Back to skill

Security audit

Smara Memory

Security checks for vulnerabilities and agentic risk

Overview

This skill does what it claims, but it can automatically store and retrieve user context in a third-party memory service without enough user control or sensitivity limits.

Review before installing. Use this only if users understand that selected conversation facts and lookup queries may be sent to Smara and stored across sessions. Configure clear opt-in, avoid storing secrets or sensitive personal information, prefer narrow searches over full-context retrieval, and confirm destructive deletes.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

other

Warning
Location
SKILL.md:36
Finding

Unbounded Third-Party Storage of Conversation Data

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 36–44 and 64–65
Vulnerability Type: other: Privacy Data Disclosure
Risk Level: Medium

Vulnerable Code

bash
curl -X POST https://api.smara.io/v1/memories \
  -H "Authorization: Bearer $SMARA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "user_id": "user_id_here",
    "fact": "User prefers dark mode and uses vim keybindings",
    "importance": 0.7
  }'
markdown
1. **After conversations**: Extract key facts (preferences, decisions, context) and store them as memories with relevant tags
2. **Before responding**: Search for relevant memories to personalize responses

Technical Analysis

The Skill instructs the agent to extract facts, preferences, decisions, and context from conversations and transmit them to the external Smara service at api.smara.io. Although third-party persistence is part of the declared functionality, the storage instruction is broad and does not require informed user consent, review of the proposed memory, or data minimization.

No instructions prevent the agent from storing credentials, authentication tokens, private communications, health information, financial details, precise identifying information, or other sensitive content. The Skill also lacks retention limits, purpose restrictions, redaction requirements, and guidance for deleting all information associated with a user.

The documented API credential is transmitted only to the declared HTTPS endpoint, and no credential theft was identified. The issue is the potentially unbounded disclosure and retention of conversation-derived data rather than misuse of the API key itself.

Attack Path

  1. A user provides personal or sensitive information during a conversation.
  2. The agent interprets that information as a relevant fact, preference, decision, or contextual detail under the broad storage instruction.
  3. The agent send ...[truncated 1180 chars]
Remediation
View remediation

Remediation Suggestions

  1. Require explicit, informed user opt-in before enabling external persistent memory.
  2. Show the exact proposed memory, destination service, and associated user identifier before transmission, and require confirmation for sensitive or ambiguous content.
  3. Define a strict allowlist of information suitable for storage and prohibit secrets, credentials, authentication data, financial information, health information, private communications, and other sensitive personal data.
  4. Minimize stored content by redacting unnecessary identifiers and converting free-form conversation excerpts into narrowly scoped facts.
  5. Add retention limits and automatic expiration appropriate to the memory's purpose.
  6. Provide users with mechanisms to list, correct, export, and delete all stored memories, including account-wide deletion rather than only deletion by individual memory ID.
  7. Require confirmation before retrieving or applying full user context in a new conversation or materially different context.
  8. Document the third-party service's privacy, retention, access-control, and deletion behavior so users can make an informed decision.
  9. Log storage and retrieval operations without recording the sensitive memory contents or API credential.
  10. Apply least privilege to the Smara API key and rotate or revoke it if exposure is suspected.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (9)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly encourages storing user facts and context in a third-party memory service but does not instruct the agent to obtain user consent, minimize sensitive data, or disclose external sharing. This creates a real privacy and data-governance risk because personal information may be retained and transmitted off-platform without the user's awareness.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
93% confidence
Finding

This example sends user facts to an external API endpoint using an authorization token, which is a true external transmission path for conversational data. In this skill's context, the transfer is the core feature, but it is still security-relevant because it can expose personal data to a third party if used without consent and proper scoping.

Content

Scanner excerpt · SKILL.md (reported line 41)May include surrounding context.

Store a memory

bash
curl -X POST https://api.smara.io/v1/memories \
  -H "Authorization: Bearer $SMARA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{

External Transmission

Medium
Category
Data Exfiltration
Confidence
93% confidence
Finding

This example sends user facts to an external API endpoint using an authorization token, which is a true external transmission path for conversational data. In this skill's context, the transfer is the core feature, but it is still security-relevant because it can expose personal data to a third party if used without consent and proper scoping.

Content

Scanner excerpt · SKILL.md (reported line 41)May include surrounding context.

Store a memory

bash
curl -X POST https://api.smara.io/v1/memories \
  -H "Authorization: Bearer $SMARA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{

External Transmission

Medium
Category
Data Exfiltration
Confidence
88% confidence
Finding

The search endpoint transmits user identifiers and queries to an external service, which can reveal interests, preferences, or other personal context even if it is not creating new records. Because this skill uses retrieval to personalize responses, the external lookup increases privacy exposure and should be treated as a real security concern.

Content

Scanner excerpt · SKILL.md (reported line 54)May include surrounding context.

Search memories

bash
curl "https://api.smara.io/v1/memories/search?user_id=user_id_here&query=editor+preferences&limit=5" \
  -H "Authorization: Bearer $SMARA_API_KEY"

External Transmission

Medium
Category
Data Exfiltration
Confidence
95% confidence
Finding

The 'get full user context' action requests a broad set of stored information from an external service, which materially increases the blast radius of any misuse or mis-scoping. Broad retrieval is more dangerous than targeted lookup because it can surface unrelated or stale personal data and make accidental disclosure in responses more likely.

Content

Scanner excerpt · SKILL.md (reported line 61)May include surrounding context.

Get full user context

bash
curl "https://api.smara.io/v1/users/user_id_here/context" \
  -H "Authorization: Bearer $SMARA_API_KEY"

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 68)May include surrounding context.

Delete a memory

bash
curl -X DELETE "https://api.smara.io/v1/memories/MEMORY_ID" \
  -H "Authorization: Bearer $SMARA_API_KEY"

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 94)May include surrounding context.

Delete a memory

bash
curl -X DELETE "https://api.smara.io/v1/memories/MEMORY_ID" \
  -H "Authorization: Bearer $SMARA_API_KEY"

Ssd 3

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The instructions tell the agent to persist conversational facts and retrieve full user context for future responses, establishing a durable retention-and-recall pattern for natural-language personal data. Without constraints on consent, sensitivity, tenant separation, or least-privilege retrieval, this increases the chance of overcollection, excessive retention, and accidental exposure of prior user information.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

The delete example is a destructive operation presented without guardrails such as confirmation, scope validation, or warning that deletion may be irreversible from the user's perspective. While lower risk than data exfiltration, it can still lead to unintended loss of stored memory or user context.

Content

No source excerpt is available for this finding.