Back to skill

Security audit

Video Sourcing Agent

Security checks for vulnerabilities and agentic risk

Overview

This skill appears purpose-built for video sourcing, but it runs downloaded code outside the sandbox with API keys and without strong integrity checks.

Install only if you trust the Memories.ai Labs runtime and are comfortable with it running outside the sandbox. Use restricted API keys, avoid exposing unrelated secrets in the environment, and prefer reviewing or pinning the runtime to an immutable commit before use.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • 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
  • 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)

T03 · Remote Payload Retrieval and Execution

Warning
Location
scripts/run_video_query.sh:57
Finding

Mutable Remote Runtime Is Retrieved and Executed with API Credentials

Content
View full analysis
/dev/null 2>&1 || { log_error "Failed to clone ${PINNED_REPO_URL} at tag ${PINNED_TAG}." log_error "Check network access and that the release tag exists." exit 2 } else ( cd "${MANAGED_RELEASE_DIR}" git fetch --depth 1 origin "refs/tags/${PINNED_TAG}:refs/tags/${PINNED_TAG}" >/dev/null 2>&1 || true git checkout --detach "${PINNED_TAG}" >/dev/null 2>&1 || { log_error "Failed to checkout pinned runtime tag ${PINNED_TAG}." exit 2 } ) fi ( cd "${MANAGED_RELEASE_DIR}" uv sync --frozen --no-dev >/dev/null 2>&1 || { log_error "Failed to install runtime dependencies with uv sync." exit 2 } ) } ``` ```bash main() { require_binary uv require_env_key GOOGLE_API_KEY require_env_key YOUTUBE_API_KEY RUNTIME_ROOT="$(resolve_runtime_root)" cd "${RUNTIME_ROOT}" uv run python -m video_sourcing_agent.integrations.openclaw_runner "$@" } main "$@" ``` ### Technical Analysis The Skill retrieves application code from an external Git repository and executes it on the host. Although the source is labeled as pinned to `v0.2.3`, a Git tag is not an immutable integrity identifier: an upstream maintainer or an attacker controlling the repository can move or replace t ...[truncated 2978 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (5)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The free-form trigger condition is broad enough that ordinary user requests about videos could implicitly activate this skill and cause host-side execution. Because the skill explicitly runs outside the sandbox and may bootstrap code plus use API keys, ambiguous activation increases the chance of unintended execution of privileged behavior without clear user consent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill states that it expects host runtime execution with sandbox mode off and can auto-bootstrap a runtime, but it does not prominently warn the user that external code may be fetched and run with access to local tooling and API keys. This creates a dangerous trust gap: users may invoke the skill assuming normal in-sandbox behavior when it actually expands execution privileges and secret exposure risk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The script self-bootstrapstraps by cloning code from a remote GitHub repository and installing its dependencies at runtime, which expands the trust boundary far beyond a deterministic local wrapper. Even with a pinned tag, there is no cryptographic verification of the fetched source or lockfile contents, so repository compromise, tag retargeting, dependency confusion, or a malicious upstream release could lead to arbitrary code execution on the host.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
80% confidence
Finding

For a skill presented as a deterministic /video_sourcing chat entrypoint, requiring git and uv on the host and reading API secrets from environment variables introduces execution and credential-dependency capabilities beyond a narrow chat wrapper. While the API keys may be expected for video sourcing itself, the host bootstrap/tooling dependency is not clearly justified by the stated purpose alone.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The script silently clones and prepares remote code without clear user-facing disclosure, which can mislead users into believing they are running only a concise local skill wrapper. In this context, hidden network retrieval and dependency installation reduce informed consent and make supply-chain compromise or unexpected host changes more dangerous.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.