agentpond-instrumentation
Warn
Audited by Gen Agent Trust Hub on Aug 2, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the user to execute 'eval "$(npx agentpond env get local)"' in the Files SDK reference. Evaluating the output of a dynamic CLI command directly in the shell is a security risk as it allows for arbitrary command execution if the tool's output is compromised.
- [EXTERNAL_DOWNLOADS]: The workflow requires installing various packages from external registries, such as @agentpond/firebase, @agentpond/supabase, and @agentpond/vercel, along with their dependencies.
- [REMOTE_CODE_EXECUTION]: The skill utilizes 'npx agentpond' commands, which involve fetching and executing code from a remote registry at runtime.
- [PROMPT_INJECTION]: The skill processes untrusted local files, such as package manifests and platform configuration files, to generate instrumentation proposals, creating a surface for indirect prompt injection.
- Ingestion points: File inspection in Phase 1 and platform-specific references (firebase.json, supabase/config.toml, .vercel/project.json).
- Boundary markers: Absent.
- Capability inventory: Shell execution (npx, eval) and file-writing capabilities.
- Sanitization: Not specified for the ingested configuration data.
Audit Metadata