v1.0.0

Proactive Self Improving

ReviewClawScan verdict for this skill. Analyzed May 1, 2026, 8:40 AM.

Analysis

The skill is coherent and instruction-only, but it asks the agent to persist learning records and automatically change future agent behavior, so it should be reviewed carefully before installation.

GuidanceInstall only if you want the agent to keep persistent learning files and potentially update long-term agent guidance. Before using it, set rules that require approval for promotions or skill creation, review diffs, and prevent secrets or sensitive command output from being written into .learnings/.

Findings (3)

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.

Abnormal behavior control

Checks for instructions or behavior that redirect the agent, misuse tools, execute unexpected code, cascade across systems, exploit user trust, or continue outside the intended task.

Rogue Agents

SeverityHighConfidenceHighStatusConcern

SKILL.md

同一模式出现 ≥3 次 → 触发自动晋升，写入永久文件 ... 创建 `skills/<skill-name>/SKILL.md`

The skill authorizes autonomous promotion into permanent agent files and creation of new skills based on recurrence thresholds, without an explicit human review gate.

User impactThe agent’s future behavior or installed skill set could change automatically based on past interactions, including mistakes or misleading feedback.

RecommendationKeep automatic records inside .learnings/ only, and require a user-approved diff before modifying AGENTS.md, TOOLS.md, SOUL.md, or creating new skills.

Cascading Failures

SeverityMediumConfidenceHighStatusConcern

SKILL.md

有没有什么值得其他 agent 也知道的？ ... 跨项目通用 ... 可提取为独立 skill

The instructions aim to reuse learned rules across agents, projects, and skills. If a wrong or poisoned learning is promoted, it can propagate beyond the original task.

User impactOne incorrect correction, bad tool result, or misleading task review could influence later sessions, other agents, or newly generated skills.

RecommendationScope learnings to the originating project, record provenance, validate promoted rules, and require manual approval before sharing them across agents or projects.

Sensitive data protection

Checks for exposed credentials, poisoned memory or context, unclear communication boundaries, or sensitive data that could leave the user's control.

Memory and Context Poisoning

SeverityMediumConfidenceHighStatusConcern

SKILL.md

### 错误信息 ... 实际的报错输出 ... ### 上下文 - 执行的命令/操作 - 输入参数 - 环境信息（如相关）

The skill tells the agent to persist detailed error context, including command output, inputs, and environment information. Those fields can contain sensitive data, and the artifact does not define redaction, retention, or approval boundaries.

User impactPrivate commands, tool outputs, paths, tokens, or environment details could be saved into local learning files and reused later.

RecommendationRequire user confirmation before saving detailed logs, redact secrets by default, limit what fields may be stored, and define retention or cleanup rules.