Back to skill

Security audit

Ourmem

Security checks for vulnerabilities and agentic risk

Overview

This appears to be a real shared-memory integration, but it handles credentials, imports, and sharing with too much automatic authority and too little user review.

Review carefully before installing. Use a self-hosted or trusted endpoint when possible, do not paste another user's API key for sharing, avoid displaying full API keys in chat, preview any files before importing, and pin or verify packages, Docker images, and release binaries before running them.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
Findings (4)

T03 · Remote Payload Retrieval and Execution

Error
Location
references/selfhost-setup.md:57
Finding

Mutable, Unverified Remote Payloads Are Downloaded and Executed

Content
View full analysis
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Error
Location
SKILL.md:62
Finding

Bearer API Keys Are Reused and Requested as Cross-User Identifiers

Content
View full analysis
agent needs Bob's tenant ID (API key) 2. If the agent doesn't know Bob's ID, ask the user for it 3. Call `share-to-user` with the memory ID and Bob's tenant ID 4. Report: "Shared to Bob via team space {space_id}. Bob can now find it when searching." ``` ### Technical Analysis The same UUID is used both as a bearer authentication secret and as an identifier that other users must provide for sharing. These roles require different security properties: an identifier must be safe to disclose, while a bearer credential must remain confidential. The Skill explicitly tells agents to ask for another user's API key when initiating a share. Supplying that value through chat, agent context, logs, tool invocations, or stored memory exposes a credential that can authenticate to the recipient's entire memory account. The requested privilege therefore exceeds what is necessary to identify a sharing recipient. This is a least-privilege and access- ...[truncated 1326 chars]
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Error
Location
SKILL.md:253
Finding

Automatic Workspace Scanning Can Bulk-Upload Sensitive Session Files Without File-Level Approval

Content
View full analysis
find memory files -> upload 20 most recent via `/v1/imports` in parallel -> report results. ``` ### Technical Analysis The workflow interprets a general request to import memories as permission to scan the workspace, locate existing memory or session files, select the 20 most recent, and upload them in parallel. It does not require the agent to first present the discovered paths, preview their contents, distinguish hosted from self-hosted destinations, run a secret scan, or obtain per-file approval. Memory and session files commonly contain prompts, source code, filesystem paths, personal information, access tokens, API keys, proprietary material, and unrelated conversations. Recency does not establish relevance or authorization. Parallel asynchronous import further reduces the opportunity to stop an inappropriate transfer before server-side processing begins. Although the user initiates an import operation, the breadth of automatic discovery and transmission exceeds the minimum access necessary for a controlled file import. ### Attack Path 1. The user asks to import memories without naming specific files. 2. The agent follows the Skill and scans the workspace or platform-specific session storage. 3. The agent selects the 20 most recent matching files, including potentially unrelated or sensitive sessions. 4. The files are uploaded in parallel to `/v1/imports`. 5. The hosted or self-hosted service asynchronously stores, parses, extracts, and reconciles their contents. 6. Sensitive information becomes available to the memory service before the user receives the results or can meaningfully ...[truncated 679 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:166
Finding

Setup Handoff Requires Displaying the Full Bearer API Key in Chat

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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (42)

Tp4

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

The description promises a broad persistent-memory skill with onboarding, storage/recall, sharing, analytics, import/export, and deployment-related use cases. The actual code chunk is much narrower: it is a diagnostic script for verifying that an ourmem service is reachable, authenticated, and able to store/search/delete a test memory. While this is related to 'memory not working' or setup validation, it does not substantiate the broader declared functionality and has a materially different primary purpose from the full skill description. Therefore this chunk does not accurately represent the declared behavior.

Content

No source excerpt is available for this finding.

Vague Triggers

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The trigger phrases are broad everyday language such as 'remember this' and 'save this for later', which can cause the skill to activate in many normal conversations where persistent storage was not intended. In this skill, activation can lead to networked persistence and sharing behavior, so over-triggering materially raises privacy and consent risks.

Content

No source excerpt is available for this finding.

Vague Triggers

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The manifest explicitly instructs activation even when the user does not mention the product, based only on ambiguous memory-related intent. Given the skill's ability to persist data across sessions and share it externally, this increases the risk of silently turning ordinary chat content into long-lived or shared records.

