Security audit

Soul Dreaming

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent local memory skill, but it asks agents to persist and prune sensitive workspace notes without enough user control or redaction guidance.

Install only if you want an agent to maintain plaintext memory files in your workspace. Before use, set explicit rules to review every memory write, never store passwords, tokens, keys, session cookies, private connection strings, or precise credential locations, and require confirmation before deletion, merging, archival, heartbeat, or cron-style maintenance.

SkillSpector

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 (11)

Vague Triggers

Medium

Confidence: 92% confidence
Finding: The activation conditions are broad enough that the skill may trigger on common words like "memory" or inferred context loss, causing it to run when the user did not explicitly request persistent memory behavior. In this skill's context, unintended activation is more dangerous because the skill then instructs the agent to read and write memory files, increasing the chance of unauthorized persistence of sensitive or irrelevant conversation data.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The skill explicitly directs the agent to persist session content, promote entries across files, delete absorbed journal entries, and archive older data without a clear disclosure or consent flow. This is dangerous because it can silently retain sensitive user data and modify local state in ways the user may not expect, creating privacy, integrity, and retention risks.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The checklist explicitly tells the agent to write down 'a security-relevant finding or credential location' to persistent memory files, but provides no guardrails to avoid storing secrets, secret paths, or other sensitive operational metadata. In a memory-persistence skill, that guidance increases the chance that credentials or highly actionable secret-discovery information will be retained in broadly accessible files, expanding exposure through logs, sync, backups, or later retrieval.

Missing User Warnings

Medium

Confidence: 92% confidence
Finding: The document explicitly instructs deletion of daily journals after promotion/expiry, but it provides no safeguard such as confirmation, backup, retention exception, or dry-run guidance. In a memory-management skill, these instructions could lead an agent to irreversibly remove records that may still contain needed context, audit history, or user data, causing data loss and reducing recoverability.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: The weekly cleanup repeats a destructive deletion action for old journals without warning that removal may be irreversible or that entries may still have future value. Because this skill is meant to preserve memory across compaction and session loss, unsafe cleanup guidance directly undermines its purpose and can destroy useful or sensitive records without review.

Missing User Warnings

Medium

Confidence: 92% confidence
Finding: This protocol explicitly instructs agents to delete old entry content after supersession, but it does not require user confirmation, backup retention guarantees, or a safe rollback path. In a memory-management skill, that creates a real risk of unintended data loss or silent destruction of context the user may still need.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: The duplicate-consolidation rule tells the agent to delete other versions after merging, but provides no verification checklist, diff review, or confirmation requirement. Because this skill is designed to persist and manage memory, accidental omission of unique details during merge could permanently degrade stored knowledge.

Missing User Warnings

Medium

Confidence: 90% confidence
Finding: The REM procedure operationalizes moving and removing entries from source files based on scores, yet it lacks any user-facing warning, dry-run mode, or confirmation gate before modifying stored data. In context, this is more dangerous because the skill's purpose is long-term memory preservation, so automated archival/removal mistakes directly undermine the core safety objective.

Missing User Warnings

Medium

Confidence: 91% confidence
Finding: The consolidation step instructs summing hit counts and deleting duplicates, but does not define how to validate semantic equivalence or preserve discarded records. This can cause irreversible loss of nuanced or conflicting information, especially in an agent memory system where near-duplicates may contain materially different context.

Ssd 3

Medium

Confidence: 96% confidence
Finding: The skill mandates persistent capture of decisions, discoveries, pending tasks, and context anchors into long-lived markdown files, which naturally encourages storing sensitive session data in unstructured text. In a memory-persistence skill, this is more dangerous because the whole design normalizes retention across sessions, increasing the chance of collecting secrets, personal data, internal context, or confidential decisions that can later be exposed, over-retained, or misused.

Ssd 3

Medium

Confidence: 95% confidence
Finding: The taxonomy explicitly creates a `people.md` store for user preferences, roles, and communication habits, encouraging accumulation of personal or quasi-personal profiles in natural language. This is especially risky in a memory skill because these records are intended for long-term reuse, making privacy leakage, profiling, unauthorized persistence, and policy noncompliance more likely.

Static analysis

No suspicious patterns detected.