Briefing-First Startup

Make agents start fast without forgetting what matters.

This repository packages a reusable skill for designing low-bloat, high-recall session startup flows. It is meant for agent setups where full-memory preload has become slow, noisy, or brittle.

What problem this solves

Many agents degrade over time because session startup drifts into one of two bad extremes:

1. Load everything → huge prompts, slower reasoning, diluted focus
2. Load almost nothing → amnesia, repeated questions, broken continuity

This skill helps you build a third path:

Read a compact briefing first, scan active work, then expand context only when the current message actually requires it.

What is included

briefing-first-startup/
├── SKILL.md
├── references/
│   ├── startup-patterns.md
│   └── validation-checks.md
└── scripts/
    ├── briefing_lint.py
    └── scan_active_tasks.sh

Skill highlights

• Briefing-first startup instead of full-memory preload
• Active task scanning to recover current work quickly
• Message-type routing to decide what to load next
• On-demand expansion for daily memory, long-term memory, rules, and task files
• Validation checks so prompt slimming does not quietly reduce agent effectiveness

Good fit for

• OpenClaw / Codex / Claude Code style agent workspaces
• AI assistants that wake up fresh each session
• Long-running personal or project agents
• Teams trying to reduce prompt bloat without losing continuity

Quick install

Copy the skill folder into your workspace skills directory:

cp -R briefing-first-startup /path/to/your/skills/

Or clone this repo and reference the skill folder directly.

Quick use

1) Scan active tasks

bash briefing-first-startup/scripts/scan_active_tasks.sh /path/to/tasks

2) Check briefing quality

python3 briefing-first-startup/scripts/briefing_lint.py /path/to/BRIEFING.md

Core idea in one line

Use a short, curated startup briefing as the default context source; recover deeper context only when the incoming message proves you need it.

Why this looks simple but matters

Session startup is one of the easiest places for an agent to get slower while feeling safer. This repository is opinionated about avoiding that trap.

It is not a giant memory system. It is a practical startup discipline.

License

MIT