Agent Memory Tools
Security checks across malware telemetry and agentic risk
Overview
This looks like a purpose-aligned memory/indexing tool, but it can persistently cache workspace content and can use cloud services if you configure them.
Install only if you are comfortable with the tool indexing and caching your workspace notes. For local-only use, keep the Ollama preset, leave convexUrl and cloud API keys unset, set MEMORY_WORKSPACE to a specific folder, and prefer --scan before enabling --watch or OS-level auto-triggers.
VirusTotal
No VirusTotal findings
Risk analysis
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.
Private or untrusted workspace notes can become long-lived agent memory and influence future answers.
The auto-ingest path intentionally reads workspace Markdown, extracts facts, and stores them for later memory recall.
WATCH_DIRS = ["memory", "agents", "projects", "docs", "notes"] ... facts = extract_facts(context, cfg, debug) ... stored = store_to_convex(facts, agent=agent, debug=debug)
Set MEMORY_WORKSPACE narrowly, review ignore patterns, inspect or delete the .cache outputs when needed, and avoid ingesting sensitive or untrusted documents.
If enabled, the memory index can continue updating in the background as workspace files change.
The skill documents a continuous watcher mode that keeps processing file changes after it is started.
python3 scripts/auto_ingest.py --watch # Daemon: poll for changes every 30s
Use one-shot --scan mode unless you explicitly want continuous ingestion, and remove any LaunchAgent/systemd/Task Scheduler entry when you no longer need it.
Facts extracted from workspace documents may leave the local machine if the cloud backend is enabled.
When a Convex URL is configured, stored facts are sent to the configured Convex backend instead of only local JSON storage.
result = _convex_call("agentMemory:store", {"fact": fact, "category": category, "agent": agent, "confidence": confidence, "source": source}, is_mutation=True, convex_url=convex_url)Leave convexUrl unset for local-only operation, and only configure a trusted Convex endpoint for data you are comfortable syncing.
If you enable cloud presets, the agent may use your provider account and incur usage or data-sharing implications.
The default path needs no credentials, but optional cloud presets require API keys.
| `openai` | gpt-4o-mini | text-embedding-3-small | OPENAI_API_KEY | ... | `openrouter` | gemma-3-4b:free | text-embedding-3-small | OPENROUTER_API_KEY + OPENAI_API_KEY |
Only set these API keys when you intentionally want cloud models, and prefer scoped or separate keys where possible.
Running setup delegates installation trust to the remote Ollama install script.
On Linux, the optional setup script executes Ollama's remote installer if Ollama is missing.
curl -fsSL https://ollama.com/install.sh | sh
Run setup.sh only if you trust the source, or install Ollama manually using your preferred package manager before running the skill.