Overkill Memory System

Provides a neuroscience-inspired 6-tier automated memory system with WAL protocol, semantic search, emotional tagging, and value-based retention for OpenClaw...

This package contains a complete, locally runnable memory system and many helper modules — but its registry metadata understates what it needs. Before installing or enabling it: 1) Review install.sh and every script that writes to crontab or ~/.openclaw to see what will be created and scheduled. 2) Search the code for references to ACC_MODELS, AGENT_ID, ChromaDB, Ollama, Redis, fd, rg, and any network endpoints (Supermemory) — these indicate credentials and binaries the system will use. 3) If you plan to run it, run inside a contained environment (container or isolated user account) first; do not blindly export API keys into your global environment. 4) If you only want limited features, ask the author for a minimal configuration or disable cron/background sync and cloud backup. 5) Prefer setting required env vars locally and narrowly, and avoid giving broad credentials (cloud backups, LLM keys) unless you trust the source. Finally, request the upstream source/homepage or a reproducible build/release — the package currently lists 'source: unknown' and no homepage, which makes verification harder.

Type: OpenClaw Skill Name: overkill-memory-system Version: 1.9.5 The skill is classified as suspicious due to multiple critical vulnerabilities, primarily prompt injection vectors and potential remote code execution (RCE). User-controlled input (e.g., `content`, `fact`, `description`, `context`, `mitigation`, `task`, `notes`, `action`) is written directly into various markdown and JSON files (`SESSION-STATE.md`, `CRON-INBOX.md`, `diary/*.md`, `git-notes/index.json`, `ERRORS.md`, `LEARNINGS.md`, `FEATURE_REQUESTS.md`, `acc-state.json`, `habits.json`, `internal_state.json`, `reflections/*.json`). These files are explicitly designed to be read and interpreted by the AI agent (e.g., `ACC_STATE.md` is 'for prompt injection'), creating direct prompt injection vulnerabilities. Additionally, `acc-error-memory/scripts/calibrate-patterns.sh` and `acc-error-memory/scripts/haiku-screen.sh` execute external LLM CLI commands via `subprocess.run` using the `ACC_MODELS` environment variable. If `ACC_MODELS` can be controlled by an attacker, this leads to RCE. While the skill's stated purpose is benign, these vulnerabilities present significant attack surfaces.