knowledge-fetch
Pass
Audited by Gen Agent Trust Hub on Jul 25, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes local Python scripts including
candidate-store.py,fetch-cache.py, andrun-metrics.pyvia shell commands to manage project metadata and cache statistics. These scripts are executed within the defined plugin environment. - [EXTERNAL_DOWNLOADS]: The skill dispatches the
source-fetchertask to retrieve content from remote URLs identified during the curation phase. While the skill delegates the network requests, it handles the batching and processing of these external resources. - [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) due to its handling of external data.
- Ingestion points: The skill reads untrusted URL data and status flags from
.metadata/candidates.json. - Boundary markers: The instructions do not define boundary markers or explicit isolation for the URLs passed to the
source-fetchertask. - Capability inventory: The skill has access to Bash, Write, and Task capabilities, which are used to execute scripts and dispatch other agents.
- Sanitization: There is no documented validation or sanitization of the URLs or metadata retrieved from the candidate store before they are used in subsequent task dispatches.
Audit Metadata