Content

No source excerpt is available for this finding.

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
92% confidence
Finding

The skill directs modification of agent configuration files such as ~/.claude/settings.json, which are sensitive because they can alter agent behavior and often contain secrets or environment configuration. Changing files in an agent config directory is high risk if not tightly scoped and explicitly approved, as it can persist capabilities or leak credentials.

Content

Scanner excerpt · SKILL.md (reported line 157)May include surrounding context.

md
| Platform | Install | Config |
|----------|---------|--------|
| Claude Code | `/plugin marketplace add ourmem/omem` | `~/.claude/settings.json` env field |
| OpenCode | `"plugin": ["@ourmem/opencode"]` in opencode.json | `plugin_config` in opencode.json |
| OpenClaw | `openclaw plugins install @ourmem/ourmem` | `openclaw.json` with apiUrl + apiKey |
| MCP | `npx -y @ourmem/mcp` in MCP config | `OMEM_API_URL` + `OMEM_API_KEY` in env block |

Self-Modification

High
Category
Rogue Agent
Confidence
90% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · SKILL.md (reported line 340)May include surrounding context.

md
## Update

Do not set up automatic daily self-updates for this skill.

Only update the local skill file when the user or maintainer explicitly asks for a refresh from a reviewed source.

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
94% confidence
Finding

The skill directs the agent to access and modify ~/.claude/settings.json, a sensitive agent configuration location. Access to agent config directories is high risk because it can alter agent behavior, persist secrets, and create lasting changes outside the current task scope.

Content

Scanner excerpt · references/hosted-setup.md (reported line 35)May include surrounding context.

Claude Code

[AGENT] Write credentials to ~/.claude/settings.json (Claude Code's native config):

bash
python3 -c "

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
93% confidence
Finding

The embedded Python command opens and rewrites ~/.claude/settings.json, confirming direct programmatic modification of a privileged agent configuration file. Such automated access is dangerous because it can silently persist behavior changes and secrets in a trusted location that influences future agent sessions.

Content

Scanner excerpt · references/hosted-setup.md (reported line 40)May include surrounding context.

bash
python3 -c "
import json, os
p = os.path.expanduser('~/.claude/settings.json')
try: d = json.load(open(p))
except: d = {}
d.setdefault('env', {})

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · references/selfhost-setup.md (reported line 38)May include surrounding context.

bash
git clone https://github.com/ourmem/omem.git
cd omem
cp .env.example .env
docker-compose up -d

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
90% confidence
Finding

The instructions target an agent configuration file in the user's home directory and direct the agent to modify it with credentials. Accessing and changing agent config paths is sensitive because it establishes durable trust/configuration state that can outlive the current session and may be abused if done without clear consent and safeguards.

Content

Scanner excerpt · references/selfhost-setup.md (reported line 156)May include surrounding context.

Claude Code

[AGENT] Write credentials to ~/.claude/settings.json (Claude Code's native config):

bash
python3 -c "

Agent Config Directory Access

High
Category
Agent Snooping
Confidence
90% confidence
Finding

The embedded Python snippet programmatically edits ~/.claude/settings.json, which is a high-trust configuration location. Automating writes to such files can silently alter agent behavior and persist sensitive values beyond the intended setup flow.

Content

Scanner excerpt · references/selfhost-setup.md (reported line 161)May include surrounding context.

bash
python3 -c "
import json, os
p = os.path.expanduser('~/.claude/settings.json')
try: d = json.load(open(p))
except: d = {}
d.setdefault('env', {})

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
95% confidence
Finding

The skill clearly instructs the agent to run shell commands such as curl, npx, and plugin installers, but it does not declare an explicit tool scope or allowed-tools boundary. That increases the chance an agent executes command-capable actions without clear sandboxing or consent expectations, especially during onboarding and troubleshooting flows.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
90% confidence
Finding

Using an unpinned npx package reference means the executed code can change over time and may resolve to a compromised or unexpected version at install time. This creates a software supply-chain risk because the skill encourages direct execution of remote package code without version immutability.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The skill recommends running npx -y @ourmem/mcp without a pinned version, which executes remote package code immediately and non-interactively. Because this is an MCP component that may be granted broad agent access, a compromised upstream package could result in arbitrary code execution or credential exposure.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
91% confidence
Finding

The cross-user sharing API transmits stored memory content to another user's space via a remote service. Although sharing is a core product feature, it is still a real external transmission risk because mistaken identity, wrong target user IDs, or overbroad sharing could disclose sensitive data beyond the original user boundary.

