From CLI Harness to Product

PROJECT CONTEXT:
I have an agent loop that works in a script. The gap between that and something people adopt - session management, config, resume, distribution - is exactly the gap the polished terminal agents crossed.

GOAL:
Productize my agent harness: named sessions with resume, layered configuration, first-run experience, and clean packaging.

REQUIREMENTS:
- Persistent sessions: every run saved, listable, and resumable by ID
- Layered config: defaults < config file < environment < flags
- A first-run flow that validates credentials and writes a starter config
- Clean CLI surface: run, resume, list, config commands with helpful errors
- Distribution as an installable package with a versioned changelog

CONSTRAINTS:
- Sessions must store enough to truly resume (history, todo state, cwd), not just a chat log
- Secrets never written into session files or logs

DEFINITION OF DONE:
- `agent resume <id>` continues a previous task with full context restored
- Config precedence works and `agent config` shows the effective merged values
- A new user reaches a successful first run in under 2 minutes
- The package installs cleanly on a fresh machine and reports its version

COMMON FAILURES TO AVOID:
- Resume that replays a chat log but loses tool state and working directory
- Config spread across hardcoded values nobody can override
- First run failing cryptically on a missing API key
- Session files leaking API keys into plaintext on disk