Multi Agent Daily Report

A local-first CLI that collects activity from multiple AI coding agents and engineering sources, then prepares structured context for Claude Code, Codex, Cursor, or any other agent to summarize into a daily report.

Design

• CLI collects, cleans, and normalizes data from local sources.
• Skill / Prompt guides the current agent to generate a human-readable daily report.
• No remote LLM calls in the default CLI path — lower cost, no vendor lock-in, better privacy.

Architecture

~/.claude/projects/  ──┐
~/.codex/            ──┤  Collectors     ┌──────────┐     ┌──────────────┐
Cursor workspace     ──┼──────────────►  │  JSON    │ ──► │  Markdown    │
Git repos            ──┘                 │ context  │     │  context     │
                                         └──────────┘     └──────┬───────┘
                                                                │ Agent reads
                                                                ▼
                                         ┌──────────────────────────────┐
                                         │  Final daily report (.md)     │
                                         │  + optional QQ notification   │
                                         └──────────────────────────────┘

Quick Start

pip install -e .
daily-report init
daily-report collect --date yesterday
daily-report render --date yesterday

Without pip install, run from the repository with:

PYTHONPATH=src python -m multi_agent_daily_report.cli collect --date yesterday
PYTHONPATH=src python -m multi_agent_daily_report.cli render --date yesterday

Use /multi-agent-daily-report in Claude (Slash Command)

This repo includes a Claude skill (see skills/SKILL.md). Once installed, you can type /multi-agent-daily-report ... in a Claude chat and have Claude run collection/rendering, read the context, and generate the final daily report.

1) Install the skill into ~/.claude/skills/

Recommended: symlink the skills/ directory so changes in this repo take effect immediately:

mkdir -p ~/.claude/skills
ln -snf "$(pwd)/skills" ~/.claude/skills/multi-agent-daily-report

Or copy it (if you prefer a pinned snapshot):

mkdir -p ~/.claude/skills/multi-agent-daily-report
cp skills/SKILL.md ~/.claude/skills/multi-agent-daily-report/SKILL.md

2) Use it in a Claude chat

Basic usage (no args defaults to yesterday):

/multi-agent-daily-report

Examples:

/multi-agent-daily-report yesterday
/multi-agent-daily-report today
/multi-agent-daily-report 2026-04-24
/multi-agent-daily-report 2026-04-24 --sources claude,codex
/multi-agent-daily-report yesterday --sources claude,codex,cursor,git --compact
/multi-agent-daily-report 2026-04-26 --compact --send qq

Arguments (kept consistent with skills/SKILL.md):

• date: today / yesterday / YYYY-MM-DD (defaults to yesterday)
• sources: --sources claude,codex,cursor,git (defaults to all enabled)
• compact: --compact (deduplicated/aggregated context; recommended)
• send: --send qq (send the final report file via QQ bot; requires notify.qq.* config)

3) Prerequisites

• The slash command expects the daily-report CLI to be available; if not, it will run pip install -e . in the project directory.
• Requires Python 3.10+. If system Python is older, the skill falls back to PYTHONPATH=src python -m multi_agent_daily_report.cli ....
• Outputs: context reports/YYYY-MM-DD_context.md, final report final_reports/YYYY-MM-DD.md.

4) Troubleshooting

• /multi-agent-daily-report not found in Claude: ensure ~/.claude/skills/multi-agent-daily-report/SKILL.md exists. If you used a symlink, verify it still points to this repo’s skills/ directory.
• daily-report: command not found: run pip install -e . in the repo root, then verify with command -v daily-report.
• Python version error (< 3.10): install/run with a Python 3.10+ interpreter (venv recommended), or use the module form PYTHONPATH=src python -m multi_agent_daily_report.cli ... with a compatible Python.

Sources

• Claude Code — session history from ~/.claude/projects/**/*.jsonl
• Codex — local history (history.jsonl) and thread records (state_5.sqlite)
• Cursor — workspace metadata and local file history under ~/Library/Application Support/Cursor/User
• Git — commit log from configured repos (auto-discovers repos under ~/work_space)

Collectors are intentionally conservative and skip missing sources.

Configuration

The CLI reads cfg/config.yaml first (project-level), falling back to ~/.config/multi-agent-daily-report/config.yaml.

daily-report init              # generate default config
daily-report init --path ~/.config/multi-agent-daily-report/config.yaml

Config key highlights:

Key	Description
sources.*.enabled	Toggle individual collectors
sources.*.path	Override default source paths
output.directory	Where reports are saved
output.timezone	Timezone for date filtering (default: Asia/Shanghai)
state.backend	sqlite (default) or mysql
notify.*	QQ official bot notification settings

See cfg/config.example.yaml for the full reference.

CLI Commands

Command	Description
daily-report init	Generate default config
daily-report collect --date <date>	Collect activities to JSON
daily-report render --date <date>	Render Markdown context
daily-report render --date <date> --compact	Render deduplicated/aggregated context
daily-report collect --sources claude,git	Collect from specific sources only
daily-report send --date <date> --channel qq	Send report file via QQ Bot

Dates support: today, yesterday, YYYY-MM-DD.

Recommended Workflow

# 1. Collect activities from all sources
daily-report collect --date yesterday

# 2. Render compact context
daily-report render --date yesterday --compact

# 3. Have your current agent read reports/YYYY-MM-DD_context.md, generate the final report → final_reports/YYYY-MM-DD.md

Skill Integration

The included skill (skills/SKILL.md) teaches any agent (Claude Code, Codex, Cursor) how to:

• Run the CLI commands to collect and render context
• Read the rendered context and extract technical signals
• Generate a well-structured daily report in Chinese or English
• Optionally send the report via QQ Bot

This means the same CLI can be reused across multiple agents — each just needs a thin skill/prompt wrapper.

Output Files

Directory	File	Description
reports/	YYYY-MM-DD.json	Raw activity data (collect)
reports/	YYYY-MM-DD_context.md	Agent-readable context (render)
final_reports/	YYYY-MM-DD.md	Final human-readable daily report (generated by agent)

Intermediate activity data is also stored in SQLite (~/.config/multi-agent-daily-report/drep.db) for cross-date queries.

Workday Scheduled Sending

The script scripts/claude-send-yesterday-qq.sh checks whether today is a Chinese workday before sending:

• Weekends and statutory holidays are skipped
• Adjusted workdays (e.g., Saturday makeup days) are treated as workdays

python3 scripts/workday.py should-send --date 2026-04-27
python3 scripts/workday.py previous-workday --date 2026-04-27

Calendar files are in cfg/calendar/.

Privacy & Security

• No remote model calls by default
• Only summary context is generated — no requirement to upload raw conversation logs
• The skill instructs agents to avoid verbatim sensitive prompts, preferring project-level summaries
• Secrets (QQ keys, DB passwords) are resolved from environment variables, not stored in committed config

Development

pip install -e ".[dev]"    # install with dev tools
ruff check src/ tests/      # lint
mypy src/                   # type check
pytest tests/ -v            # run tests

Limitations

• Claude and Codex local records are well-supported.
• Cursor is based on workspace metadata and file history — full Cursor Chat/Agent conversation coverage is not guaranteed.
• The current render output is "activity context", not a final report. Final reports should be generated by an agent following skills/SKILL.md.

License

MIT — see LICENSE.