ClawHub
Back to skill

Security audit

the-librarian

Security checks for vulnerabilities and agentic risk

Overview

The main skill is a document search tool, but the package also includes local instructions for privileged Supabase access to private conversations and contacts, so it needs careful review before installation.

Review this before installing. Use only with documents you are comfortable storing in the generated index and sending to the configured embedding endpoint. Do not use the packaged local Supabase instructions unless you intentionally want an agent to access private conversations, contacts, and a service-role database credential; replace that with scoped read-only credentials first.

A.I.G

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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 (5)

T05 · Unauthorized Access and Privilege Escalation

Error
Location
SKILL.local.md:20
Finding

Supabase Service-Role Credential Access Violates Least Privilege

Content
View full analysis
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Error
Location
SKILL.local.md:24
Finding

Skill Scope Includes Broad Access to Private Conversations and Contacts

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/build_index.py:52
Finding

Document Contents Can Be Sent to an Arbitrary Embedding Endpoint

Content
View full analysis
list[float]: """Get embedding from Ollama API.""" response = requests.post( f"{api_url}/api/embeddings", json={"model": model, "prompt": text}, timeout=120 ) response.raise_for_status() return response.json()["embedding"] ``` The endpoint is user-selectable: ```python parser.add_argument("--api", "-a", default=DEFAULT_OLLAMA_API, help=f"Ollama API URL (default: {DEFAULT_OLLAMA_API})") ``` ### Technical Analysis The index builder sends every plaintext document chunk in the `prompt` field of an HTTP POST request. The endpoint is controlled by the `--api` argument. The safety function only prints a warning; it neither rejects the destination nor requires affirmative consent before transmission. The validation treats all `10.*` and `192.168.*` destinations as safe even though private-network ownership does not establish trust. It also includes `0.0.0.0` as a safe destination and does ...[truncated 1668 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/search.py:62
Finding

Sensitive Search Queries Can Be Sent to an Arbitrary HTTP Endpoint

Content
View full analysis
list[float]: """Get embedding from Ollama API.""" response = requests.post( f"{api_url}/api/embeddings", json={"model": model, "prompt": text}, timeout=60 ) response.raise_for_status() return response.json()["embedding"] ``` The query reaches this sink through vector search: ```python def vector_search(query: str, index, chunks: list[dict], model: str, api_url: str, top_k: int = 5) -> tuple: """Pure vector search with TurboVec.""" # Get query embedding query_emb = get_embedding(query, model, api_url) ``` ### Technical Analysis Raw search queries are sent to the configured embedding endpoint. Queries can contain names, legal matters, health information, trade secrets, credentials, customer identifiers, or other sensitive intent. As in the index-building path, the endpoint check only displays a warning. Execution proceeds automatically, all `10.*` and `192.168.*` addresses are implicitly ...[truncated 1310 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
SKILL.md:40
Finding

Unpinned Third-Party Dependencies Create Supply-Chain Exposure

Content
View full analysis
Remediation
View remediation

SkillSpector

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
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • YARA SignaturesMalware Match, Webshell Match, Cryptominer Match
Findings (21)

Credential Access

High
Category
Privilege Escalation
Confidence
99% confidence
Finding

The documentation shows how to extract a Supabase service-role key directly from a local .env file and use it in REST calls. Service-role keys are highly privileged secrets; exposing their file path and encouraging shell extraction increases the risk of credential disclosure, misuse, and full database compromise if the skill is invoked or copied in an unsafe environment.

Content

Scanner excerpt · SKILL.local.md (reported line 50)May include surrounding context.

REST API Examples

bash
# Full-text search
SERVICE_ROLE_KEY=$(grep SERVICE_ROLE_KEY /home/endar/docker-stack/supabase/.env | cut -d= -f2)
curl "http://host.docker.internal:8000/rest/v1/knowledge_entries?content=ilike.*GEM*&select=id,title,category" \
  -H "apikey: $SERVICE_ROLE_KEY" \
  -H "Authorization: Bearer $SERVICE_ROLE_KEY"

YARA rule 'backdoor_persistence': Backdoor persistence with malicious payloads (shell commands, SSH key injection, hidden root users) [malware]

High
Category
YARA Match
Confidence
75% confidence
Finding

YARA rule matched a known malware signature (reverse shell, backdoor, ransomware, C2 framework, or info stealer).

