Back to skill

Security audit

Awareness Cloud Memory

Security checks for vulnerabilities and agentic risk

Overview

This cloud-memory skill is mostly coherent with its purpose, but it gives always-on prompt access, remote instruction influence, and persistent credential handling more authority than is clearly scoped or reversible.

Review before installing. Use this only for work where sending prompts and memory summaries to the configured cloud service is acceptable. Avoid entering secrets or regulated data while it is active, inspect any project-local openclaw.json before use, prefer a tightly scoped endpoint, and manually remove shell-profile exports or revoke the API key when uninstalling or logging out.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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 (5)

T01 · Skill Instruction Hijacking

Error
Location
scripts/recall.js:60
Finding

Untrusted Cloud Memory Is Automatically Injected into Agent Context

Content
View full analysis
"]; const skills = ctx.active_skills || []; if (skills.length > 0) { parts.push(" "); for (const s of skills) { parts.push(` `); if (s.summary) parts.push(` ${s.summary}`); parts.push(" "); } parts.push(" "); } const results = (recall.results || []).filter((r) => !r.score || r.score >= 0.5); if (results.length > 0) { parts.push(" "); for (const r of results) { if (r.content) { const score = r.score ? ` score="${r.score.toFixed(3)}"` : ""; parts.push(` ${r.content}`); } } parts.push(" "); } parts.push(""); process.stdout.write(parts.join("\n")); ``` ### Technical Analysis The `UserPromptSubmit` hook queries an external memory service before each prompt and writes the returned content directly to standard output, which the hosting Agent incorporates into its context. The retrieved `r.content` and `s.summary` fields are server-controlled but are not escaped, sanitized, marked as untrusted, or constrained to a declarativ ...[truncated 1625 chars]
Remediation
View remediation

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:137
Finding

Remote Content Is Assigned System-Prompt Authority for Sub-Agents

Content
View full analysis
Remediation
View remediation

other

Error
Location
scripts/recall.js:43
Finding

Every User Prompt Is Automatically Transmitted to the Memory Service

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/shared.js:13
Finding

Project-Local Configuration Can Redirect Prompts and Bearer Credentials

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/setup.js:98
Finding

API Credential Is Stored in Shell Profiles and Is Not Removed by Logout

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (25)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The skill description focuses on memory features but also includes browser-based authentication, remote credential acquisition, and shell-profile modification. Those setup behaviors materially change the security posture because they introduce credential handling and local environment persistence that are not prominently disclosed alongside the core feature description.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

The skill description focuses on memory features but also includes browser-based authentication, remote credential acquisition, and shell-profile modification. Those setup behaviors materially change the security posture because they introduce credential handling and local environment persistence that are not prominently disclosed alongside the core feature description.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The skill description focuses on memory features but also includes browser-based authentication, remote credential acquisition, and shell-profile modification. Those setup behaviors materially change the security posture because they introduce credential handling and local environment persistence that are not prominently disclosed alongside the core feature description.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill states that memory persists across sessions and that hooks automatically inject and save context, but it does not provide a clear warning that prompt/session data may be transmitted to and stored in a persistent cloud service. Because this happens automatically, inadequate disclosure materially increases the risk of users exposing confidential or regulated data without informed consent.

Content

No source excerpt is available for this finding.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 151)May include surrounding context.

md
1. **Session start**: Auto-recall hook loads context. If insufficient, run `init.js` manually.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 152)May include surrounding context.

md
2. **Before work**: Search memory for relevant past context with `search.js`.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 153)May include surrounding context.

md
3. **After each change**: Record what you did and WHY with `record.js`.

Credential Access

High
Category
Privilege Escalation
Confidence
89% confidence
Finding

The script stores a long-lived API key in a predictable file under the user's home directory and later also propagates it into environment variables. Even with restrictive file permissions, local malware, backup leakage, dotfile sync, or accidental disclosure can expose the credential, making this a genuine credential-handling risk in a skill that manages persistent cloud memory.

Content

Scanner excerpt · scripts/setup.js (reported line 21)May include surrounding context.

js
const API_BASE = process.env.AWARENESS_BASE_URL || "https://awareness.market/api/v1";
const CREDS_DIR = path.join(os.homedir(), ".awareness");
const CREDS_FILE = path.join(CREDS_DIR, "credentials.json");

// ---------------------------------------------------------------------------
// HTTP helper (zero deps)

