Back to skill

Security audit

Test B

Security checks for vulnerabilities and agentic risk

Overview

This skill is not malicious, but it broadly saves conversation and tool-output details into persistent memory without enough user control or redaction safeguards.

Install only if you are comfortable with the agent saving summarized task details across sessions. Avoid using it in conversations that may contain tokens, passwords, private keys, personal data, internal hostnames, or confidential configuration unless you manually review and redact entries before they are written.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Findings (2)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:41
Finding

Unfiltered Persistence of Potentially Sensitive Conversation Data

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:41-50 and SKILL.md:87-95
Vulnerability Type: Unfiltered plaintext persistence of sensitive information
Risk Level: Medium

Vulnerable Instructions

The following is an English translation of the relevant source instructions:

markdown
1. Extract information worth persisting from tool output and the conversation:
   - Newly discovered facts: addresses, configuration values, endpoints, and file paths
   - Decisions and reasons: why option A was selected instead of option B
   - Errors and solutions: records of encountered problems
   - User preferences: explicitly stated preferences or requirements
   - Task progress: completed and incomplete work

2. Append the extracted information to memory/YYYY-MM-DD.md
   - Use edit in append mode and never overwrite existing content
   - Attach a brief source description to every memory entry

The persistence rules further state:

markdown
- Never discard critical information that has not been recorded; saving too much is preferable to omitting anything.
- Append only: use edit to append to the memory file and never overwrite it.
- Must save: important configurations, addresses, endpoints, file paths, and error solutions.
- Recommended to save: decision reasons, user preferences, and task progress.

Technical Analysis

The Skill intentionally scans conversation history and tool results, then writes selected information to persistent, date-based memory files. This is its declared function, but the extraction policy is excessively broad and does not define safeguards for secrets or personal data.

Configuration values, endpoints, error output, and tool results frequently contain API tokens, credentials, authorization headers, cookies, private hostnames, private IP addresses, database connection strings, personal information, or portions of environment files. The instructions do not require: ...[truncated 1993 chars]

Remediation
View remediation

Remediation Suggestions

  1. Add an explicit denylist for passwords, API tokens, authorization headers, cookies, session identifiers, private keys, recovery codes, database URLs, .env contents, and personal data.
  2. Detect and redact common secret formats before presenting or writing memory entries.
  3. Display the exact proposed memory entries and require explicit user confirmation before persisting them.
  4. Replace the “save more rather than less” policy with data minimization: store only information required for the continuing task.
  5. Store references to secret locations rather than secret values, such as “credential is stored in the configured secret manager.”
  6. Require restrictive file permissions and prevent memory files from being exposed through shared directories or source control.
  7. Define retention and deletion procedures for dated memory files.
  8. Treat all tool output as potentially sensitive, particularly output from environment inspection, configuration files, HTTP clients, cloud tools, and authentication systems.

T08 · Insecure Dependencies

Warning
Location
README.md:39
Finding

Mutable and Unverified Remote Skill Installation

Content
View full analysis

Vulnerability Details

File Location: README.md:39-52
Vulnerability Type: Unpinned remote dependency installation
Risk Level: Medium

Vulnerable Instructions

bash
git clone https://github.com/wavmson/openclaw-skill-smart-compact.git \
  ~/.openclaw/skills/smart-compact
bash
mkdir -p ~/.openclaw/skills/smart-compact
curl -o ~/.openclaw/skills/smart-compact/SKILL.md \
  https://raw.githubusercontent.com/wavmson/openclaw-skill-smart-compact/main/SKILL.md

Technical Analysis

The documented installation methods retrieve Skill instructions from a mutable Git repository or directly from its mutable main branch. Neither method pins an immutable commit or release, and no checksum or cryptographic signature is verified.

A Skill file is interpreted as operational instructions by the Agent. Consequently, changing the remotely hosted SKILL.md can change effective Agent behavior even when no conventional executable script is downloaded. A compromise of the upstream account, repository, release process, DNS path, or hosting platform could therefore replace the reviewed content with malicious instructions.

The current audited artifact contains no confirmed malicious remote payload, and the commands do not pipe downloaded content directly into a shell. The vulnerability is the absence of integrity and immutability controls in the recommended installation process.

Attack Path

  1. An attacker compromises the upstream repository, maintainer account, or content publication process.
  2. The attacker modifies SKILL.md on the default branch to include malicious Agent instructions.
  3. A user follows the documented git clone or raw main download procedure.
  4. The mutable attacker-controlled version is written into the local OpenClaw Skill directory.
  5. The user restarts the Gateway as instructed by the installation guide.
  6. OpenClaw loads the modified Skill instructions.
  7. When in ...[truncated 670 chars]
Remediation
View remediation

Remediation Suggestions

  1. Publish versioned releases and recommend installation from a specific immutable release tag and commit.
  2. Pin raw downloads to a full commit hash instead of the mutable main branch.
  3. Publish SHA-256 checksums or cryptographic signatures through a separately protected release channel.
  4. Require users or the installer to verify the checksum or signature before placing the Skill in the active Skill directory.
  5. Download into a staging directory, display the content for review, and activate it only after successful verification.
  6. Document a trusted update process that shows content differences before replacing an installed Skill.
  7. Protect the upstream repository with multi-factor authentication, branch protection, signed commits or releases, and restricted publication permissions.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (7)

Ssd 3

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

This section directs the agent to extract and append a wide range of information—facts, configuration values, endpoints, file paths, errors, preferences, and task progress—from both tool output and conversation into dated memory files. Because no minimization, redaction, consent gating, or data-classification controls are required, the skill can persist sensitive user information and operational details long after the original context, making accidental disclosure or later misuse more likely.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The heading '中文说明' indicates the skill documentation is presented in Chinese only, and no alternative language option or opt-in is provided in the visible content. This can violate language/locale policy requirements when a skill implicitly forces a specific language without user choice.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

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.

Content

Scanner excerpt · README.md (reported line 49)May include surrounding context.

方式三:只复制核心文件

bash
mkdir -p ~/.openclaw/skills/smart-compact
curl -o ~/.openclaw/skills/smart-compact/SKILL.md \
  https://raw.githubusercontent.com/wavmson/openclaw-skill-smart-compact/main/SKILL.md

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · README.md (reported line 50)May include surrounding context.

bash
mkdir -p ~/.openclaw/skills/smart-compact
curl -o ~/.openclaw/skills/smart-compact/SKILL.md \
  https://raw.githubusercontent.com/wavmson/openclaw-skill-smart-compact/main/SKILL.md

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The trigger list includes generic phrases like 'pre-compact' and 'compact check,' which may match ordinary conversation rather than a deliberate request to invoke the skill. Ambiguous triggers increase the risk of accidental execution of scanning and persistence behavior, especially because the skill processes prior tool outputs and writes memory files.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill explicitly recommends proactive or automatic suggestion based on context fullness and heartbeat signals, which creates overly broad invocation boundaries for a state-changing workflow. Even though the action is framed as a suggestion, autonomous activation guidance can cause the agent to run a memory-persistence routine in situations the user did not clearly request, increasing the chance of unintended data retention or workflow interference.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The instruction to 'save before compressing' and prefer over-retention encourages broad extraction of conversation and tool-derived content into persistent memory without any sensitivity screening. In this skill's context, tool outputs may contain secrets, personal data, credentials, internal paths, or operational details, so blanket retention materially raises confidentiality and compliance risk.

Content

No source excerpt is available for this finding.