Context

# Output context for AI agents (used in Claude Code SessionStart hook)
agkan context

# Output in JSON format (for SessionStart hook integration)
agkan context --hook

Note: agkan agent-guide --hook is deprecated. Use agkan context --hook instead.

Task Operations

# Create task
agkan task add "Title" "Body"
agkan task add "Title" --status ready --author "agent"
agkan task add "Subtask" --parent 1
agkan task add "Title" --file ./spec.md  # Read body from file
agkan task add "Title" --blocked-by 1,2  # Set tasks that block this task
agkan task add "Title" --blocks 3,4      # Set tasks that this task blocks
agkan task add "Title" --assignees "alice,bob"  # Set task assignees (comma-separated)
agkan task add "Title" --branch feature/my-branch  # Assign a branch to this task
agkan task add "Title" --priority high            # Set task priority (critical/high/medium/low)
agkan task add "Title" --tag frontend,3           # Attach tags at creation (comma-separated names or IDs)
agkan task add "Title" --model-planning opus      # Model used for planning runs
agkan task add "Title" --model-run sonnet         # Model used for implementation runs
agkan task add "Title" --effort-planning high     # Reasoning effort for planning runs
agkan task add "Title" --effort-run xhigh         # Reasoning effort for implementation runs

# List tasks
agkan task list                    # All tasks
agkan task list --status in_progress
agkan task list --tree             # Hierarchical view
agkan task list --root-only        # Root tasks only
agkan task list --tag 1,2          # Filter by tags
agkan task list --dep-tree         # Dependency (blocking) tree view
agkan task list --sort title       # Sort by field (id / title / status / priority / created_at / updated_at), default: created_at
agkan task list --order asc        # Sort order (asc / desc), default: desc
agkan task list --assignees "alice,bob"  # Filter by assignees (comma-separated)
agkan task list --all              # Include all statuses (including done and closed)
agkan task list --priority high,critical  # Filter by priority
agkan task list --sort priority           # Sort by priority
agkan task list --archived         # Include archived tasks in the results

# Get details
agkan task get <id>

# Search
agkan task find "keyword"
agkan task find "keyword" --all  # Include done/closed
agkan task find "keyword" --status ready,in_progress  # Filter by status (validated: icebox/backlog/ready/in_progress/review/done/closed)

# Update (positional argument form - backward compatible)
agkan task update <id> status in_progress

# Update (named option form - v1.6.0+)
agkan task update <id> --status in_progress
agkan task update <id> --title "New Title"
agkan task update <id> --body "New body text"
agkan task update <id> --author "agent"
agkan task update <id> --assignees "alice,bob"
agkan task update <id> --file ./spec.md  # Read body from file
agkan task update <id> --status done --title "Updated Title"  # Multiple options
agkan task update <id> --branch feature/my-branch  # Set the branch column
agkan task update <id> --model-run sonnet --effort-run high  # Set the task's run model / effort (planning: --model-planning / --effort-planning)
agkan task update <id> --priority high            # Update task priority
agkan task update <id> --priority ""              # Clear task priority
agkan task update <id> --model-planning opus      # Set the planning model
agkan task update <id> --model-run sonnet         # Set the implementation model
agkan task update <id> --effort-planning high     # Set the planning reasoning effort
agkan task update <id> --effort-run xhigh         # Set the implementation reasoning effort
agkan task update <id> --model-run ""             # Pass an empty string to clear the value

# Count
agkan task count
agkan task count --status ready --quiet  # Output numbers only

# Update parent-child relationship
agkan task update-parent <id> <parent_id>
agkan task update-parent <id> null  # Remove parent

# Copy task
agkan task copy <id>
agkan task copy <id> --status ready    # Specify destination status (default: backlog)
agkan task copy <id> --no-tags         # Do not copy tags
agkan task copy <id> --json            # Output in JSON format

# Delete task
agkan task delete <id>
agkan task delete <id> --dry-run              # Preview the impact of deletion without deleting

