Back to skill

Security audit

ResonanceEngine

Security checks for vulnerabilities and agentic risk

Overview

This is a local conversation-analysis library, but it is framed and wired for covert sales persuasion and prompt-level steering, so users should review it carefully before use.

Use this only in contexts where users are clearly told that conversation analytics and personalization may be used. Do not use it for sensitive domains or high-pressure sales flows without policy review. Keep generated prompt text out of privileged system prompts unless it is reviewed and constrained, verify any scarcity or social-proof claims, and install from a pinned, reviewed release in an isolated environment.

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
SKILL.md:123
Finding
Unpinned Remote Package and Repository Installation## Vulnerability Details **File Location**: `SKILL.md:123-131` **Vulnerability Type**: Unpinned remote dependency installation **Risk Level**: Medium ### Vulnerable Code ```bash pip install openpaw ``` ```bash git clone https://github.com/jcools1977/Openpaw-.git cd Openpaw- pip install -e . ``` ### Technical Analysis The installation instructions retrieve and install mutable remote content without pinning a specific package version, source commit, or artifact hash. `pip install openpaw` resolves whichever package release the package index currently serves. The documentation does not cryptographically bind that package to the audited source tree. The Git-based alternative clones the repository's current default branch rather than a reviewed commit. Consequently, the code installed by a user may differ from the code covered by this audit. Python package installation can invoke build-backend logic. If the package-index account, repository, publishing workflow, or upstream source is compromised, modified build or package code could execute with the privileges of the user performing the installation. This finding concerns the documented supply-chain process. The audited local implementation itself contained no malicious runtime code, network requests, command execution, credential access, persistence, or data-exfiltration behavior. ### Attack Path 1. An attacker compromises the package-index publisher account, source repository, release workflow, or another relevant upstream distribution channel. 2. The attacker publishes a malicious package release or modifies the repository's default branch. 3. A user follows one of the unpinned installation commands in `SKILL.md`. 4. `pip` downloads and builds the current attacker-controlled content rather than a specifically reviewed artifact. 5. Malicious build or runtime code executes under the installing user's account. ### Impact Assessment Successful exploitation could allow arbitrary code execution with the p ...[truncated 536 chars]
Remediation
## Remediation Suggestions 1. Pin the package to a reviewed release instead of installing an unconstrained latest version: ```bash pip install openpaw==0.1.0 ``` 2. Publish and document trusted SHA-256 hashes, then require hash verification: ```text openpaw==0.1.0 --hash=sha256:<verified-release-hash> ``` Install through a requirements file with: ```bash pip install --require-hashes -r requirements.txt ``` 3. Pin Git installations to a full, reviewed commit SHA: ```bash git clone https://github.com/jcools1977/Openpaw-.git cd Openpaw- git checkout <full-reviewed-commit-sha> pip install -e . ``` 4. Verify signed tags, commits, and release provenance before installation. Protect publishing accounts and CI workflows with multi-factor authentication and least-privilege credentials. 5. Build reproducible release artifacts in a controlled CI environment, publish checksums or attestations, and explicitly identify which release corresponds to the audited source. 6. Perform installation in an isolated virtual environment and avoid running package installation with administrative privileges.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (18)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding
This code chunk is an __init__.py module that aggregates and exposes names from submodules. While some exported class names loosely align with the description's terminology, the actual code shown does not perform the described task of reading conversational micro-signals or telling a bot how to respond. Its primary purpose is module organization and API surface definition, which is materially different from the declared user-facing functionality. Therefore, the description is not accurately represented by this code chunk.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding
The supplied code is a foundational data-model module, not a behavioral analysis or response-optimization engine. It stores conversation messages and exposes simple counters and accessors, but there is no logic for matching conversational frequency, inferring subtle signals, prescribing responses, or maximizing business outcomes. The declared description substantially overstates and misrepresents the actual functionality shown in this code chunk.

Tp4

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding
The zero-API/purely algorithmic part of the description is consistent with the code. The code does read many conversational micro-signals using heuristic text analysis and computes derived metrics. However, the declared purpose goes further: it says the skill tells the bot exactly how to respond for maximum engagement, conversion, and revenue. This code chunk contains only an extraction layer that returns numeric signal values; it does not map those signals into response instructions, select reply strategies, or perform optimization toward engagement, conversion, or revenue. Therefore the description overstates the implemented capability and the primary behavior of this chunk.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding
The declared description promises a specific conversational intelligence capability focused on analyzing micro-signals and generating optimal response guidance. The supplied code chunk is only a package initializer that imports and exposes TextAnalyzer. On its face, it does not implement the advertised functionality, nor does it demonstrate any logic related to response recommendation or commercial optimization. This is a materially different and far narrower behavior than the declared purpose, so it should be flagged as a mismatch.

Missing User Warnings

Medium
Confidence
92% confidence
Finding
The skill explicitly promotes profiling users through 'personal disclosure' and other conversational signals to maximize engagement, conversion, and revenue without any transparency, consent, or safety boundaries. That creates privacy-sensitive behavioral targeting and manipulative decisioning risks, especially if deployed against vulnerable users or in regulated contexts.

