model gpt-4o --base-url https://api.openai.com/v1 "Explain this code"

# Sandboxed execution
odek run --sandbox "npm test"

# Custom sandbox image
odek run --sandbox --sandbox-image node:20-alpine "node --version"

# Interactive REPL with sandbox
odek repl --sandbox --model deepseek-v4-pro

# Sandbox REPL with custom image and no network
odek repl --sandbox --sandbox-image node:20-alpine --sandbox-network none

# Resume a sandboxed session in REPL mode
odek repl --id 20260518-abc123

# Custom system prompt
odek run --system "You are a Go expert. Answer with code only." "Write HTTP server"

# Run with skill learning (on by default — use --no-learn to disable)
odek run "Set up CI with GitHub Actions"

# File attachments
odek run --ctx go.mod "check go version"
odek run -c main.go,util.go "refactor both files"
odek run "&#64;schema.sql design a migration plan"

# Cron integration: deliver agent result to Telegram
odek run --deliver "Daily weather forecast for Berlin"
odek run --deliver "Check the CI pipeline status"

# Systemd cron example (crontab -e):
# */5 * * * * /usr/local/bin/odek run "Say hello" --deliver >> /tmp/odek-cron.log 2>&1

Config priority

Config sources from lowest to highest priority:

1.  ~/.odek/config.json    ← Global defaults
2.  ./odek.json           ← Project overrides
3.  KODE_* env vars       ← Runtime overrides
4.  CLI flags             ← Explicit invocation (highest)

See Configuration for details.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

CLI Reference

Commands

Run flags

File attachments

`--ctx` / `-c` (CLI)

`@file` inline references

Web UI

Implementation

Sandbox file injection

Shell tool schema

Dangerous operations

Skills

How skills work

Skill commands

Skill file format

Curation

Sandbox flags

Init flags

Examples

Config priority

FilesExpand file tree

CLI.md

Latest commit

History

CLI.md

File metadata and controls

CLI Reference

Commands

Run flags

File attachments

--ctx / -c (CLI)

@file inline references

Web UI

Implementation

Sandbox file injection

Shell tool schema

Dangerous operations

Skills

How skills work

Skill commands

Skill file format

Curation

Sandbox flags

Init flags

Examples

Config priority

`--ctx` / `-c` (CLI)

`@file` inline references