Memory Self-Heal

General-purpose self-healing loop that learns from past failures, retries safely, and records reusable fixes.

This skill appears to do what it says, but it performs broad filesystem reads (memory/, tasks/, logs/, skills/*/SKILL.md) and appends entries to memory. Before enabling it: 1) Restrict scan roots to only the project-level memory/task directories you trust; avoid allowing it to read system-wide paths. 2) Ensure the agent enforces redaction rules (don’t include secrets/tokens in memory writebacks) and add explicit exclude patterns for known secret/config locations. 3) Limit the skill's write target (single memory file or dedicated folder) and review permissions so it cannot alter other skills or system files. 4) Test the skill in a sandboxed project to observe what it reads/writes. 5) If you need stronger guarantees, require user confirmation before any writeback or before scanning directories that might contain credentials. If you want a different assessment, provide the exact runtime root paths the agent will have access to and any platform-level sandboxing/config restrictions — that would raise confidence to high.

Type: OpenClaw Skill Name: memory-self-heal Version: 1.1.0 The skill bundle provides instructions for an AI agent to perform self-healing and diagnostic tasks. It instructs the agent to scan local directories like `memory/` and `tasks/` for error patterns using shell commands (`Get-ChildItem`/`rg` and `Select-String`). These operations are read-only and confined to expected diagnostic paths. Crucially, the `SKILL.md` includes explicit 'Safety Rules' instructing the agent to 'Never auto-run destructive operations without confirmation' and 'Never log secrets/tokens in memory files', which actively mitigate potential prompt injection risks and prevent malicious actions. There is no evidence of data exfiltration, malicious execution, persistence mechanisms, or obfuscation.