T02 · Agent Memory Poisoning
- Location
SKILL.md:29- Finding
Persistent Modification of Agent Governance and Memory Files
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This skill openly sets up or audits persistent OpenClaw workspace governance files, with no evidence of hidden execution, exfiltration, or system-level persistence.
Install only if you want this skill to create or update persistent governance and memory files in an OpenClaw workspace. Before applying it to an existing workspace, review the proposed AGENTS.md and MEMORY.md changes, keep a backup, and verify any missing template content because the referenced references/ templates are not included in the artifact inspected here.
SKILL.md:29Persistent Modification of Agent Governance and Memory Files
The main skill instructions from L013 onward are written entirely in Chinese, while the front-matter metadata is in English. This creates a language/locale constraint for users without any opt-in, alternative language, or stated region-specific requirement, which matches the policy-violation criteria for forced language use.
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.
WS="$1"
mkdir -p "$WS/memory"
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.
WS="$1"
mkdir -p "$WS/memory"