# Archive done/closed tasks (hide but recoverable via unarchive)
agkan task archive                            # Archive done/closed tasks older than 3 days ago (default)
agkan task archive --before 2026-01-01        # Archive tasks last updated before the given date (ISO 8601)
agkan task archive --status done              # Target specific statuses (default: done,closed)
agkan task archive --dry-run                  # Preview tasks that would be archived without archiving
agkan task archive --json                     # Output in JSON format

# Unarchive a specific task (restore from archived state)
agkan task unarchive <id>                     # Unarchive a task by ID
agkan task unarchive <id> --dry-run           # Preview without unarchiving
agkan task unarchive <id> --json              # Output in JSON format

# Purge old done/closed tasks (permanent deletion, irreversible)
agkan task purge                        # Delete done/closed tasks older than 3 days ago (default)
agkan task purge --before 2026-01-01    # Purge tasks last updated before the given date (ISO 8601)
agkan task purge --status done          # Target specific statuses (default: done,closed)
agkan task purge --dry-run              # Preview tasks that would be purged without deleting
agkan task purge --json                 # Output in JSON format

-p is the short form of --priority, not --parent. Specify a parent with the long form --parent <id> only.

Model / effort values

Models: fable, opus, sonnet, haiku, gpt-6-astra, gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, gemini-3.8-flash, gemini-3.7-flash, claude-sonnet-4-6, claude-opus-4-6-thinking, gpt-oss-120b-medium

Efforts: low, medium, high, xhigh, max

The accepted set comes from the modelCatalog config key. Run agkan task add --help to see the list this installation actually accepts.

archive vs purge

archive: Sets the is_archived flag to hide tasks from default views. Recoverable with unarchive.

purge: Permanently deletes tasks. Irreversible.

Blocking Relationships

# task1 blocks task2 (task2 cannot be started until task1 is complete)
agkan task block add <blocker-id> <blocked-id>
agkan task block remove <blocker-id> <blocked-id>
agkan task block list <id>

Comment Operations

# Add a comment to a task
agkan task comment add <task-id> <content>
agkan task comment add <task-id> <content> --author "agent"

# List all comments for a task
agkan task comment list <task-id>

# Update a comment by ID
agkan task comment update <comment-id> <content>

# Delete a comment by ID
agkan task comment delete <comment-id>

Tag Operations

# Tag management
agkan tag add "frontend"
agkan tag list
agkan tag delete <tag-id-or-name>
agkan tag delete <tag-id-or-name> --dry-run  # Preview the impact of deletion without deleting
agkan tag rename <id-or-name> <new-name>

# Tag tasks
agkan tag attach <task-id> <tag-id-or-name>
agkan tag detach <task-id> <tag-id-or-name>
agkan tag show <task-id>

Export / Import Operations

# Export all tasks to JSON (output to stdout)
agkan export

# Export to file
agkan export > backup.json

# Import tasks from a JSON export file
agkan import <file>
agkan import backup.json

Export behavior:

Outputs all tasks in JSON format to stdout
Redirect to a file for backup or migration

Import behavior:

Imports tasks from a JSON file created by agkan export
Preserves timestamps from the exported data
Automatically creates any tags that are missing in the target project
ID mapping is handled internally (new IDs are assigned in the target DB)

Use cases:

Task backup before destructive operations
Migrating tasks to another project
CI/CD pipeline task management

Process Operations

agkan ps                   # Show list of running Claude processes
agkan ps --port <number>   # Specify port
agkan ps --json            # Output in JSON format

Board Operations

agkan board start [--port <number>] [--title <title>]   # Start the board
agkan board stop                                         # Stop the board
agkan board restart [--port <number>]                   # Restart the board
agkan board status [--port <number>]                    # Check board status

Metadata Operations

# Set metadata
agkan task meta set <task-id> <key> <value>

# Get metadata
agkan task meta get <task-id> <key>

# List metadata
agkan task meta list <task-id>

# Delete metadata
agkan task meta delete <task-id> <key>

Priority

Task priority is managed with the native --priority flag:

# Set priority when creating a task
agkan task add "Title" --priority high

# Update priority on an existing task
agkan task update <id> --priority high
agkan task update <id> --priority ""  # Clear priority

Value	Meaning
`critical`	Requires immediate attention. Blocking issue
`high`	Should be prioritized
`medium`	Normal priority (default)
`low`	Work on if there is time

When to set priority: Priority is set during the planning phase (agkan-planning-subtask), at the same time the task is moved from backlog to ready. This is the responsibility of the planning skill. Skills that select tasks for execution (e.g., agkan-run) read this value to determine which task to work on next.

