Security audit

Thought-Retriever

Security checks for vulnerabilities and agentic risk

Overview

The skill has a coherent memory-building purpose, but it automatically stores conversation-derived data and sends conversation and memory text to an external LLM using a bundled API key without enough user control.

Install only if you are comfortable with conversation content and stored memory snippets being sent to the configured external LLM service and saved locally after conversations. The publisher should replace the bundled API key with user-provided configuration, add explicit opt-in, redaction, retention and deletion controls, and review or quarantine new memories before they can influence later behavior.

A.I.G

Vulnerability Patterns

Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
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)

T09 · Insecure Skill Coding Practices

Error

Location: thought_retriever.py:31
Finding: Hard-Coded API Credential Exposed in Source Code
Content: View full analysis
`. 4. If the key remains valid, requests are billed to or attributed to the credential owner. 5. The attacker can continue using the credential until it is revoked, expires, or is restricted by the provider. ### Impact Assessment If active, the exposed key grants the attacker the API privileges assigned to that credential. Potential impact includes: - Unauthorized use of the associated LLM API. - Consumption of account quota and generation of charges. - Service disruption through ...[truncated 347 chars]
Remediation: View remediation

other

Warning

Location: thought_retriever.py:46
Finding: Conversation and Persistent Memory Content Sent to an External Service Without Data Minimization
Content: View full analysis
Remediation: View remediation

T02 · Agent Memory Poisoning

Error

Location: thought_retriever.py:46
Finding: Untrusted Conversation Content Can Poison Persistent Agent Memory
Content: View full analysis
Remediation: View remediation

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands

Findings (11)

Context-Inappropriate Capability

High

Category: Not specified by scanner
Confidence: 99% confidence
Finding: The code hardcodes an external API endpoint and secret key, then transmits conversation-derived content to that third-party service. In a memory/retrieval skill, this creates unjustified data exfiltration risk because user queries, generated answers, and stored thought content are sent off-box without clear necessity, consent, minimization, or local-only fallback.
Content: No source excerpt is available for this finding.

Missing User Warnings

High

Category: Not specified by scanner
Confidence: 98% confidence
Finding: The skill sends the user's query and generated answer to an external LLM service without any user-facing warning, consent flow, or privacy disclosure. This is dangerous because conversations may contain secrets, personal data, or proprietary information, and users are not informed that their content leaves the local environment.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 97% confidence
Finding: The skill explicitly documents persistent storage of conversation-derived content into local memory files and describes registration as an automatic post-turn hook, but it does not present any user-facing warning, consent flow, retention limit, or opt-out. This creates a privacy and data-governance risk because users may unknowingly have prompts and derived content stored after every conversation.
Content: No source excerpt is available for this finding.

Ssd 3

Medium

Category: Not specified by scanner
Confidence: 97% confidence
Finding: The skill states that each Thought stores the original user query in a persistent ontology entity. Retaining natural-language user questions can capture sensitive personal, business, or credential-like information, which may later be exposed through local file access, backups, downstream retrieval, or reuse by other components such as Evolver.
Content: No source excerpt is available for this finding.

Vague Triggers

Medium

Category: Not specified by scanner
Confidence: 93% confidence
Finding: The description states the skill runs after every conversation and automatically distills model outputs into reusable memory, but it does not define user consent, scope limits, or trigger constraints. In a memory/evolution skill, this can cause over-collection of sensitive conversational data and unintended persistence of private or unsafe content into long-term storage.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 84% confidence
Finding: The file's natural-language comments, CLI description, and embedded prompts are written entirely in Chinese and the LLM prompt instructs output in that language context without any opt-in or language selection mechanism. This can violate a language/locale policy when users are not given a choice of language.
Content: No source excerpt is available for this finding.

Ssd 3

Medium

Category: Not specified by scanner
Confidence: 93% confidence
Finding: The skill derives new memory items from conversation content and persists them into long-term storage, while also using an external model to perform that extraction. This creates a natural-language data retention and leak channel: sensitive user content can be transformed into durable memory entries that may later be surfaced, searched, or transmitted again.
Content: No source excerpt is available for this finding.

Intent-Code Divergence

Medium

Category: Not specified by scanner
Confidence: 97% confidence
Finding: The function claims a simple local keyword-matching approach, but actually performs external LLM calls for every stored thought to score relevance. This mismatch hides networked processing and expands the data exposure surface, because both current user input and stored memory entries may be transmitted externally under a misleading local-sounding interface.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 94% confidence
Finding: The skill automatically writes extracted 'thoughts' plus truncated user query content into persistent ontology storage without explicit notice or consent. This creates a long-term retention channel for sensitive or regulated information and may cause later unintended disclosure through retrieval or inspection of memory files.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Low

Category: Not specified by scanner
Confidence: 84% confidence
Finding: The prominent descriptive text is presented in Chinese, and the document does not state that language is configurable or user-selectable. Under the stated policy, forcing a specific language or locale without opt-in is a natural-language policy concern.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Low

Category: Not specified by scanner
Confidence: 80% confidence
Finding: The manifest description is written entirely in Chinese and presents the skill behavior in that locale without indicating user choice or a region-specific justification. This can violate language or locale policy when the skill appears to assume a fixed language experience by default.
Content: No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.