claude-cron-automation

name	claude-cron-automation
description	Run Claude Code commands (including /insights, /doctor, custom prompts) via system crontab in non-interactive mode. Use when: (1) setting up daily/weekly automated Claude Code tasks, (2) scheduling reports or analysis via cron, (3) running Claude Code with third-party API backends (DeepSeek, Qwen, etc.) in headless environments, (4) "set up a daily report", "automate this in cron", "schedule insights". Covers: cron PATH setup, third-party API env vars for non-interactive mode, nvm/node auto-discovery, and wechat-reminder integration for notification delivery.
author	Jing Mai
version	1.0.0
date	"2026-07-27T00:00:00.000Z"

Claude Code Cron Automation

Problem

Running Claude Code in cron requires solving three challenges that don't apply to interactive sessions:

1. cron's minimal PATH doesn't include nvm/node or user binaries
2. Shell aliases and functions (like dsccpro) are unavailable
3. Third-party API backends need explicit environment variables

Context / Trigger Conditions

• Setting up automated daily/weekly Claude Code tasks via cron
• claude -p non-interactive mode with API backends other than Anthropic
• /insights, /doctor, or custom prompts need to run on a schedule
• cron environment lacks DEEPSEEK_API_KEY, node, or wechat-reminder in PATH
• Need to deliver results via wechat-reminder (Feishu/WeChat)

Solution

1. Crontab Entry Pattern

Always use zsh -l (login shell) to source the user's full environment:

7 22 * * * /bin/zsh -l '/absolute/path/to/script.sh'

The -l flag ensures .zshenv, .zprofile, and .zshrc are sourced, making API keys and PATH available.

2. Script Self-Setup (Defense in Depth)

Even with zsh -l, include a self-setup section in the script for robustness:

#!/bin/zsh
set -euo pipefail

# Auto-discover nvm/node (covers cron's minimal PATH)
for nvm_dir in "$HOME/.nvm" "/usr/local/nvm"; do
    if [[ -s "$nvm_dir/nvm.sh" ]]; then
        export NVM_DIR="$nvm_dir"
        break
    fi
done

NODE_BIN="$HOME/.nvm/versions/node/$(ls "$HOME/.nvm/versions/node/" 2>/dev/null | sort -V | tail -1 || echo '')/bin"
export PATH="$NODE_BIN:$HOME/.local/bin:/usr/local/bin:/usr/bin:/bin:$PATH"

# Verify essentials
command -v claude >/dev/null 2>&1 || { echo "claude not found"; exit 1; }

3. Third-Party API Env Vars (Non-Interactive)

Shell aliases like dsccpro don't work in scripts. Use the raw env var pattern that _claude_with_api expands to:

# DeepSeek API (replaces dsccpro alias)
REPORT=$(
    env -u ANTHROPIC_API_KEY \
        ANTHROPIC_BASE_URL="https://api.deepseek.com/anthropic" \
        ANTHROPIC_AUTH_TOKEN="${DEEPSEEK_API_KEY}" \
        API_TIMEOUT_MS=600000 \
        ANTHROPIC_MODEL="deepseek-v4-pro[1m]" \
        ANTHROPIC_DEFAULT_SONNET_MODEL="deepseek-v4-pro[1m]" \
        ANTHROPIC_DEFAULT_OPUS_MODEL="deepseek-v4-pro[1m]" \
        ANTHROPIC_DEFAULT_HAIKU_MODEL="deepseek-v4-pro[1m]" \
        ANTHROPIC_SMALL_FAST_MODEL="deepseek-v4-pro[1m]" \
        CLAUDE_CODE_SUBAGENT_MODEL="deepseek-v4-pro[1m]" \
        CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1 \
        CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 \
        claude -p "$PROMPT" \
            --print \
            --output-format text \
            --dangerously-skip-permissions \
            --add-dir "$HOME/.claude/projects" \
            --add-dir "$HOME/.claude/usage-data" \
            2>&1
)

Key flags:

• --dangerously-skip-permissions — required for non-interactive; skips all permission prompts
• --add-dir <path> — explicitly allow file access to specific directories
• --print --output-format text — non-interactive text output (no TUI)

4. Built-in Commands in Non-Interactive Mode

These Claude Code built-in commands work with claude -p:

Command	Behavior in -p mode
/insights	Generates HTML report at ~/.claude/usage-data/report-YYYY-MM-DD-HHMMSS.html
/doctor	Outputs diagnostic report to stdout
/context	Outputs context usage to stdout
Custom prompt	Full tool access (Read, Bash, etc.) for multi-step workflows

After running /insights, use Read to access the generated HTML report.

5. Multi-Step Prompt Design

Combine steps into a single -p call to avoid multiple API invocations:

你是一个[角色描述]。请严格按以下步骤操作：

## 步骤1: [描述]
[具体指令]

## 步骤2: [描述]
运行 /insights 命令，然后用 Read 工具读取最新的 report HTML。

## 步骤3: 输出报告
[输出格式要求]

### 风格要求
[约束条件]

Verification

1. Run the script manually first: zsh -l ./script.sh
2. Check the log: tail -f ~/.claude/logs/<script>.log
3. Verify the notification arrived (Feishu/WeChat)
4. Only then install the crontab

Example

See self-reflection/daily-insights.sh in this dotfiles repo for a complete working example that:

• Scans today's memory files
• Runs /insights
• Generates a structured daily reflection report
• Sends via wechat-reminder

Notes

• macOS cron caveat: On macOS, cron needs Full Disk Access permission. System Settings → Privacy & Security → Full Disk Access → add Terminal.app
• Avoid :00 and :30 minutes in cron schedules to reduce API fleet contention
• Log everything: cron's stdout/stderr can be lost; always write to a log file
• Cross-platform date/find: Use a temp reference file with touch -t (macOS BSD) or touch -d (Linux GNU) rather than -newermt for cross-platform find -newer
• Idempotent install: Write an install.sh that checks for existing crontab entries before adding, and supports --uninstall and --status
• Prompt passing: Use a temp file to build multi-line prompts rather than complex shell quoting

References

• Claude Code Commands
• User's _claude_with_api function in common_shell_setup.sh
• claude-end-reminder.sh hook for wechat-reminder integration pattern