Content

Scanner excerpt · SKILL.md (reported line 219)May include surrounding context.

The agent should call share-to-user which auto-creates a bridging Team Space if needed, adds the target user as a member, and shares the memory in one step.

bash
curl -sX POST "$API_URL/v1/memories/MEMORY_ID/share-to-user" \
  -H "Content-Type: application/json" -H "X-API-Key: $KEY" \
  -d '{"target_user": "TARGET_USER_TENANT_ID"}'
# Returns: { "space_id": "team/xxx", "shared_copy_id": "yyy", "space_created": true }

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill tells the agent to scan the user's workspace for memory/session files and upload them, but it does not require an explicit warning or informed consent before accessing local data. This can expose sensitive project files, transcripts, or metadata to the agent and then to the remote memory service.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · references/api-quick-ref.md (reported line 13)May include surrounding context.

bash
# Create tenant (returns API key)
curl -sX POST $API_URL/v1/tenants \
  -H "Content-Type: application/json" \
  -d '{"name": "my-workspace"}'
# -> {"id": "...", "api_key": "...", "status": "active"}

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The quick reference documents cross-space and cross-user sharing endpoints without any warning about privacy boundaries, authorization expectations, or the risk of propagating sensitive memories to other users or teams. In a memory-sharing product, examples that normalize broad sharing can lead integrators to expose personal, confidential, or tenant-scoped data through misuse or overbroad automation.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The GitHub connector example encourages submission of an access token directly in a request body but does not warn that this credential is highly sensitive and must be stored, rotated, and scoped carefully. Users may copy this pattern into logs, client-side code, or insecure transport paths, leading to repository compromise if the token is exposed.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · references/api-quick-ref.md (reported line 190)May include surrounding context.

md
# -> {"id": "...", "status": "completed|partial|failed", ...}

# Trigger intelligence on past import
curl -sX POST "$API_URL/v1/imports/IMPORT_ID/intelligence" -H "X-API-Key: $KEY"

# Cross-reconcile (discover relations via vector similarity)
curl -sX POST $API_URL/v1/imports/cross-reconcile -H "X-API-Key: $KEY"

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The rollback and batch deletion examples demonstrate broad destructive operations, including delete-all, without prominent warnings about data loss, confirmation semantics, or operational safeguards. In a persistent memory system, operators may accidentally erase large datasets or trigger irreversible cleanup through automation, copy-paste usage, or misunderstood filters.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · references/hosted-setup.md (reported line 10)May include surrounding context.

Step 1: Create API Key

bash
curl -sX POST https://api.ourmem.ai/v1/tenants \
  -H "Content-Type: application/json" \
  -d '{"name": "my-workspace"}' | jq .

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 · references/hosted-setup.md (reported line 10)May include surrounding context.

Step 1: Create API Key

bash
curl -sX POST https://api.ourmem.ai/v1/tenants \
  -H "Content-Type: application/json" \
  -d '{"name": "my-workspace"}' | jq .

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 · references/hosted-setup.md (reported line 198)May include surrounding context.

Step 1: Create API Key

bash
curl -sX POST https://api.ourmem.ai/v1/tenants \
  -H "Content-Type: application/json" \
  -d '{"name": "my-workspace"}' | jq .

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 · references/hosted-setup.md (reported line 202)May include surrounding context.

Step 1: Create API Key

bash
curl -sX POST https://api.ourmem.ai/v1/tenants \
  -H "Content-Type: application/json" \
  -d '{"name": "my-workspace"}' | jq .

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 · references/hosted-setup.md (reported line 205)May include surrounding context.

Step 1: Create API Key

bash
curl -sX POST https://api.ourmem.ai/v1/tenants \
  -H "Content-Type: application/json" \
  -d '{"name": "my-workspace"}' | jq .