Content

Scanner excerpt · scripts/build_index.py (reported line 30)May include surrounding context.

python
E = 'turbovec'
except ImportError:
    print("ERROR: TurboVec not available. Install with: pip install turbovec")
    print("BLAS library required. Install with: sudo apt install libblas3")
    print("")
    print("Run via the wrapper script (recommended):")
    print("  ./scripts/librarian.sh build <input> <output>")
    print("")
    print("Or set LD_PRELOAD before Python starts:")
    print("  LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libblas.so.3 python build_index.py ...")
    sys.exit(1)

import numpy as np
import requests

# Optional: BM25 for hybrid search
try:
    from rank_bm25 import BM25Okapi
    BM25_AVAILABLE = True
except ImportError:
    BM25_AVAILABLE = False
    print("Note: BM25 not available. Install with: pip install rank-bm25")

# Configuration
DEFAULT_OLLAMA_API = "http://host.docker.internal:11434"
DEFAULT_MODEL = "nomic-embed-text:v1.5"
CHUNK_SIZE = 650  # Characters per chunk
CHUNK_OVERLAP = 100  # Characters overlap between chunks
BIT_WIDTH = 4  # 2, 3, or 4 bits

YARA rule 'backdoor_persistence': Backdoor persistence with malicious payloads (shell commands, SSH key injection, hidden root users) [malware]

High
Category
YARA Match
Confidence
75% confidence
Finding

YARA rule matched a known malware signature (reverse shell, backdoor, ransomware, C2 framework, or info stealer).

Content

Scanner excerpt · SKILL.local.md (reported line 168)May include surrounding context.

md
E = 'turbovec'
except ImportError:
    print("ERROR: TurboVec not available. Install with: pip install turbovec")
    print("BLAS library required. Install with: sudo apt install libblas3")
    print("")
    print("Run via the wrapper script (recommended):")
    print("  ./scripts/librarian.sh search <query> <index>")
    print("")
    print("Or set LD_PRELOAD before Python starts:")
    print("  LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libblas.so.3 python search.py ...")
    sys.exit(1)

import numpy as np
import requests

# Optional: BM25 for hybrid search
try:
    from rank_bm25 import BM25Okapi
    BM25_AVAILABLE = True
except ImportError:
    BM25_AVAILABLE = False

# Optional: Flashrank for reranking
try:
    from flashrank import Ranker, RerankRequest
    FLASHRANK_AVAILABLE = True
except ImportError:
    FLASHRANK_AVAILABLE = False

# Configuration
DEFAULT_OLLAMA_API = "http://host.docker.internal:11434"
DEFAULT_MODEL = "nomic-embed-text:v1.5"

# Security: Warn if embedding API is

YARA rule 'backdoor_persistence': Backdoor persistence with malicious payloads (shell commands, SSH key injection, hidden root users) [malware]

High
Category
YARA Match
Confidence
75% confidence
Finding

YARA rule matched a known malware signature (reverse shell, backdoor, ransomware, C2 framework, or info stealer).

Content

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

md
E = 'turbovec'
except ImportError:
    print("ERROR: TurboVec not available. Install with: pip install turbovec")
    print("BLAS library required. Install with: sudo apt install libblas3")
    print("")
    print("Run via the wrapper script (recommended):")
    print("  ./scripts/librarian.sh search <query> <index>")
    print("")
    print("Or set LD_PRELOAD before Python starts:")
    print("  LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libblas.so.3 python search.py ...")
    sys.exit(1)

import numpy as np
import requests

# Optional: BM25 for hybrid search
try:
    from rank_bm25 import BM25Okapi
    BM25_AVAILABLE = True
except ImportError:
    BM25_AVAILABLE = False

# Optional: Flashrank for reranking
try:
    from flashrank import Ranker, RerankRequest
    FLASHRANK_AVAILABLE = True
except ImportError:
    FLASHRANK_AVAILABLE = False

# Configuration
DEFAULT_OLLAMA_API = "http://host.docker.internal:11434"
DEFAULT_MODEL = "nomic-embed-text:v1.5"

# Security: Warn if embedding API is

YARA rule 'backdoor_persistence': Backdoor persistence with malicious payloads (shell commands, SSH key injection, hidden root users) [malware]

High
Category
YARA Match
Confidence
75% confidence
Finding

YARA rule matched a known malware signature (reverse shell, backdoor, ransomware, C2 framework, or info stealer).

