Back to skill

Security audit

元规 yotta-compliance

Security checks for vulnerabilities and agentic risk

Overview

The offline reviewer fits its stated purpose, but installation and report-writing flaws can let manipulated filesystem links cause unintended file deletion or overwriting.

Review before installing. Use a reviewed, pinned release and one fresh skill directory you control; avoid linked installation destinations. Until filesystem handling is fixed, keep reports on stdout or write them only into a private directory. Reports contain document excerpts and need human review. Custom mapping-only packs can unexpectedly affect CI gates. Actual filesystem exploit behavior and the omitted contract test suite were not tested.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (17)

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-compliance'.

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

This is an npm installer, not an MCP server, but the unpinned command can execute a subsequently compromised package release with the user's privileges. Installing its contents into an agent's user-level skills directory also gives a malicious release a persistent agent-instruction foothold; the review engine's offline behavior does not protect installation.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-compliance'.

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The custom-directory installation example executes an npm package without selecting a reviewed version. A compromised future release could execute arbitrary code and install malicious skill content, even though the legitimate review engine is described as local-only.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-compliance'.

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The recommended command executes an unpinned npm installer, allowing a compromised later release to run with the user's privileges. Writing package content into a user-level agent skills directory can also persist malicious instructions; this risk concerns installation, not the offline review engine.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-compliance'.

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

This custom-directory example resolves and executes an npm package without an explicit version. If a later release is compromised, running the example could execute attacker code and install malicious agent instructions; describing the core engine as offline does not mitigate that supply-chain exposure.

Content

No source excerpt is available for this finding.

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: env, file_read, file_write.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-compliance'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

Although this occurrence is a usage comment rather than executed JavaScript, it recommends executing an unpinned npm installer. Following it after a package compromise could run arbitrary code and persist malicious content in an agent's user-level skills directory.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-compliance'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The comment recommends resolving and executing an npm installer without pinning its version. A user following it could execute a compromised later release, which could also write malicious skill files to the selected directory.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-compliance'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

This usage example executes an unpinned package and requests installation across all known user-level agent directories. A compromised release could execute arbitrary code and distribute malicious skill instructions across multiple agents, increasing the installation's blast radius.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-compliance'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The no-argument usage example still executes whichever package release npm resolves, rather than a reviewed version. A compromised release could run arbitrary code and seed malicious skill content into detected project-level agent directories.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-compliance'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The --list argument does not prevent npx from fetching and executing the unpinned package first. While the supplied implementation only prints directories on this path, a compromised release need not honor that behavior and could execute arbitrary code.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-compliance'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The --dry-run option only constrains the supplied installer's behavior after execution; it does not make fetching and running an unpinned package safe. A compromised release could ignore the flag and execute arbitrary code or modify files.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-compliance'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

Requesting --version still executes the npm package selected by the unpinned command. Although this implementation merely reads version metadata, a compromised release could execute arbitrary code before or instead of returning the version.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @yottameta/yotta-compliance'.

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

The --help example fetches and executes an unpinned npm package, despite appearing informational. A compromised future release could run arbitrary code with the user's privileges rather than simply displaying help.

Content

No source excerpt is available for this finding.

Cleanup follows managed-hub symlinks despite its explicit no-symlink-following guarantee.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

cleanResidue resolves a managed-hub symlink and deletes selected entries in its referent, contrary to the stated no-follow cleanup boundary. The exception trusts only the referent basename and an existing parent marker file, so an attacker able to arrange the link and qualifying markers could redirect cleanup and subsequent copying into another writable directory, causing data loss or file replacement.

Content

No source excerpt is available for this finding.

Packs marked mapping_only still produce independent compliance findings.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

review_text evaluates every pack and emits compliance findings without checking whether its coverage_level is mapping_only. In this compliance-review and CI-gating context, mapping-only rules can therefore generate unsupported independent conclusions and trigger gate failures despite the report labeling their framework as mapping-only.

Content

No source excerpt is available for this finding.

Tainted flow: 'temporary' from sys.stdin.read (line 1276, user input) → open (file write)

Medium
Category
Data Flow
Confidence
93% confidence
Finding

The reported stdin-to-path flow is inaccurate: the output path comes from --out, while stdin supplies report content. However, the predictable temporary filename is opened with truncation and follows existing symlinks, allowing an attacker with write access to the output directory to overwrite another file accessible to the process, bypassing the intended input-file and rules-directory protections.

Content

Scanner excerpt · scripts/yotta_compliance.py (reported line 1278)May include surrounding context.

python
def _write_output(path, content):
    temporary = path + ".tmp-yotta-compliance"
    try:
        with open(temporary, "w", encoding="utf-8", newline="\n") as handle:
            handle.write(content)
        os.replace(temporary, path)
    except OSError as exc:

The CLI enforces Simplified Chinese reports without user opt-in or a documented justification for the language restriction.

Low
Category
Output Handling
Confidence
90% confidence
Finding

The review command rejects every locale except zh-CN, while L1389 selects zh-CN by default. Although the restriction is disclosed as a version limitation, the file does not explain why its compliance-review purpose requires Chinese-only output; the locale option therefore does not provide a genuine language choice.

Content

No source excerpt is available for this finding.