reviewing-presentation
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied slide decks and 'Deck Folders,' which are explicitly identified as untrusted input. Malicious instructions could potentially be embedded within speaker notes or slide content to influence the agent's behavior during the review process. However, the skill instructions include specific guidance to review the content without making unauthorized changes and to treat the input as untrusted.\n
- Ingestion points:
deck.md,brief.md, and other files within the user-specified folder.\n - Boundary markers: The agent is instructed to 'Review the deck as it is; change nothing unless the person asks'.\n
- Capability inventory: Reading local files via Node scripts and reporting findings to the conversation. The provided scripts do not have network or file-write capabilities.\n
- Sanitization: The
scripts/deck.mjsfile uses specific regular expressions to parse content rather than executing or evaluating the input.\n- [COMMAND_EXECUTION]: The skill instructs the agent to execute a local Node.js script (scripts/reviewing-presentation.mjs) to estimate speaking time. To mitigate command injection risks, the instructions require the agent to use absolute paths and forbid the use of shell chaining characters (&&,$(...)) or heredocs when invoking the script.
Audit Metadata