task-tracking（任务跟踪）

This skill enables agents to manage tasks using the bundled bash script task_tracking.sh, located in the skill's scripts folder.

Prerequisites

Requires jq for JSON processing. Install via your system package manager if not already available.

How It Works

The script stores task data in a .tasks directory. It automatically finds the correct location by:

1. Walking up from your current directory to find an existing .tasks directory
2. If not found, creates .tasks in your current directory

This means:

• ✅ Works from any subdirectory once .tasks exists
• ✅ No need to remember or navigate to workspace root
• ⚠️ Initial setup: be in your desired workspace root when creating the first task list

Discovery: Run bash <path-to-skill>/scripts/task_tracking.sh --help to see all available commands.

Quick Start

Note: In all examples below, replace <path-to-skill> with the absolute path to where this skill is loaded from (the directory containing this SKILL.md file).

• Create a task list: bash <path-to-skill>/scripts/task_tracking.sh create-list <list-name>
• List available task lists: bash <path-to-skill>/scripts/task_tracking.sh list-lists
• Add a task: bash <path-to-skill>/scripts/task_tracking.sh add-task <list-name> "<description>" [--file <path>]... [--doc <path>]... [--skill <name>]... [--ac "<criterion>"]... [--verify-command "<cmd>"] [--verify-instruction "<text>"] [--depends-on <task-id>]...
• Claim or get next task: bash <path-to-skill>/scripts/task_tracking.sh next <list-name> [--skip-failed] [--claim <AGENT_ID>]
• Update task status: bash <path-to-skill>/scripts/task_tracking.sh update-status <list-name> <task-id> <status> "<note>"
• Update task metadata: bash <path-to-skill>/scripts/task_tracking.sh update-task <list-name> <task-id> [--description "<text>"] [--file <path>]... [--doc <path>]... [--skill <name>]... [--ac "<criterion>"]... [--verify-command "<cmd>"] [--verify-instruction "<text>"] [--depends-on <task-id>]...
• Add a dependency: bash <path-to-skill>/scripts/task_tracking.sh add-dependency <list-name> <task-id> <depends-on-task-id>
• Remove a dependency: bash <path-to-skill>/scripts/task_tracking.sh remove-dependency <list-name> <task-id> <depends-on-task-id>
• List all tasks in a list: bash <path-to-skill>/scripts/task_tracking.sh list-tasks <list-name> [--format summary|full|json]
• Query/filter tasks: bash <path-to-skill>/scripts/task_tracking.sh query <list-name> [--status <status>] [--search <term>] [--depends-on <task-id>] [--blocked] [--claimed-by <agent-id>] [--limit <n>]
• Count tasks: bash <path-to-skill>/scripts/task_tracking.sh count <list-name> [--status <status>] [--exclude-status <status>]...

Script Reference

Run bash <path-to-skill>/scripts/task_tracking.sh --help for complete CLI documentation.

Key concepts:

• The script automatically finds or creates the .tasks directory by walking up from your current location. All operations are performed via the CLI interface.
• Task statuses: todo, in_progress, completed, failed, skipped

- completed and skipped are terminal — next will not pick them up, and they satisfy dependency requirements for downstream tasks - skipped means "won't implement" — use update-status <list> <id> skipped "<reason>" to mark a task as deliberately bypassed

• update-status changes status and writes a log entry. Both <status> and <note> are positional and always required.
• update-task edits metadata only — description, context, acceptance criteria, verification, dependencies. It never changes status or writes a log entry. Use it to fix typos, add context, or clarify scope at any time.
• Each task has a context object with files, docs, and skills arrays (via repeatable --file, --doc, --skill flags). All are optional and default to empty arrays.
• Tasks may include acceptance_criteria (list of strings, via repeatable --ac) and a verification object (via --verify-command or --verify-instruction). All are optional.
• --verify-command stores {"type":"command","value":"..."} — a shell command the agent can execute.
• --verify-instruction stores {"type":"instruction","value":"..."} — a free-text instruction for manual verification.
• Only one of --verify-command or --verify-instruction may be specified per task.
• Metadata arrays (--file, --doc, --skill, --ac) are replaced, not appended, when passed to update-task.

