feed-digest
Pass
Audited by Gen Agent Trust Hub on Aug 25, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses several shell commands including
curl,grep,head, andpython3 -cto discover RSS feeds and fetch data from external APIs (Hacker News, Reddit, and iTunes). - [DATA_EXPOSURE]: The skill is instructed to read a configuration file (
digest.config.md) from the user's home directory. Accessing the home directory is a sensitive operation as it may expose other user data or preferences. - [INDIRECT_PROMPT_INJECTION]: The skill acts on data ingested from external, untrusted sources such as blogs, RSS feeds, and Reddit comments. This data could potentially contain malicious instructions intended to influence the agent's behavior during the digest creation process.
- Ingestion points: The skill reads external content through URLs resolved in the adapters (e.g.,
references/rss.md,references/blogs.md). - Boundary markers: No explicit boundary markers or "ignore instructions" prompts are used when interpolating external content into the agent's context.
- Capability inventory: The skill possesses the ability to execute shell commands for fetching and to generate arbitrary summarized text for the user.
- Sanitization: While the RSS adapter mentions stripping HTML tags, there is no broad validation or sanitization of the natural language content fetched from external sources.
Audit Metadata