self-improving-agent
ReviewAudited by ClawScan on May 10, 2026.
Overview
The skill is coherent, but it asks the agent to store full errors and promote learnings into long-term agent memory/instruction files without clear review or redaction controls.
Install only if you want an agent memory/logging workflow. Before using it, decide where learnings may be written, review entries before promotion to MEMORY.md or agent instruction files, and redact secrets from errors or environment details.
Findings (4)
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.
Sensitive command output, API errors, local paths, or transient misunderstandings could be saved and reused in future sessions, affecting future agent behavior or exposing details in persistent context.
The skill tells the agent to store complete errors and environment details, then potentially promote repeated learnings into persistent agent instruction or memory files.
Error template: "完整报错文本" and "关键环境信息"; promotion targets: "AGENTS.md", "SOUL.md", "TOOLS.md", "MEMORY.md"
Require user confirmation before recording or promoting entries, redact secrets and private data, keep entries project-scoped, and periodically review or expire persistent memory files.
Enabled hooks may add persistent reminders to sessions and influence the agent’s priorities, though the content is disclosed and aligned with the skill purpose.
When the optional hook is enabled, it injects a reminder into the agent bootstrap context, which can steer the agent toward self-improvement logging.
event.context.bootstrapFiles.push({ path: 'SELF_IMPROVEMENT_REMINDER.md', content: REMINDER_CONTENT, virtual: true })Enable the hook only if you want this behavior, and disable or edit it if the reminder becomes intrusive.
Information logged in one session could be sent to another session if the agent follows this guidance.
The skill recommends sharing key learnings across sessions, which may include context from errors, corrections, or project-specific details.
在需要跨会话协作时，使用 `sessions_send` 分享关键学习。
Share only non-sensitive summaries, confirm the target session, and avoid sending secrets, private logs, or proprietary details.
Users may have less confidence that the reviewed package metadata matches the advertised registry entry.
The registry metadata and packaged internal metadata do not align, creating some ambiguity about the exact package identity/version.
Registry: slug self-improving-agent-next, version 2.0.1; _meta.json: slug self-improving-agent, version 1.0.11
Verify the publisher, version, and file contents before enabling hooks or copying scripts into global OpenClaw locations.