Security audit

Soul Pack

Security checks for vulnerabilities and agentic risk

Overview

The skill is purpose-aligned, but importing packages can install unreviewed agent instructions and handles archives/overwrites too loosely, so it needs Review before use.

Install only if you trust the SOUL packages you will import. Review a package's SOUL.md before importing, avoid --force unless you have backed up the workspace, use a test agent first, and avoid sharing exported archives if the embedded local workspace path is sensitive.

A.I.G

Vulnerability Patterns

Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
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
Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally

Findings (3)

T01 · Skill Instruction Hijacking

Error

Location: scripts/import-soul.sh:42
Finding: Untrusted SOUL instructions are installed and registered without security review
Content: View full analysis
Remediation: View remediation

T09 · Insecure Skill Coding Practices

Warning

Location: scripts/export-soul.sh:20
Finding: Unvalidated package name permits output path traversal and tar option injection
Content: View full analysis
"$PKG_DIR/preview.md" < "$PKG_DIR/manifest.json" <
Remediation: View remediation

T09 · Insecure Skill Coding Practices

Error

Location: scripts/import-soul.sh:27
Finding: Untrusted packages are extracted and copied without archive safety controls or full schema validation
Content: View full analysis
Remediation: View remediation

SkillSpector

Vulnerability Patterns

MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (4)

Tp4

High

Category: MCP Tool Poisoning
Confidence: 96% confidence
Finding: The script’s behavior matches the export portion of the description: it validates a workspace, copies SOUL.md, creates package metadata files, and archives the package. However, the declared purpose says the skill can also import/install soul packages and batch list local packages. Those capabilities are absent from the supplied code chunk. Since the description claims broader functionality than what this code actually implements, this is a description-behavior mismatch.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 89% confidence
Finding: The skill documentation is entirely written in Chinese, which imposes a specific language on users without indicating that other languages are available. Under the policy, a language constraint should either be optional or clearly justified as region-specific.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 84% confidence
Finding: This shell script performs file writes into the target workspace and allows overwriting existing content when --force is set. Although there is a refusal message when overwrite is blocked, there is no explicit confirmation prompt or warning describing the consequences of --force before replacing SOUL.md, preview.md, and soul-manifest.json.
Content: No source excerpt is available for this finding.

Missing User Warnings

Low

Category: Not specified by scanner
Confidence: 92% confidence
Finding: The script embeds the full workspace path into both preview.md and manifest.json, then packages those files into the exported archive. If the archive is shared externally, it can disclose local filesystem structure, usernames, project names, or other environment details that the user may not realize are being published.
Content: No source excerpt is available for this finding.