novoads-update
Pass
Audited by Gen Agent Trust Hub on Sep 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes standard git commands (
git fetch,git rev-parse,git log,git diff) and a local repository script (./scripts/update.sh) to facilitate the update process. These actions are restricted to the repository root through explicit directory verification logic. - [COMMAND_EXECUTION]: The skill utilizes shell utilities including
mkdir,grep,printf, andmvto manage its internal state files located in the.update-state/directory. These operations are used to record and verify user consent for update behaviors. - [INDIRECT_PROMPT_INJECTION]: The skill represents a potential surface for indirect prompt injection as it ingests and summarizes data from external sources (git logs and the
CHANGELOG.mdfile). - Ingestion points: Commit messages and the content of
CHANGELOG.mdare read during the update phase inSKILL.md. - Boundary markers: Instructions specify that the agent should summarize the changes into a specific number of bullets, acting as a functional delimiter.
- Capability inventory: The skill is capable of executing the local
./scripts/update.shscript and writing to specific configuration files within the pack root. - Sanitization: The agent is tasked with summarizing the information for the user, which serves as a natural language filter for the ingested data.
Audit Metadata