Listing, Querying, and Counting Tasks

• list-tasks <list> [--format summary|full|json] — Lists all tasks in a list.

- summary (default): returns [{id, description, status}] from the index — fast, no file reads - full: returns full task objects in order, reading each per-task file - json: returns the raw _task-list.json index

• count <list> [options] — Returns a single integer: the number of tasks matching the filters. Reads only the index file (fast). Supports --status <status> and --exclude-status <status> (repeatable; excludes each listed status).
• query <list> [options] — Filters tasks matching ALL provided criteria. Returns [{id, description, status, claimed_by, depends_on}]. Examples: # All todo tasks bash <path-to-skill>/scripts/task_tracking.sh query my-list --status todo # Search for auth-related tasks bash <path-to-skill>/scripts/task_tracking.sh query my-list --search "auth" # Tasks blocked by task-01 being incomplete bash <path-to-skill>/scripts/task_tracking.sh query my-list --depends-on task-01 # All blocked tasks bash <path-to-skill>/scripts/task_tracking.sh query my-list --blocked # Tasks claimed by a specific agent bash <path-to-skill>/scripts/task_tracking.sh query my-list --claimed-by agent-123 # Combine filters: in-progress tasks claimed by agent-123 bash <path-to-skill>/scripts/task_tracking.sh query my-list --status in_progress --claimed-by agent-123

- --status <status> — match exact status (todo, in_progress, completed, failed) - --search <term> — case-insensitive substring match in description, acceptance criteria, or notes - --depends-on <task-id> — tasks that list the given ID in their depends_on - --blocked — non-completed tasks with at least one non-completed dependency - --claimed-by <agent-id> — tasks claimed by the specified agent - --limit <n> — cap results at N (default: 0 = unlimited)

Dependency Tracking

Tasks can declare dependencies on other tasks via depends_on (a list of task IDs).

• next --claim rejects a task if any of its depends_on tasks are not completed. Complete blocking tasks first.
• update-status... completed prints a warning listing any pending tasks that depend on the one being completed, so you know what is now unblocked.
• Use --depends-on <task-id> (repeatable) with add-task or update-task to set/replace the full dependency list.
• Use add-dependency / remove-dependency for surgical edits to an existing task's dependencies.
• Circular dependencies are not auto-detected; avoid them when designing task lists.

Error Handling

CRITICAL: If the task tracking script returns an error or fails to execute:

1. IMMEDIATELY notify the user about the error and include the full error message
2. DO NOT attempt to modify task files manually (e.g., editing .tasks/ directory JSON files directly)
3. DO NOT work around the script by using jq, cat, or other tools to manipulate task data
4. WAIT for explicit user instruction before taking any alternative action

The task tracking system is designed to maintain data integrity through the script interface only. Manual modifications can corrupt the task list, status logs, or task metadata.

Data location: The script automatically finds the .tasks directory by walking up from your current location, or creates it in the current directory if not found.

Best Practices

Initial Setup

• Create your first task list from your intended workspace root directory
• The .tasks directory will be created there
• Subsequent commands work from any subdirectory

Task management

• Always provide meaningful context for each task
• Use agent IDs to claim tasks for specific agents
• Log progress and errors using the update-task command with notes
• If the script fails, stop and notify the user - do not attempt manual workarounds

After completing a task

• Provide a git commit message summarizing the work done for the user to use when committing changes

Writing Self-Sufficient Tasks

When creating a task, populate enough context for a zero-context agent to succeed without follow-up questions:

Description: State the current state and desired outcome, not just the task name.

• Poor: "Fix free book button"
• Good: "Free products show 'Add to Cart' — change button label to 'Get Free' for products with price = 0"

Before running add-task, search the codebase to identify relevant files, then include them via --file.

Acceptance criteria (--ac) should describe observable outcomes, not restate the description.

Verification: Use --verify-command for automated checks, --verify-instruction when human judgment is needed.