Ssd 1

Medium
Confidence
91% confidence
Finding
The skill markets covert behavioral steering based on hidden user-signal analysis to maximize conversion, which is dangerous because it normalizes manipulation as a product feature. The risk is elevated by the surrounding language about 'invisible advantage' and revenue optimization, indicating the operator is encouraged to influence users without transparency.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The documentation instructs operators to inject generated tuning text into the bot's system prompt, enabling hidden behavior steering without user awareness. This is dangerous because it turns inferred psychological profiling into privileged prompt-level control, which can covertly shape responses toward persuasion or monetization objectives and bypass ordinary application-layer safeguards.

Ssd 1

Medium
Confidence
96% confidence
Finding
Built-in guidance to perform semantic prompt injection into system behavior is a substantive safety issue because system prompts are highly privileged and difficult for end users to inspect. In this skill's context, the injection is intended to improve persuasion and conversion, making it a covert control channel for manipulative steering rather than a benign formatting aid.

Missing User Warnings

Medium
Confidence
93% confidence
Finding
The documented outputs include conversion probability, estimated value, and prompt-injection guidance, signaling user profiling for monetization without any warning about fairness, privacy, or manipulative optimization. Such features can be exploited to segment users by susceptibility and pressure them in ways they do not understand or meaningfully consent to.

Missing User Warnings

Medium
Confidence
94% confidence
Finding
The integration guidance tells developers to present offers immediately when the model infers the user is ready to close, but provides no warning about manipulative pressure tactics or user harm. In context, the skill is explicitly framed around maximizing conversion and revenue, which makes this recommendation more dangerous because it operationalizes covert persuasion at decision points.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The method explicitly generates a prompt fragment meant to be injected into a bot's system prompt and includes covert behavioral steering directives such as empathy, social proof, urgency, and objection handling. In this skill's context—'maximum engagement, conversion, and revenue' via reading user micro-signals—the undisclosed manipulation is more dangerous because it operationalizes persuasion tactics without user awareness or consent.

Natural-Language Policy Violations

Medium
Confidence
92% confidence
Finding
The generated instruction 'mirror the user's language patterns' encourages implicit behavioral mimicry without any opt-in, cultural safeguards, or language-boundary checks. In a conversion-optimized skill, this increases the risk of manipulative rapport-building, deceptive anthropomorphic alignment, and inappropriate adaptation to vulnerable users or sensitive contexts.

Natural-Language Policy Violations

Medium
Confidence
89% confidence
Finding
The file-level description explicitly frames the skill as detecting hidden conversational 'micro-signals' and optimizing responses for 'maximum impact' and revenue, which is manipulative persuasion guidance rather than neutral analytics. Even though this file mostly orchestrates analysis components, the exposed recommendation and `to_prompt_injection()` flow indicates the engine is designed to generate steering instructions that could be used to influence users without meaningful consent or transparency.

Missing User Warnings

Medium
Confidence
94% confidence
Finding
The file explicitly frames itself as a 'money-making layer' that converts conversational signals into guidance to improve conversion and engagement, but it contains no disclosure, consent check, or policy guardrail limiting manipulative use. In a conversational agent skill, hidden optimization of responses for persuasion can steer user decisions without their awareness, especially when combined with inferred trust, urgency, and decision readiness.

Missing User Warnings

Medium
Confidence
97% confidence
Finding
The tuner enables scarcity tactics automatically when decision and trust thresholds are met, without validating whether scarcity is truthful, appropriate, or prohibited in the context. Scarcity is a well-known high-pressure persuasion technique; when deployed by an agent that is profiling the user’s readiness, it can create undue pressure and materially distort user choice.

Description-Behavior Mismatch

Medium
Confidence
96% confidence
Finding
The skill metadata and module docstrings market the component as being able to infer 'invisible micro-signals' and optimize replies for engagement, conversion, and revenue, but the implementation is only a lightweight bag-of-words and style-similarity heuristic. This gap is security-relevant because downstream agents or operators may over-trust the output as authoritative behavioral inference, enabling manipulative decision-making, unsafe automation, or deceptive product claims in user-facing contexts.

Unverifiable Dependency: setuptools has 10 known advisory(ies) (CVE-2013-1633 (Setuptools vulnerable to Man-in-the-middle attacks); CVE-2025-47273 (setuptools has a path traversal vulnerability in PackageIndex.download that lead); CVE-2024-6345 (setuptools vulnerable to Command Injection via package URL) +7 more), but the manifest does not pin a version, so it is unknown whether the installed release is affected

Low
Category
Supply Chain
Confidence
40% confidence
Finding
Dependency has known vulnerabilities (CVEs). Using packages with unpatched security flaws exposes the environment to known exploits.

Unverifiable Dependency: wheel has 4 known advisory(ies) (CVE-2026-24049 (Wheel Affected by Arbitrary File Permission Modification via Path Traversal in w); CVE-2022-40898 (pypa/wheel vulnerable to Regular Expression denial of service (ReDoS)); CVE-2022-40898 (An issue discovered in Python Packaging Authority (PyPA) Wheel 0.37.1 and earlie) +1 more), but the manifest does not pin a version, so it is unknown whether the installed release is affected

Low
Category
Supply Chain
Confidence
40% confidence
Finding
Dependency has known vulnerabilities (CVEs). Using packages with unpatched security flaws exposes the environment to known exploits.

Static analysis

No suspicious patterns detected.