Transform AI agents from task-followers into proactive partners that anticipate needs and continuously improve. Now with WAL Protocol, Working Buffer, Autonomous Crons, and battle-tested patterns. Part of the Hal Stack 🦞
Security Analysis
medium confidenceThe skill's files and instructions mostly match a proactive-agent purpose, but there are conflicting guardrails and a few behaviors (reading system/workspace config, tailing logs, and pre-scan prompt-injection strings present in SKILL.md) that merit caution before installing or enabling autonomously.
The name/description (proactive, stateful agent) lines up with the artifacts: workspace files, WAL/working-buffer protocols, and a security-audit script. Nothing requests unrelated cloud credentials or external services. Note: AGENTS.md contains the phrase "Don't ask permission. Just do it." which conflicts with other guardrails and could escalate autonomous actions if followed.
The SKILL.md and assets instruct the agent to read/write many workspace files (SESSION-STATE.md, MEMORY.md, ONBOARDING.md, AGENTS.md, memory/ working-buffer), scan logs (tail /tmp/clawdbot/*.log), and check $HOME/.clawdbot/clawdbot.json. Reading and updating workspace files is expected, but tailing system/log paths and inspecting user home config is broader than a simple 'helper' and may expose sensitive context. There are also contradictory instructions across files: most files emphasize requiring human approval for external actions, but some statements (e.g., "Don't ask permission. Just do it.") remove that gate — this inconsistency is an operational risk.
No install spec and only one simple shell audit script included. There are no downloads or extract steps and no external packages being pulled in. Instruction-only skills + a local script minimize supply-chain risk compared with arbitrary remote installs.
The skill declares no required environment variables or credentials. The included security patterns and script check local .credentials and .gitignore; those checks are reasonable for a self-auditing agent and proportional to the stated purpose.
The skill does not request always:true and is user-invocable only. It instructs persistent behavior (writing SESSION-STATE.md, working buffers, memory files) which is consistent with a stateful agent. However, the mixed messaging about when to seek human approval could grant the agent broader autonomy in practice — combine this with the platform-default autonomous invocation and the agent could act without the explicit manual gating the user expects.
Guidance
What to check before installing/using this skill: - Inspect the script: open scripts/security-audit.sh and confirm it only reads local files and performs benign checks (it does). Run it in a sandbox first to see what it flags. - Review AGENTS.md and SKILL.md for contradictory rules. The skill mostly enforces human approval for actions that leave the machine, but one place says "Don't ask permission. Just do it." Decide which policy you want and edit the files to enforce explicit gating if you prefer. - Confirm scope of file access: this skill expects to read and write many workspace files and may read clawdbot configs and /tmp logs; if you keep secrets or sensitive data in the workspace or in .clawdbot, consider isolating the skill in a disposable workspace. - Test in an isolated environment first (no network, test account, or VM) to observe any autonomous behavior before granting it in your primary environment. - Because the SKILL.md contains prompt-injection example phrases, ensure those are only used as detection examples; search for any lines that might cause the agent framework to treat them as active instructions. - If you want to let the agent act autonomously, explicitly audit and tighten approval gates (especially for external actions, sends, deletions, or credential use). If you want, I can produce a short checklist of specific lines to search/replace in these files to harden the skill before enabling it.
Latest Release
v1.0.0
- Initial public release of Joko Proactive Agent. - Rebranded skill (new slug, name, and author). - Introduced proactive, self-improving agent architecture featuring WAL Protocol, Working Buffer, Autonomous Crons, and best practices from the Hal Stack. - Comprehensive documentation included for setup, memory management, and proactive behavior patterns.
More by @oyi77
Published by @oyi77 on ClawHub
