space-image-studio
Fail
Audited by Gen Agent Trust Hub on Jul 4, 2026
Risk Level: HIGHCOMMAND_EXECUTIONDATA_EXFILTRATIONCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill is highly vulnerable to command injection in its image generation workflow. Step 5 executes a shell command that uses shell substitution (
$(cat /tmp/space-image-studio-prompt.txt)) to pass the prompt as an argument. Since the prompt includes unsanitized user input (titles, descriptions, etc.), an attacker can craft inputs with shell metacharacters (e.g.,;,$(...),`) to execute arbitrary code on the host system. - [COMMAND_EXECUTION]: The output path for generated images is constructed using a
{slug}variable derived from user-provided content:~/Documents/GitHub/obsidian/09 image/space-image-studio/{YYYY-MM-DD}/{type}-{slug}-{HHMMSS}.png. The absence of sanitization or validation on this variable allows for path traversal attacks, potentially enabling the skill to overwrite sensitive files outside the designated output directory. - [DATA_EXFILTRATION]: User-provided descriptions and the resulting prompts are sent to an external API (
https://api.labnana.com) for processing. This involves the transmission of potentially sensitive information to a third-party service not included in the platform's trusted domain list. - [CREDENTIALS_UNSAFE]: The skill relies on an API key stored in a
.labnana.envfile and instructs the agent tosourcethis file. This pattern loads secrets directly into the shell environment, exposing them to any subsequent processes executed in the same session. - [PROMPT_INJECTION]: The skill has an indirect prompt injection surface due to its ingestion and processing of untrusted user data that is later used in high-privilege shell operations.
- Ingestion points:
SKILL.mdStep 2 (user input for image content fields). - Boundary markers: No delimiters or safety warnings are present in the prompt templates (e.g.,
references/prompt-templates/base.md). - Capability inventory: Execution of shell commands in
SKILL.mdand file system write operations inscripts/generate_image.py. - Sanitization: The skill lacks any mechanism for escaping or validating user input before it is interpolated into shell commands or used in file path construction.
Recommendations
- AI detected serious security threats
Audit Metadata