Security audit
Security checks for vulnerabilities and agentic risk
This skill is a straightforward fal.ai integration, but users should know that prompts and selected files are sent to fal.ai and generated outputs are saved locally.
Install only if you are comfortable sending prompts, parameters, and any uploaded media files to fal.ai. Do not upload confidential files or sensitive personal media unless that use is approved, and periodically clean ~/.fal/sessions/${CLAUDE_SESSION_ID}/ if generated outputs should not remain on disk.
The upload command posts a user-supplied local file to a remote fal CDN without any embedded user-facing warning or confirmation requirement. Because local files may contain sensitive or proprietary data, silent upload behavior materially increases the risk of unintended exfiltration.
The README includes very broad natural-language examples such as 'Generate an image of a sunset over mountains' and 'Turn this photo into a video,' which can encourage automatic invocation of the skill without the user explicitly choosing the external fal.ai service. In a skill that sends prompts and potentially files to a third-party API and downloads outputs locally, overly generic activation examples increase the chance of unintended external data transfer and side effects.
The README describes uploading files, submitting jobs, and polling fal.ai, but it does not clearly warn that prompts, media, and uploaded files may be transmitted to external fal.ai services. In this context, omission of a third-party data-transfer warning is dangerous because users may provide sensitive images, audio, or text without realizing that the content leaves the local environment.
The README describes uploading files, submitting jobs, and polling fal.ai, but it does not clearly warn that prompts, media, and uploaded files may be transmitted to external fal.ai services. In this context, omission of a third-party data-transfer warning is dangerous because users may provide sensitive images, audio, or text without realizing that the content leaves the local environment.
The skill persists generated media to a session-scoped directory under the user's home folder, which can retain potentially sensitive images or videos beyond the immediate interaction. Without retention guidance, permission controls, or cleanup behavior, this increases exposure to local data leakage or unintended reuse.
Save generated files to session folder:
mkdir -p ~/.fal/sessions/${CLAUDE_SESSION_ID}
Downloaded images/videos go to: ~/.fal/sessions/${CLAUDE_SESSION_ID}/
The skill sends user prompts, model inputs, and likely generated outputs to fal.ai APIs, but the description does not clearly warn users that their data leaves the local environment and is processed by a third party. This creates a privacy and consent risk, especially if users provide sensitive text, images, audio, or other files under the assumption the operation is local.
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
Search for models matching $1:
curl -s "https://api.fal.ai/v1/models?q=$1&limit=15" \
-H "Authorization: Key $FAL_KEY" | jq -r '.models[] | "• \(.endpoint_id) — \(.metadata.display_name) [\(.metadata.category)]"'
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
Search for models matching $1:
curl -s "https://api.fal.ai/v1/models?q=$1&limit=15" \
-H "Authorization: Key $FAL_KEY" | jq -r '.models[] | "• \(.endpoint_id) — \(.metadata.display_name) [\(.metadata.category)]"'
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
Search for models matching $1:
curl -s "https://api.fal.ai/v1/models?q=$1&limit=15" \
-H "Authorization: Key $FAL_KEY" | jq -r '.models[] | "• \(.endpoint_id) — \(.metadata.display_name) [\(.metadata.category)]"'
This command sends a JSON payload to fal.ai's remote queue endpoint, which is expected functionality for a hosted AI service but still constitutes external data transmission. In this skill context, prompts and other parameters may contain sensitive information, so the lack of explicit guardrails and disclosure makes the transmission security-relevant.
Step 2: Submit to queue
curl -s -X POST "https://queue.fal.run/$1" \
-H "Authorization: Key $FAL_KEY" \
-H "Content-Type: application/json" \
-d '<JSON_PAYLOAD>'
The result-handling flow explicitly instructs saving downloaded images or videos to persistent session storage, but does not mention retention, cleanup, or sensitivity of stored outputs. Since generated media can contain private user content or proprietary prompts rendered into assets, disk persistence is a meaningful confidentiality risk.
RESULT=$(curl -s "https://queue.fal.run/$1/requests/$REQUEST_ID" \
-H "Authorization: Key $FAL_KEY")
# Create session output folder
mkdir -p ~/.fal/sessions/${CLAUDE_SESSION_ID}
# Download images/videos
This markdown file includes a natural-language/config example that sets "language": "en" for Whisper input. Because the document does not explain that language should be selected based on user preference or locale, it can encourage a fixed-language behavior that conflicts with language/locale choice expectations.
No suspicious patterns detected.