Memory System V2
PassAudited by VirusTotal on May 12, 2026.
Overview
Type: OpenClaw Skill Name: memory-system-v2 Version: 1.0.0 The skill bundle implements a file-based memory system for AI agents using bash and jq. All file operations are confined to a dedicated directory (`~/clawd/memory`), and JSON processing with `jq` is done securely using `--arg` and `--argjson` to prevent command injection. The `SKILL.md` includes instructions for the AI agent to use its memory system (`run memory_search on MEMORY.md + memory/*.md`), which is a form of prompt injection, but its clear intent is to guide the agent to utilize the skill's core functionality for self-management, not to perform unauthorized actions or exfiltrate data. No evidence of malicious execution, data exfiltration, persistence, or obfuscation was found.
Findings (0)
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.
Memories may contain sensitive or outdated information and can be recalled in later sessions, potentially shaping future answers.
The tool stores captured content in persistent local markdown files and a JSON index for later search and consolidation.
MEMORY_DIR="$HOME/clawd/memory" ... echo "$content" >> "$daily_file" ... '.memories += [$entry]'
Only capture information you are comfortable retaining, periodically review or delete files under $HOME/clawd/memory, and verify recalled memories against current facts.
If memories are stale, incorrect, or adversarially captured, they could steer answers in those topic areas.
The skill recommends a pre-answer recall rule for certain topics, making stored memory context part of the agent’s response workflow.
Before answering anything about prior work, decisions, dates, people, preferences, or todos: run memory_search on MEMORY.md + memory/*.md
Use memory recall as supporting context, not as authority over the user’s latest instructions or current project files.