Tag Priority

When selecting or tagging tasks, use the following priority order:

Priority	Tag Name
1	bug
2	security
3	improvement
4	test
5	performance
6	refactor
7	docs

This is the canonical definition. All skills refer to this table.

JSON Output

Use the --json flag when machine processing is needed:

agkan task list --json
agkan task get 1 --json
agkan task count --json
agkan tag list --json

# Combine with jq
agkan task list --status ready --json | jq '.tasks[].id'

JSON Output Schema

`agkan task list --json`

{
  "totalCount": 10,
  "filters": {
    "status": "ready | null",
    "author": "string | null",
    "assignees": "string | null",
    "tagIds": [1, 2],
    "rootOnly": false,
    "priority": "critical | high | medium | low | null"
  },
  "sort": "id | title | status | priority | created_at | updated_at | null",
  "order": "asc | desc | null",
  "tasks": [
    {
      "id": 1,
      "title": "Task Title",
      "body": "Body | null",
      "author": "string | null",
      "status": "icebox | backlog | ready | in_progress | review | done | closed",
      "priority": "critical | high | medium | low | null",
      "parent_id": "number | null",
      "created_at": "2026-01-01T00:00:00.000Z",
      "updated_at": "2026-01-01T00:00:00.000Z",
      "parent": "object | null",
      "tags": [{ "id": 1, "name": "bug" }],
      "metadata": []
    }
  ]
}

Note: priority is a first-class column, not metadata. metadata holds arbitrary key/value pairs unrelated to priority (e.g. pr).

`agkan task get <id> --json`

{
  "success": true,
  "task": {
    "id": 1,
    "title": "Task Title",
    "body": "Body | null",
    "author": "string | null",
    "assignees": "string | null",
    "status": "icebox | backlog | ready | in_progress | review | done | closed",
    "priority": "critical | high | medium | low | null",
    "parent_id": "number | null",
    "is_archived": "boolean",
    "branch": "string | null",
    "model_planning": "string | null",
    "model_run": "string | null",
    "effort_planning": "string | null",
    "effort_run": "string | null",
    "created_at": "2026-01-01T00:00:00.000Z",
    "updated_at": "2026-01-01T00:00:00.000Z"
  },
  "parent": "object | null",
  "children": [],
  "blockedBy": [{ "id": 2, "title": "..." }],
  "blocking": [{ "id": 3, "title": "..." }],
  "tags": [{ "id": 1, "name": "bug" }],
  "comments": [],
  "metadata": []
}

`agkan task count --json`

{
  "counts": {
    "icebox": 0,
    "backlog": 0,
    "ready": 2,
    "in_progress": 1,
    "review": 0,
    "done": 8,
    "closed": 5
  },
  "total": 16
}

`agkan task find <keyword> --json`

{
  "keyword": "Search keyword",
  "excludeDoneClosed": true,
  "totalCount": 3,
  "tasks": [
    {
      "id": 1,
      "title": "Task Title",
      "body": "Body | null",
      "author": "string | null",
      "status": "ready",
      "priority": "critical | high | medium | low | null",
      "parent_id": "number | null",
      "created_at": "2026-01-01T00:00:00.000Z",
      "updated_at": "2026-01-01T00:00:00.000Z",
      "parent": "object | null",
      "tags": [],
      "metadata": []
    }
  ]
}

`agkan task block list <id> --json`

{
  "task": {
    "id": 1,
    "title": "Task Title",
    "status": "ready"
  },
  "blockedBy": [{ "id": 2, "title": "...", "status": "in_progress" }],
  "blocking": [{ "id": 3, "title": "...", "status": "ready" }]
}

`agkan task meta list <id> --json`

{
  "success": true,
  "data": [
    { "key": "pr", "value": "https://github.com/org/repo/pull/1" }
  ]
}

Note: priority is a first-class task field (see task list / task get schemas above), not stored via task meta. The example above uses pr to avoid implying otherwise.

