Training Manager

AdvisoryAudited by Static analysis on May 10, 2026.

Overview

Detected: suspicious.prompt_injection_instructions

Findings (2)

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.

NoteHigh Confidence

ASI02: Tool Misuse and Exploitation

What this means

Installing and using this skill can change how your agent behaves in later sessions.

Why it was flagged

The skill includes a writer that can modify core OpenClaw workspace files. The filename whitelist, overwrite protection, rate limiting, and injection checks make this purpose-aligned, but these are still important behavior-control files.

Skill content

Allowed: SOUL.md, AGENTS.md, USER.md, TOOLS.md, IDENTITY.md, MEMORY.md

Recommendation

Review proposed edits to core workspace files before accepting them, and keep backups before major changes.

NoteHigh Confidence

ASI06: Memory and Context Poisoning

What this means

Incorrect, sensitive, or overly broad training notes could persist and shape future agent responses.

Why it was flagged

The skill intentionally stores user preferences, corrections, behavioral rules, and facts into persistent memory/context files that may influence future agent behavior.

Skill content

corrections and preferences get categorized and logged automatically: Behavioral rules → `AGENTS.md` ... Facts → `MEMORY.md` or daily logs

Recommendation

Do not store secrets in training files, periodically review MEMORY.md and daily logs, and remove or correct stale instructions.

NoteMedium Confidence

ASI04: Agentic Supply Chain Vulnerabilities

What this means

You have less independent context for who maintains the code or where updates come from.

Why it was flagged

The registry metadata does not provide a clear source or homepage for a skill that ships executable bash scripts. The provided scripts are readable and local, but provenance is still limited.

Skill content

Source: unknown; Homepage: none

Recommendation

Inspect the included scripts before use and install updates only from a trusted ClawHub listing or verified repository.

Findings (2)

warn

suspicious.prompt_injection_instructions

Location: README.md:180
Finding: Prompt-injection style instruction pattern detected.

warn

suspicious.prompt_injection_instructions

Location: SKILL.md:295
Finding: Prompt-injection style instruction pattern detected.