focus-ai-brand
Warn
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill's
/reportcommand implementation inreferences/print.mdexecutes a PDF generation binary located at a computed path within the user's home directory (~/.claude/plugins/cache/focus-marketplace/chrome-driver/*/bin/pdf). This pattern of discovering and running executables from sensitive directories at runtime is a potential security risk.\n- [DYNAMIC_EXECUTION]: The/deckcommand described inreferences/presentations.mdinvolves generating and executing a custom JavaScript script using thepptxgenjslibrary to convert markdown content into PowerPoint slides.\n- [REMOTE_CODE_EXECUTION]: The skill requires the global installation of multiple Node.js packages (pptxgenjs,react,react-dom,react-icons,sharp) without version pinning. This practice introduces supply chain risks by pulling unverified code from public registries at installation time.\n- [INDIRECT_PROMPT_INJECTION]: The branding commands ingest external markdown files which serves as a potential vector for indirect prompt injection attacks.\n - Ingestion points: Markdown files passed via the
fileparameter in/reportand/deckcommands.\n - Boundary markers: None identified; no instructions are provided to the agent to treat input content as untrusted data or to use delimiters.\n
- Capability inventory: Shell command execution, execution of binaries from the home directory, file writing to
/tmp, and dynamic code generation.\n - Sanitization: The skill does not define any sanitization or validation logic for the input markdown data before it is processed by the generation tools.\n- [COMMAND_EXECUTION]: The skill uses shell commands (
ls,tail,sort) to identify and prepare the execution environment for document generation, which increases the complexity and potential for command injection if inputs are not properly handled.
Audit Metadata