Agent Memory
ReviewAudited by ClawScan on May 1, 2026.
Overview
This skill is coherent and local-only, but it intentionally creates persistent agent memory that can shape future sessions and be shared across agents.
Install only if you want the agent to maintain local long-term memory. Keep sensitive details out of memory files, review shared multi-agent memory before relying on it, and treat exported memory ZIPs as private data.
Findings (3)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
Memory entries may influence later answers and can retain private work or preference information across sessions.
The skill is designed to persistently store and reuse learned preferences/corrections, which is purpose-aligned but can preserve mistaken or sensitive context until the user removes it.
HOT — hot.md (≤100 lines, always loaded) ... Delete | Never (unless user says "forget X")
Review memory files periodically, avoid storing sensitive information, and use the documented 'forget' or 'forget everything' controls when needed.
A memory entry created by one agent could affect another agent's behavior in later tasks.
The multi-agent design intentionally shares domain and project memory across agents, which can propagate useful context but also propagate incorrect or overly sensitive entries.
domains/ # SHARED across all agents ... projects/ # SHARED, loaded by context
Keep shared domain/project files limited to non-sensitive team knowledge and review them when adding or changing agents.
Running the helper will create or modify local memory files in the selected directory.
The included bootstrap script creates local directories and memory template files, and can overwrite them if explicitly run with --force. This is disclosed setup behavior and is scoped to the memory directory by default.
DEFAULT_ROOT = os.path.expanduser("~/agent-memory") ... overwrite = "--force" in sys.argv ... path.write_text(content, encoding="utf-8")Run the bootstrap only from the reviewed skill directory, use the default path or a path you intentionally choose, and avoid --force unless you have backed up existing memory files.