Content

Scanner excerpt · scripts/search.py (reported line 35)May include surrounding context.

python
E = 'turbovec'
except ImportError:
    print("ERROR: TurboVec not available. Install with: pip install turbovec")
    print("BLAS library required. Install with: sudo apt install libblas3")
    print("")
    print("Run via the wrapper script (recommended):")
    print("  ./scripts/librarian.sh search <query> <index>")
    print("")
    print("Or set LD_PRELOAD before Python starts:")
    print("  LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libblas.so.3 python search.py ...")
    sys.exit(1)

import numpy as np
import requests

# Optional: BM25 for hybrid search
try:
    from rank_bm25 import BM25Okapi
    BM25_AVAILABLE = True
except ImportError:
    BM25_AVAILABLE = False

# Optional: Flashrank for reranking
try:
    from flashrank import Ranker, RerankRequest
    FLASHRANK_AVAILABLE = True
except ImportError:
    FLASHRANK_AVAILABLE = False

# Configuration
DEFAULT_OLLAMA_API = "http://host.docker.internal:11434"
DEFAULT_MODEL = "nomic-embed-text:v1.5"

# Security: Warn if embedding API is

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill metadata and opening description frame this as a lightweight document-indexing/search tool, but the body also exposes a broader Supabase-backed knowledge/conversation system with message ingestion, contacts, and scheduled review workflows. This scope expansion can cause an agent or user to invoke the skill in contexts where sensitive conversations and knowledge data are accessed or processed without clear expectation or consent.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger phrases include broad language like 'search my knowledge', 'what do we know about', and 'remember this', which can match routine user requests outside the intended narrow use case. That over-breadth raises the likelihood that the skill will activate inappropriately and access sensitive knowledge stores or conversations without clear user intent.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The documentation introduces WhatsApp ingestion, contact handling, and weekly review/summary workflows that are materially broader than a document-indexing skill. In practice, this increases the chance of over-collection and automated processing of personal data, especially if the skill is triggered for ordinary knowledge-search requests.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

Claiming 'Local, free, no API calls' while also documenting HTTP endpoints for embeddings and Supabase access is misleading. Even if the endpoints are local-network services, they are still API calls and may transmit sensitive document text or queries over HTTP, which can undermine user assumptions about privacy and data flow.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill describes automatic ingestion of WhatsApp conversations, contact population, and weekly summaries without an equally prominent user warning in the skill description. This creates a significant transparency and consent problem because users invoking a 'librarian' or document-search skill may not expect continual ingestion and summarization of personal communications.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
89% confidence
Finding

The skill describes capabilities that read and write local files and send document/query content over the network, but it does not declare any explicit tool scope such as permissions or allowed-tools. That mismatch can cause the agent runtime or reviewer to underestimate what the skill can access, increasing the risk of unintended file exposure or outbound data transfer.

Content

No source excerpt is available for this finding.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · SKILL.local.md (reported line 149)May include surrounding context.

bash
# Install BLAS library (required for TurboVec)
sudo apt install libblas3

# Create venv and install dependencies
cd /path/to/the-librarian

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · SKILL.md (reported line 42)May include surrounding context.

bash
# Install BLAS library (required for TurboVec)
sudo apt install libblas3

# Create venv and install dependencies
cd /path/to/the-librarian

Unrestricted Tool Access

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill grants unrestricted tool access without appropriate constraints. An agent with unfettered tool access can perform arbitrary actions including file modification, network requests, and code execution.

