Back to skill

Security audit

Feelings Framework

Security checks for vulnerabilities and agentic risk

Overview

This skill appears to be a benign local feelings-state library, with disclosed mood persistence and some install/privacy cautions.

Install from the bundled reviewed source or a verified exact registry version. Before enabling it for an agent, decide whether persistent mood data is appropriate, where the JSON/localStorage state will live, who can read it, and how users can reset or delete it.

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:28
Finding

Unpinned Third-Party Package Installation

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:28-31; additional occurrences in README.md:34-36 and README.md:60-62
Vulnerability Type: Supply-chain exposure through unpinned package installation
Risk Level: Medium

Vulnerable Code

SKILL.md:28-31:

markdown
Or pip install if published:

```bash
pip install feelings-framework
text

`README.md:34-36`:

```markdown
```bash
pip install feelings-framework
text

`README.md:60-62`:

```markdown
```bash
npm install feelings-framework
text

### Technical Analysis

The documentation instructs users to install `feelings-framework` by package name without pinning an exact version or verifying the retrieved artifact with a hash, signature, lockfile, or trusted publisher identity.

These commands resolve mutable content from the package registry configured in the user's environment. Consequently, the installed package is not guaranteed to correspond to the local source code reviewed in this audit. The wording “if published” in `SKILL.md` also leaves the ownership and expected identity of the registry package unclear.

This creates a supply-chain risk if the package name is unclaimed, transferred, compromised, shadowed through dependency confusion, or updated with malicious content. Package installation or subsequent import can then execute code that is absent from the audited project.

### Attack Path

1. An attacker gains control of the relevant Python or npm package name, compromises its publisher account, or influences package resolution through a malicious registry configuration.
2. The attacker publishes a package version containing malicious installation, build, import-time, or runtime code.
3. A user follows the documented `pip install feelings-framework` or `npm install feelings-framework` command.
4. The package manager resolves and installs the attacker-controlled or compromised artifact because no version or integ
...[truncated 1024 chars]
Remediation
View remediation

Remediation Suggestions

  1. Prefer installation from the bundled, reviewed source rather than an external registry:

    bash
    python -m pip install .
    

    For development use:

    bash
    python -m pip install -e .
    
  2. If registry installation is required, pin an exact reviewed version:

    bash
    python -m pip install feelings-framework==1.0.0
    npm install feelings-framework@1.0.0 --save-exact
    
  3. For Python distribution, publish and document cryptographic hashes and install with a hash-enforced requirements file:

    text
    feelings-framework==1.0.0 --hash=sha256:<verified-digest>
    
    bash
    python -m pip install --require-hashes -r requirements.txt
    
  4. Commit appropriate lockfiles for consumers or examples and use reproducible installation workflows such as npm ci.

  5. Explicitly document the authoritative registry, expected publisher or organization, package URL, source repository, and release verification procedure.

  6. Protect publisher accounts with multi-factor authentication and trusted publishing, and verify that built registry artifacts correspond to tagged source releases.

  7. Remove the external installation instructions until package ownership and artifact provenance have been confirmed.

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (9)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The specification explicitly instructs implementations to persist an agent’s mood and feelings state across sessions in a workspace JSON file, but provides no user-facing notice, consent model, retention limits, or guidance on handling potentially sensitive interaction-derived state. Because this framework is designed to remember interaction patterns and carry them forward, the stored data can reveal behavioral history and inferred emotional state, creating a privacy risk if users are unaware of the persistence or if local files are exposed to other processes or users.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The 'When It Activates' section uses subjective phrases like 'when interactions should feel like they matter' and 'when you want agents to have consistent emotional memory' without defining clear trigger phrases, scope boundaries, or exclusion conditions. In a markdown skill description, this can cause unintended invocation because the activation guidance overlaps with many ordinary conversational or design contexts.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
55% 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 · library/js/feelings/core.js (reported line 233)May include surrounding context.

js
this._state = this._makeState(agentId, initialMood);
    this._triggerCounts = {};

    // Load saved state
    const loaded = this.memory.load();
    if (loaded) {
      this._state = this._makeState(loaded.agentId, loaded.mood);

Session Persistence

Medium
Category
Rogue Agent
Confidence
55% 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 · library/python/feelings/core.py (reported line 208)May include surrounding context.

python
self._current_calibration_key = "default"
        self._state = FeelingsState(agent_id=agent_id, mood=initial_mood)

        # Try to load saved state
        loaded = self.memory.load()
        if loaded:
            self._state = FeelingsState.from_dict(loaded)

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
40% confidence
Finding

Dependencies lack version pinning, allowing potential malicious package updates. Consider pinning versions.

Content

Scanner excerpt · library/js/feelings/package.json (reported line 35)May include surrounding context.

json
"node": ">=18.0.0"
  },
  "devDependencies": {
    "typescript": "^5.0.0"
  }
}

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.

Content

No source excerpt is available for this finding.

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.

Content

No source excerpt is available for this finding.

Vague Triggers

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

This is a TOML manifest file, so vague-trigger review applies. The description "A general-purpose AI agent feelings engine" is extremely broad and lacks any specific trigger scope, constraints, or negative examples, which could contribute to unintended invocation if tooling uses manifest descriptions for skill matching.

Content

No source excerpt is available for this finding.

Unverifiable Dependency: pytest has 2 known advisory(ies) (CVE-2025-71176 (pytest has vulnerable tmpdir handling); CVE-2025-71176 (pytest has vulnerable tmpdir handling)), 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.

Content

No source excerpt is available for this finding.