Back to skill

Security audit

qapten-sentient

Security checks for vulnerabilities and agentic risk

Overview

The skill is not malicious, but it stores and reuses inferred emotional state with weak ongoing-consent and deletion controls, so it should be reviewed before installation.

Install only if you are comfortable with the agent inferring your emotional state and keeping a local tone memory between sessions. Prefer a version that asks for explicit opt-in before reading or writing that memory, provides inspect and delete commands, expires stored state clearly, and treats the file contents strictly as data.

Vulnerability Patterns
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

T02 · Agent Memory Poisoning

Warning
Location
skill.md:8
Finding

Persistent Emotional State Can Influence Future Sessions Without Robust Validation

Content
View full analysis
2 jours) : ignorer, repartir en Sérénité ``` ```markdown ## Heartbeat & activation Cette skill ne s'active pas automatiquement. Elle nécessite une autorisation explicite de l'utilisateur. **Au premier démarrage** (si tone-state.md n'existe pas encore) : Demander à l'utilisateur : "La skill qapten-sentient peut adapter mon ton en fonction de tes émotions et les mémoriser chaque jour. Tu veux l'activer ? (oui/non)" **Si l'utilisateur dit non** : rester en mode Sérénité fixe, ne pas relancer la question. **Si l'utilisateur dit oui** : activer la détection émotionnelle et créer tone-state.md à la fin de la session. **Renouvellement du consentement** : si tone-state.md existe, pas besoin de redemander — l'accord est implicite. L'utilisateur peut désactiver à tout moment avec "désactive le mode émotionnel" ou "redeviens une IA". ``` ### Technical Analysis The skill instructs the agent to read a cross-session state file at startup and use its emotional-state and calibration fields to influence future responses. It also directs the agent to create that file after opt-in and treats the file's mere existence as proof of continuing consent. No schema validation, integrity verification, ownership check, or strict enumeration of accepted state values is specified. Consequently, anyone or anything capable of modifying `tone-state.md` could place manipulated state in t ...[truncated 1703 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (4)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The entire skill description and prescribed response style are written to operate in French, including fixed French trigger and deactivation phrases, with no indication that the user can opt into another language. This creates a locale policy concern because the skill appears to force a specific language mode rather than offering a choice.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill instructs the agent to read a persisted emotional-state file at session start and use inferred emotion as a calibration signal across sessions. Emotional state is sensitive inferred user data, and the skill does not present a clear, durable privacy notice about retention, purpose, scope of reuse, or deletion, creating a meaningful privacy and data-minimization risk.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill explicitly tells the agent to persist and reuse inferred emotional state from prior sessions, including using file-based storage as a behavioral input. Reusing sensitive inferred data without renewed explicit consent increases the risk of privacy leakage, unexpected personalization, and retention of intimate user attributes beyond the original interaction.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

This section normalizes implicit ongoing consent once the file exists, allowing sensitive emotional inferences to be stored and reused indefinitely without fresh confirmation. That is dangerous because it treats inferred mental/emotional data as durable memory, which can surprise users, leak sensitive context across sessions, and violate privacy expectations.

Content

No source excerpt is available for this finding.