Back to skill

Security audit

Loom Vision

Security checks for vulnerabilities and agentic risk

Overview

The skill appears to do what it claims, but it can automatically download Loom videos and keep local copies, transcripts, and frames while relying on an unpinned downloader dependency.

Review before installing. Use it only for Looms you are comfortable downloading locally, set a private OUTPUT_ROOT when processing sensitive recordings, delete generated video/transcript/frame files after use, and prefer a pinned/reviewed loom-dl version. The artifacts do not show malicious behavior, but the current scoping and retention controls deserve user attention.

Vulnerability Patterns
  • 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
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Findings (3)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:17
Finding

Unpinned Third-Party Executable Dependency

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
process-loom.sh:58
Finding

Insufficient Validation of the Downloader URL

Content
View full analysis
/... VIDEO_ID=$(echo "$URL" | grep -oE '[a-f0-9]{32}' | head -1) if [[ -z "$VIDEO_ID" ]]; then echo "ERROR: Could not extract a 32-hex video ID from URL: $URL" >&2 echo " Expected format: https://www.loom.com/share/<32-hex-id>/..." >&2 exit 1 fi ``` The superficially validated value is later passed unchanged to the downloader: ```bash loom-dl --url "$URL" --out "$OUTPUT_DIR/video.mp4" --transcript >&2 ``` ### Technical Analysis The validation only checks whether the supplied string contains a lowercase 32-character hexadecimal substring. It does not parse the value as a URL or enforce: - The `https` scheme. - An approved Loom hostname. - The expected `/share/` path. - The absence of embedded credentials or unexpected ports. - An exact relationship between the extracted identifier and the URL path. For example, an attacker-controlled URL containing an arbitrary 32-character hexadecimal path component passes validation even if its hostname is unrelated to Loom. The entire original URL is then supplied to the network-capable `loom-dl` executable. The exact request behavior and supported schemes depend on `loom-dl`; however, the script itself does not enforce its documented Loom-only trust boundary. This can expose the runtime to unintended outbound requests and downloader-specific parsing behavior. ### Attack Path 1. An attacker supplies a URL such as: `https://attacker.example/0123456789abcdef0123456789abcdef` 2. The regular expression finds the 32-character hexadecimal substring. 3. The script derives an output directory from that substring and accepts the input. 4. The original attacker-controlled URL is passed unchanged ...[truncated 959 chars]
Remediation
View remediation
` and extract the identifier only from that path component. 5. Reject embedded credentials, unexpected ports, malformed encodings, backslashes, and ambiguous hostnames. 6. Review and restrict redirect handling. Every redirect destination should be revalidated against the same hostname and scheme policy. 7. Where possible, run the downloader with outbound network access restricted to required Loom endpoints. ]]>

T09 · Insecure Skill Coding Practices

Warning
Location
process-loom.sh:39
Finding

Predictable Shared Temporary Storage and Unsafe Output Reuse

Content
View full analysis
/dev/null ``` ### Technical Analysis The default output root is a predictable path under the system temporary directory, and each video uses a deterministic directory based on its public video identifier. The script uses `mkdir -p` rather than creating a unique directory atomically and does not: - Set a restrictive `umask`. - Verify ownership and permissions of existing directories. - Reject symbolic links in the output path. - Ensure that output files are newly created regular files. - Remove stale frames before reprocessing. - Automatically remove downloaded videos and transcripts after use. The generated video, transcript, and screenshots may contain source code, internal UI states, personal information, access details displayed on screen, or confidential business discussions. Their effective permissions depend on the invoking process's umask; common defaults can make newly created files readable by other local users. If an attacker can prepare a writable output hierarchy, replace a configured output root, or ...[truncated 1897 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (4)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The README explains that the skill downloads the source video, transcript, and sampled frames, but it does not warn users that these artifacts are stored locally under a temporary output directory. Because Loom videos often contain sensitive code, credentials, internal UI, or user data, omission of this disclosure increases the risk of unintentional retention and exposure of confidential material.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The README states the skill triggers automatically when a Loom URL is pasted or when the user asks about Loom video content, which is broad enough to cause unintended activation in normal conversation. For a skill that downloads remote content and stores video, transcript, and extracted frames locally, accidental invocation can lead to unnecessary data collection and processing of potentially sensitive recordings.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The trigger conditions are broad enough to auto-invoke on ordinary Loom-related requests, including cases where a user may only want discussion of a video rather than local processing. Because this skill downloads the video and writes transcripts and sampled frames to disk, over-broad triggering can cause unnecessary collection and retention of potentially sensitive audiovisual content without clear user awareness or 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 documentation tells the agent to run a processor that downloads the Loom video and persists the original video, transcript, raw transcript JSON, and extracted frames, but it does not include an explicit warning to the user that this local storage will occur. This creates a privacy and data-handling risk, especially for Loom videos that may contain proprietary code, credentials, internal dashboards, or other sensitive visual information.

Content

No source excerpt is available for this finding.