agent-activity-log

Agent Activity Log

Persistent agent memory for any project. Initializes a memory.md file and enforces the read-before-act, write-after-change discipline that makes multi-session agent work compound instead of reset.

See wiki/concepts/agent-activity-log.md for the full conceptual framework.

When to Use

• Setting up a new project for AI-assisted development
• Starting a session in a project that already has memory.md
• User asks "what did we do last time" or "what's the project history"
• User says "add agent memory", "set up logging", or "track changes"
• Any project where agents will work across multiple sessions

Bootstrap Protocol

When initializing memory.md in a new project:

1. Check if memory.md already exists

ls memory.md 2>/dev/null || echo "No memory.md found"

If it exists, read it and summarize recent activity to the user. Do not overwrite.

2. Create memory.md

# Agent Activity Log

> Persistent memory for AI agents working on this project. Agents must read this
> file at session start and append after every meaningful change.

## Convention

- **Read first.** Before making changes, read this file to understand project history.
- **Append after changes.** Log decisions, state transitions, and open threads.
- **Never delete.** This file is append-only.
- **Skip noise.** Don't log formatting fixes, import reordering, or trivial typos.

---

3. Add the read/write constraint to project rules

Append to the project's .cursorrules, AGENTS.md, or .cursor/rules/ (whichever the project uses for agent configuration):

## Agent Memory

This project maintains a persistent agent activity log at `memory.md`.

**On every session start:** read `memory.md` to understand what has been done,
what decisions were made, and what open threads exist.

**After every meaningful change:** append an entry to `memory.md` with the format:

    ## [YYYY-MM-DD HH:MM] action-type | Title

    **Changed:** files touched
    **Why:** rationale
    **Decision:** what was chosen and why (if applicable)
    **Open:** unfinished items or follow-ups (if any)

Log decisions, state transitions, bug fixes, and new features. Skip formatting
fixes, import reordering, and trivial typos.