T138 Imea Clawhub

整合三層記憶系統、自進化引擎與行動模式，提升 Agent 記憶完整性、自我改進及執行任務能力。

Things to check before installing: 1) Verify your workspace layout — the script expects files like SESSION-STATE.md and .learnings/* and the script uses memory/*.md, while docs reference memory-l1-daily/ etc; adjust paths or directories so the script searches the intended files. 2) Review any memory files for sensitive secrets before granting a hook that will read them (the script only reads local files with grep, but those files may contain confidential data). 3) Understand behavior on no-match: the script exits non‑zero when no records are found, which could cause automated tasks to stop if you wire it into a before_action hook — test in a safe environment. 4) The 'agent-memory-certificate' and CEO/monitor enforcement language are documentation/policy, not code that reaches out to external services; confirm you are comfortable adopting that policy and that you control any actions that report to higher authorities. 5) Confirm the other referenced skills (proactive-agent, self-improvement, etc.) are trusted and available if you intend to use them. 6) Backup your agent config.json before adding hooks and run manual tests of scripts/memory-check.sh to ensure expected behavior.

Type: OpenClaw Skill Name: t138-imea-clawhub Version: 1.0.0 The IMEA Skill bundle implements a structured three-layer memory system (L1/L2/L3) and a 'Write-Ahead Logging' (WAL) protocol for OpenClaw agents. It includes a shell script, scripts/memory-check.sh, which performs local searches of markdown files to identify past errors and learnings before the agent takes action. While the instructions in SKILL.md and config/agent-memory-certificate.md use strong prescriptive language to steer agent behavior (e.g., 'Mandatory', 'CEO requirements'), the logic is entirely focused on task consistency and error prevention, with no evidence of data exfiltration, malicious execution, or unauthorized remote access.