Content

Scanner excerpt · SKILL.local.md (reported line 240)May include surrounding context.

md
### Vulnerability History

**v1.1.0 (July 2026):**
- **Critical fix:** `pickle.load` on `bm25_index.pkl` removed. Pickle files can execute arbitrary code when loaded. BM25 index now stored as JSON. Legacy pickle files are detected but refused to load.
- **High fix:** Embedding API URL now checked against localhost. If a non-local API endpoint is configured, a privacy warning is displayed before any data is transmitted.
- **Medium fix:** SKILL.md now documents that document text and search queries are transmitted to the Ollama HTTP API for embedding generation.

Unrestricted Tool Access

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill grants unrestricted tool access without appropriate constraints. An agent with unfettered tool access can perform arbitrary actions including file modification, network requests, and code execution.

Content

Scanner excerpt · SKILL.md (reported line 141)May include surrounding context.

md
### Vulnerability History

**v1.1.0 (July 2026):**
- **Critical fix:** `pickle.load` on `bm25_index.pkl` removed. Pickle files can execute arbitrary code when loaded. BM25 index now stored as JSON. Legacy pickle files are detected but refused to load.
- **High fix:** Embedding API URL now checked against localhost. If a non-local API endpoint is configured, a privacy warning is displayed before any data is transmitted.
- **Medium fix:** SKILL.md now documents that document text and search queries are transmitted to the Ollama HTTP API for embedding generation.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · scripts/build_index.py (reported line 24)May include surrounding context.

python
ENGINE = 'turbovec'
except ImportError:
    print("ERROR: TurboVec not available. Install with: pip install turbovec")
    print("BLAS library required. Install with: sudo apt install libblas3")
    print("")
    print("Run via the wrapper script (recommended):")
    print("  ./scripts/librarian.sh build <input> <output>")

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · scripts/librarian.sh (reported line 26)May include surrounding context.

sh
ENGINE = 'turbovec'
except ImportError:
    print("ERROR: TurboVec not available. Install with: pip install turbovec")
    print("BLAS library required. Install with: sudo apt install libblas3")
    print("")
    print("Run via the wrapper script (recommended):")
    print("  ./scripts/librarian.sh build <input> <output>")

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · scripts/search.py (reported line 29)May include surrounding context.

python
ENGINE = 'turbovec'
except ImportError:
    print("ERROR: TurboVec not available. Install with: pip install turbovec")
    print("BLAS library required. Install with: sudo apt install libblas3")
    print("")
    print("Run via the wrapper script (recommended):")
    print("  ./scripts/librarian.sh build <input> <output>")

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The script sends each document chunk's full text to the configured embedding API, which can expose sensitive local documents to another service. Although the code warns on non-local URLs, the default endpoint uses host.docker.internal and the transfer still occurs over HTTP without authentication or content minimization.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
96% confidence
Finding

This POST request transmits raw document text in the prompt field to whatever API URL the user supplies, so secrets in indexed files may be disclosed to external systems. In the context of a local document indexing skill, that behavior is security-relevant because users may reasonably expect documents to stay local unless they explicitly choose otherwise.

Content

Scanner excerpt · scripts/build_index.py (reported line 69)May include surrounding context.

python
def get_embedding(text: str, model: str, api_url: str) -> list[float]:
    """Get embedding from Ollama API."""
    response = requests.post(
        f"{api_url}/api/embeddings",
        json={"model": model, "prompt": text},
        timeout=120

External Transmission

Medium
Category
Data Exfiltration
Confidence
95% confidence
Finding

The script sends the user's query text to an embedding API via HTTP, and the configured endpoint can be changed to a non-local host. Although the code warns on non-local destinations, it does not enforce localhost, HTTPS, authentication, or explicit user confirmation, so indexed content and search queries may be disclosed to remote services.

Content

Scanner excerpt · scripts/search.py (reported line 77)May include surrounding context.

python
def get_embedding(text: str, model: str, api_url: str) -> list[float]:
    """Get embedding from Ollama API."""
    response = requests.post(
        f"{api_url}/api/embeddings",
        json={"model": model, "prompt": text},
        timeout=60