`agkan tag list --json`

{
  "totalCount": 3,
  "tags": [
    {
      "id": 1,
      "name": "bug",
      "created_at": "2026-01-01T00:00:00.000Z",
      "taskCount": 2
    }
  ]
}

Typical Workflows

Icebox Review (agkan-icebox)

Icebox holds ideas and candidates that are not yet ready for planning. Review them periodically to decide whether to promote or close each one.

# Review icebox tasks
agkan task list --status icebox

# Promote to backlog when requirements become clear
agkan task update <id> status backlog

# Close if no longer needed
agkan task update <id> status closed

Icebox → Backlog conditions:

Requirements or background are now clear enough to plan
External blockers have been resolved
Circumstances have changed and the task is now relevant

Icebox → Closed conditions:

The need no longer exists
A duplicate already exists in a later stage
Superseded by another approach

Receiving Tasks as an Agent

# Check assigned tasks
agkan task list --status ready
agkan task get <id>

# Start work
agkan task update <id> status in_progress

# Complete
agkan task update <id> status done

Structuring Tasks

# Create parent task
agkan task add "Feature Implementation" --status ready

# Add subtasks
agkan task add "Design" --parent 1 --status ready
agkan task add "Implementation" --parent 1 --status backlog
agkan task add "Testing" --parent 1 --status backlog

# Set dependencies (Design → Implementation → Testing)
agkan task block add 2 3
agkan task block add 3 4

# View overall structure
agkan task list --tree

Body Conventions

PR Label

Skills that create PRs write a structured label into the task body. This label is used by subsequent skill executions to resume work on the existing PR instead of opening a new one.

Format

PR: <URL>

Rules

PR label is written by agkan-subtask after the PR is opened (Step 7).
The label is appended to the existing task body, separated by a blank line.
Skills that start work on a task must parse this label from the task body before creating a new PR.
- If a PR: label is present → push to the branch to update the existing PR instead of opening a new one.
- If no PR: label is present → open a new PR.
A new PR is always created as a draft (gh pr create --draft). It stays a draft while the task is in_progress; the skill marks it ready for review (gh pr ready) when the task advances to review.

Branch Field

The branch column is a first-class field on the task record — it is not stored in the task body.

Set: agkan task add --branch <name> or agkan task update <id> --branch <name>
Read: agkan task get <id> --json → .task.branch
Note: agkan task list --json does not include the branch field. Use agkan task get <id> --json when branch information is needed.
When branch is null, subtask skills (agkan-subtask, agkan-subtask-direct) auto-generate a branch name from the task ID and title, create the branch, and persist the name back via agkan task update <id> --branch <name>.

Configuration

Place .agkan.yml in the project root. Every key agkan reads:

agent: claude                   # AI coding agent to run: claude | codex | agy (default: claude)
path: ./.agkan/data.db          # SQLite DB path
board:
  port: 3000                    # Board server port
  title: My Project             # Board page title
models:
  planning:                     # Applies to the agent selected by `agent`
    model: opus
    effort: high
  run:
    model: sonnet
    effort: xhigh
  claude:                       # Per-agent settings: `claude` | `codex` | `agy`,
    planning:                   # each holding its own `planning` / `run`
      model: opus
      effort: high
    run:
      model: sonnet
      effort: xhigh
modelCatalog:                   # Replaces the built-in (cli, model, efforts) catalog wholesale
  - cli: claude
    model: opus
    efforts: [low, medium, high, xhigh, max]
permissionMode: auto            # Passed through to the agent CLI.
                                # `skipPermissions` maps to --dangerously-skip-permissions.

Every key is optional. The DB path can also be set with the environment variable AGENT_KANBAN_DB_PATH=/custom/path/data.db.

Config Commands

# Get all resolved config values from .agkan.yml
agkan config get

# Get a specific config value (dot notation)
agkan config get board.port
agkan config get models.planning.model

# Output in JSON format
agkan config get --json

agkan

agkan

Overview

Quick Reference

Agent Guide