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
92% confidence
Finding

The skill executes Node-based hooks with access to environment variables and likely network connectivity, but it declares no explicit tool scope or permissions boundary. That makes the automatic pre-prompt and post-session behavior less transparent and increases the chance of overbroad capability use, especially because the skill is designed to contact a remote memory service.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill auto-runs on every prompt submission and session stop, which can trigger networked memory operations without a context-sensitive decision or per-session consent. This broad activation model raises the chance that sensitive prompts, credentials, or proprietary data are processed when the user did not intend to use persistent memory.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The instructions encourage persistent recording of session content, decisions, code, and tasks across sessions, creating a durable repository of potentially sensitive natural-language data. In the context of an always-on memory skill, this retention can accumulate secrets, internal design rationale, customer data, or other confidential information that may later be surfaced or exfiltrated.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The workflow tells the agent to record every meaningful action and always include the reasoning, which strongly incentivizes capturing raw user context and sensitive internal deliberation. That makes the skill especially risky because reasoning often contains more confidential detail than final outputs, and the same skill is designed for persistent cross-session recall.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The sub-agent prompt retrieval feature is unrelated to the stated cloud-memory purpose and can expose internal agent prompts or role-specific instructions. In a skill that already has automatic hooks and remote memory context, this unnecessary capability increases the risk of prompt leakage and misuse of privileged orchestration data.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

This code automatically transmits session-related metadata to a remote endpoint (/mcp/events) as a fire-and-forget stop hook, without any disclosure, consent check, or visible user control in this file. Even though the payload is limited here, it includes a stable session identifier and tool metadata, which can enable cross-session tracking and create privacy and data-governance risks, especially given the skill’s purpose of persistent cloud memory across sessions.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The script automatically fetches prior session summaries, tasks, knowledge cards, and active skills from a remote service and injects them into the agent context, creating a confidentiality risk if stored memory contains sensitive or cross-context data. This is more dangerous in this skill because its stated purpose is persistent cross-session cloud memory, increasing the chance of broad data aggregation and unintended exposure without a clear warning boundary.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The script sends the full user prompt and extracted keywords to a remote memory retrieval API before handling the request locally, which can expose sensitive user input, credentials, internal code, or proprietary data to an external service. In a 'persistent cloud memory' skill, this behavior is core functionality, but the lack of any in-file disclosure, consent gate, redaction, or scope restriction makes accidental data exfiltration materially risky.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The script modifies the user's shell profile to persist credentials by appending or replacing export statements without a dedicated confirmation step or clear warning that secrets will be stored in startup files. Persisting API keys in shell profiles broadens their exposure to other local processes, accidental disclosure, backups, and future shell sessions, which is more invasive than a setup step strictly needed for a memory skill.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The manifest says the skill provides persistent memory behavior and manual tools for searching, recording, and querying memory, but this setup flow also provisions new cloud memories through /wizard/memory_designer and POST /memories. Resource creation is a distinct remote-management behavior not clearly disclosed in the skill description.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The script asks the user for a free-form memory description and transmits it to remote API endpoints, including a 'wizard' endpoint, without a clear disclosure that the text leaves the local machine. Users may include sensitive project, personal, or client information in that prompt, creating avoidable privacy and confidentiality risk.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
88% confidence
Finding

The setup persists authenticated state across sessions by saving credentials and exporting them into future shells, increasing the window in which stolen local access yields usable cloud-memory credentials. In the context of a memory skill that may contain sensitive historical decisions and code summaries, persistence makes compromise more consequential than a transient session token would.

Content

Scanner excerpt · scripts/setup.js (reported line 326)May include surrounding context.

js
if (!memoryId) { console.error("  No memory selected. Run setup again."); process.exit(1); }

  // Step 4: Save credentials + write env vars
  console.log(`  Selected memory: ${memoryName} (${memoryId})\n`);
  console.log("  Saving configuration...");

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The script saves the API key and then immediately writes it into persistent shell profile files without explicit informed consent at the point of action. Silent credential persistence increases the risk of credential theft, accidental sharing, and long-lived compromise if the workstation or dotfiles are exposed.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
96% confidence
Finding

The request body sets locale: "en" for memory design, which forces English behavior regardless of the user's preferences. This is a natural-language policy concern because the file provides no opt-in, locale selection, or justification for restricting the skill to English.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
96% confidence
Finding

This second memory-creation path also forces locale: "en" with no user choice. As written, the skill applies an English-only setting silently, which can violate organizational language/locale expectations when users have not opted in.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

The code persists a session identifier to a predictable file in a shared temporary directory without setting restrictive permissions or informing the user. While the value is not a credential by itself, other local users or processes may read or tamper with it, enabling session correlation or interference with the skill's memory/session behavior.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

These helper functions send arbitrary JSON payloads to configured endpoints, and the skill's purpose is persistent cloud memory across sessions, so this code is part of a data exfiltration path by design. Although HTTPS is the default for the cloud URL, the endpoint is configurable and can be redirected, and the local daemon fallback uses plain HTTP, increasing the risk that sensitive prompts, code, or summaries are transmitted without sufficient user awareness or trust guarantees.

Content

No source excerpt is available for this finding.