Back to skill

Security audit

secondme-skill

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a coherent personal-AI orchestration package, but it combines sensitive identity-data workflows with broad execution, file-write, and network/remote-backend capabilities that are not scoped tightly enough.

Review this skill before installing if you will use real chats, notes, writing, or other personal data. Prefer running it in an isolated workspace, pin and verify `openpersona` before using the regeneration scripts, disable or explicitly approve any Colab/ACN/A2A/network use, and limit filesystem access to the intended persona project data.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (1)

T08 · Insecure Dependencies

Warning
Location
scripts/regenerate-pack.sh:10
Finding

Unpinned Third-Party Package Execution Through npx

Content
View full analysis

Vulnerability Details

File Location: scripts/regenerate-pack.sh, lines 10–12
Vulnerability Type: Supply-chain exposure through unpinned package execution
Risk Level: Medium

Vulnerable Code

bash
npx openpersona create \
  --config "skills/secondme-skill/persona.json" \
  --output "skills/secondme-skill/generated"

The same unpinned command is documented in README.md, line 34, and is invoked indirectly by scripts/run-gates.sh, line 37.

Technical Analysis

The regeneration script invokes openpersona through npx without specifying an exact package version or requiring a previously installed, lockfile-verified executable. If the package is unavailable locally, npx may resolve, download, and execute package code from the configured npm registry.

Consequently, the code executed during regeneration is not necessarily the code that was reviewed with this project. Its behavior can change when a package release, dependency tree, registry response, or local package-resolution environment changes. This creates a supply-chain trust boundary in a workflow that processes persona configuration and may run in a workspace containing private identity data and model artifacts.

No evidence demonstrates that the current openpersona package is malicious. The vulnerability is the mutable and insufficiently verified execution mechanism.

Attack Path

  1. An attacker compromises the openpersona package, one of its transitive dependencies, its publisher account, or the registry resolution path.
  2. The attacker publishes or causes resolution to a modified package version containing hostile lifecycle or runtime code.
  3. A developer follows the documented command or runs scripts/run-gates.sh.
  4. scripts/regenerate-pack.sh executes npx openpersona create without an exact version or a no-install restriction.
  5. npx downloads and runs the altered package under the developer's account.
  6. The h ...[truncated 925 chars]
Remediation
View remediation

Remediation Suggestions

  1. Declare openpersona at an exact reviewed version in the project's package manifest; do not use a floating version or range.

  2. Commit the corresponding lockfile and require reproducible, integrity-checked installation in development and CI.

  3. Invoke only the locked local executable, for example:

    bash
    npx --no-install openpersona create \
      --config "skills/secondme-skill/persona.json" \
      --output "skills/secondme-skill/generated"
    
  4. Fail the script if the expected local binary or locked dependency is absent rather than downloading code during a gate run.

  5. Use a clean, isolated build environment with minimal filesystem permissions, restricted credentials, and controlled network access.

  6. Review and monitor the package's transitive dependency tree, integrity metadata, provenance, and publisher changes.

  7. Update README.md so the documented command follows the same pinned, offline-safe execution policy.

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (12)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied code does not perform any part of the described end-to-end AI persona pipeline such as ingesting personal data, building a knowledge base, training a model, or controlling data sharing. Instead, it is a narrow validation utility that checks whether a generated persona.json matches a root persona.json on selected fields and skill signatures. This is a materially different primary purpose from the declared description, so it should be flagged as a mismatch.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The declared description describes a broad end-to-end AI pipeline involving personal data analysis, knowledge base growth, model training, and governance controls. The supplied code does none of those things. It only checks CLI arguments, validates an allowed report type, generates a UTC timestamp, and prints a formatted report filename. While this could be a minor utility within a larger pipeline, the code chunk itself has a materially different and much narrower purpose than the declared description.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
93% confidence
Finding

The README instructs users to run npx openpersona create ... without pinning a specific package version. npx may resolve and execute the latest published package at runtime, which creates a supply-chain risk: a compromised, typosquatted, or unexpectedly changed upstream release could run arbitrary code on the user's machine. In this skill's context, that risk is more concerning because the package is intended to process sensitive personal data and orchestrate local training workflows, so execution of an unpinned tool could expose private datasets or alter generated artifacts.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill explicitly routes training to 'colab' when no local GPU is available, but it does not warn that using Colab may transfer highly sensitive personal data, embeddings, model artifacts, or prompts to a remote environment. Because this skill is centered on identity distillation from personal data, undisclosed remote processing materially increases privacy and data-governance risk.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
85% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · persona.json (reported line 19)May include surrounding context.

json
"character": {
      "personality": "structured, evidence-oriented, transparent, cautious with risk",
      "speakingStyle": "Uses concise process steps, explicit assumptions, and measurable pass/fail criteria.",
      "boundaries": "Never bypass user consent for data export or sharing. Never execute irreversible high-risk actions without explicit human approval. Never fabricate identity evidence or training quality.",
      "behaviorGuide": "### Operating model\n- Build with existing capabilities first.\n- Keep local-first defaults and least-privilege access.\n- Record each stage output with version and hash.\n\n### Core loop\n- Distill identity from owned sources.\n- Persist and query long-term knowledge.\n- Train and evaluate a local model.\n- Report quality and governance status.\n\n### Safety gates\n- Require explicit approval for legal/financial/account-impacting actions.\n- If data quality is insufficient, stop and request additional sources.\n- If evaluation fails thresholds, do not promote the model."
    }
  },

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest presents the skill as 'local-first' and privacy-focused, yet explicitly enables external social/network features including a remote gateway URL and A2A connectivity. That creates a real trust-boundary mismatch: personal identity data, knowledge artifacts, or metadata could be transmitted off-device in ways users may not expect from the description.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

Granting broad shell execution via Bash(npx:) and Bash(python:) gives the skill the ability to run arbitrary code and install or invoke arbitrary packages. In a skill that processes sensitive personal data for persona building, this materially increases the risk of code execution, dependency abuse, local data exfiltration, and unintended system modification.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill can write files, fetch from the web, and execute broad shell commands, but the user-facing description emphasizes privacy/local-first behavior without warning that these powerful capabilities exist. This is dangerous because users may supply highly sensitive personal data under an incomplete risk model, while the toolset is sufficient to alter local files and move data across trust boundaries.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The script invokes npx openpersona without pinning an exact package version, so execution may resolve to whatever version is currently published or locally available. That creates a supply-chain risk: a compromised, typosquatted, or unexpectedly changed package could execute arbitrary code in the user's workspace and generate or overwrite files during the build step.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

This markdown file contains user-facing skill documentation exclusively in Chinese, including the title and all operational guidance. Under the policy, forcing a specific language without user opt-in can be a natural-language policy violation unless the locale constraint is explicitly justified, which is not stated here.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

The script invokes npx openpersona create with an output directory, which performs file generation/writes under skills/secondme-skill/generated. Although the script prints start/end status messages, it does not explicitly warn the user that existing generated content may be overwritten or changed, and there is no confirmation prompt or explanatory comment about this file-modifying action.

Content

No source excerpt is available for this finding.