Full run checklist.md tasks in Claude Code skill

Automatically execute tasks from checklist.md with state management and scheduled checking

Security Scan

The skill's code and installer are mostly project-local and avoid network/credential use, but there are clear mismatches and surprising behaviors (undeclared dependency on jq, a SessionStart hook and permission rule being added, and the task runner does not actually run commands), so review before installing.

Purpose & Capability

The skill claims to "automatically execute tasks from checklist.md", but the core execute_task() implementation simply echoes the task (no commands are executed). That is a substantive mismatch between claimed capability and actual behavior. Additionally, the README/SKILL.md do not declare that the installer requires jq, yet install/uninstall scripts fail without it.

Instruction Scope

The SKILL.md and installer instruct the agent/user to copy scripts into a project-local .claude/ directory, add a permission rule and a SessionStart hook to .claude/settings.local.json, and start a background monitor. Those actions stay within the project directory (as claimed) but the SessionStart hook will cause the monitor to auto-start in sessions where checklist.md exists — this is persistent behavior and may be surprising to users who don't expect a hook to run automatically. The installer and scripts reference the current working directory and will read/write project files (.claude-status.txt, .fullrun.log, .monitor.pid, checklist.md).

Install Mechanism

There is no external network download — scripts are copied from the included files into a project-local .claude/fullrun/scripts/ directory. This is low-risk in terms of external code fetching. However, the installer and uninstaller require jq (checked at runtime), which is not listed in the skill metadata; this undeclared dependency is an incoherence and will cause install/uninstall to fail if jq is missing. Also the installer embeds shell commands into settings.local.json (via jq), which warrants inspection before applying.

Credentials

The skill does not request credentials, environment variables, or external network access. All filesystem access is limited to the current project directory as described. No unrelated secrets or system-wide credentials are requested.

Persistence & Privilege

The skill does not set always:true and is project-local, but it does modify .claude/settings.local.json to add a permission pattern (Bash(.../scripts/*.sh)) and a SessionStart hook. That grants the platform permission to run the project scripts automatically at session start for that project — intended for this skill but potentially surprising. Uninstall does attempt to remove these changes.

What to consider before installing

Key things to check before installing: - Inspect the scripts yourself. The core runner (scripts/fullrun.sh) only echoes tasks (execute_task just prints), so it does not actually execute commands — if you expect it to run commands you need to modify/extend it. - The installer requires jq but the registry metadata does not declare this. Install jq (or review/modify the install/uninstall scripts) before running install.sh. - The installer will add a permission rule and a SessionStart hook to .claude/settings.local.json in the project directory. This is project-local persistence, but it will auto-start monitoring when a Claude Code session begins and checklist.md is present — remove/inspect the hook if you don't want auto-start behavior. - The mark-as-done implementation uses sed -i '' (BSD/macOS style); on many Linux systems that invocation will fail. Test the scripts on your target OS and adapt sed usage if necessary. - Run the installer and monitor first in a disposable project directory to observe behavior, and confirm uninstall.sh correctly reverts settings.local.json and removes copied scripts. - If you want real task execution, modify execute_task to run explicit, safe commands and validate input sanitization to avoid executing untrusted content from checklist.md.

Fullrun Skill

Overview

This skill implements a task execution system with the following features:

1. State Management: Uses .claude-status.txt file to track execution state

   • 0 = Idle, ready to execute (or file doesn't exist)
   • 1 = Claude is currently executing tasks
   • 2 = All tasks completed, monitoring should exit

2. Task List: Reads pending tasks from checklist.md

3. Scripts: Executable scripts in the scripts/ directory

File Structure

project/
├── SKILL.md                  # This skill definition file
├── checklist.md              # Task list file (create in your project)
├── .claude-status.txt        # Execution state file (auto-generated)
├── .fullrun.log              # Execution log (auto-generated)
├── .monitor.pid              # Monitor process PID (auto-generated)
├── .claude/
│   └── fullrun/
│       └── scripts/          # Installed scripts (auto-generated)
│           ├── main.sh       # Main entry point
│           ├── fullrun.sh    # Task execution script
│           └── cron-manager.sh # Cron job management script
└── scripts/
    ├── main.sh               # Main entry point (source)
    ├── fullrun.sh            # Task execution script (source)
    ├── cron-manager.sh       # Cron job management script (source)
    ├── install.sh            # Installation script
    └── uninstall.sh          # Uninstallation script

Usage

Installation (run once per project)

The installer creates project-local configuration in ./.claude/ - no global settings are modified.

# In your project directory, run:
./scripts/install.sh

What the installer does:

1. Creates .claude/ directory in your project
2. Copies scripts to .claude/fullrun/scripts/
3. Creates or updates .claude/settings.local.json with:
   • Permission rule for project scripts
   • SessionStart hook for auto-monitoring

Scope: Configuration is project-local only. Other projects are not affected.

Uninstall

./scripts/uninstall.sh

This removes:

• The .claude/fullrun/ directory
• Fullrun permission rule from .claude/settings.local.json
• Fullrun hook from SessionStart (preserves other hooks)
• .claude/settings.local.json if it becomes empty

Start scheduled monitoring (recommended)

./.claude/fullrun/scripts/main.sh start

Manually execute tasks

./.claude/fullrun/scripts/main.sh run

Check status

./.claude/fullrun/scripts/main.sh status

Stop monitoring

./.claude/fullrun/scripts/main.sh stop

Execution Rules

1. If .claude-status.txt does not exist or contains 0, start executing unfinished tasks from checklist.md
2. When executing tasks, set .claude-status.txt to 1
3. When all tasks are completed, set .claude-status.txt to 2
4. Scheduled task checks every 1 minute:
   • If status = 0 or file doesn't exist + pending tasks = start execution
   • If status = 1 = Claude is running, continue waiting
   • If status = 2 = All tasks completed, exit monitoring
5. Monitoring automatically exits when all tasks are completed (status=2)

Task Marking Format

In checklist.md, tasks use the following format:

• [ ] indicates incomplete
• [x] indicates completed

Example:

# Checklist

- [ ] Task 1: Complete a feature
- [ ] Task 2: Write tests
- [x] Task 3: Completed task

Notes

• Scheduled monitoring is session-level, stops when the current terminal session ends
• For persistent monitoring, use system cron or launchd
• The SessionStart hook checks for checklist.md in the current working directory at runtime
• Configuration is project-local via .claude/settings.local.json

Security Notes

What this skill accesses:

• Reads checklist.md in your current project directory
• Creates/reads .claude-status.txt in your current project directory
• Does NOT access the internet
• Does NOT request or transmit credentials

What the installer modifies:

• .claude/settings.local.json - Project-local settings (gitignored)
• .claude/fullrun/scripts/ - Project-local scripts

Persistence:

• The SessionStart hook runs at the start of each Claude Code session in this project
• It only starts monitoring if checklist.md exists and no execution is in progress
• Uninstall removes all modifications in the project