The sensei your agent never had. 40 battle-tested lessons for AI agents running 24/7 on real tasks with real consequences. Covers ethics, safety, messaging s...
Security Analysis
high confidenceThis is an instruction-only operational manual for agents; its requirements and instructions are consistent with that purpose, but it teaches agents to read/write workspace files, run recurring jobs, and self-update instructions — so you should review what permissions you grant to any agent that uses it.
The name and description match the content: a 40-lesson field guide for agent operations. The skill is instruction-only, declares no binaries, env vars, or installs, and the guidance it contains (memory, crons, messaging safety, code vs prompts) is coherent with an 'agent sensei' purpose.
The SKILL.md and the included field-guide direct agents to read and write workspace files (e.g., memory/*.md, bank/contacts.md), manage configurations, and set up self-improving cron-like processes. That is expected for an operational manual but grants the agent authority to persist state and modify local files — review these file paths and any execution steps before allowing an agent to act autonomously.
No install spec or code is included — this is instruction-only. That minimizes installer risk because nothing is written to disk by an installer and there are no third-party packages pulled in.
The skill declares no environment variables or credentials, which is proportionate. However, the guide recommends storing contacts, preferences, and other persistent records in workspace files; those files may contain sensitive data. The skill does not explicitly ask for secrets, but using it could lead an agent to read/write sensitive data if you allow filesystem or network access.
The skill does not request 'always: true' and allows normal model invocation. The guide encourages autonomous self-improvement (META files, crons, automated updates to instructions), which is consistent with its purpose but increases persistence and potential blast radius if the agent is granted broad permissions. Confirm any cron/setup actions and limit privileges accordingly.
Guidance
This is a coherent agent operations manual and not an obvious attempt to exfiltrate data. Before installing or enabling it for autonomous use: (1) review the field-guide to identify exact file paths it suggests creating or modifying; (2) restrict the agent's filesystem and network permissions so it cannot read unrelated sensitive files or phone/address books; (3) require human review for any outbound messages (the guide itself recommends drafts), and audit any cron/self-update actions before enabling them; (4) only enable autonomy for agents you trust, and consider running the guide in a sandboxed environment first. If you want, I can list every file path and action the guide recommends so you can map them to the permissions you will grant.
Latest Release
v1.0.0
Renamed from field-guide. 40 lessons, 28 rules, 12 sections — including fractal self-evolution and epistemic hygiene. The sensei your agent never had.
More by @globalcaos
Published by @globalcaos on ClawHub
