recall
Pass
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: TROUBLESHOOTING.md instructs the agent to read local secret files (
~/.agentmemory/.envand~/.agentmemory/secret) to authenticate REST fallback requests. While it cautions against sending local-file secrets to remote URLs, it explicitly allows sending these credentials to loopback addresses (localhost,127.0.0.0/8, or[::1]). This creates a risk if an attacker can control the local network environment. - [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and display untrusted data from the
memory_smart_searchtool. - Ingestion points: External data enters via the
narrativeandtitlefields returned by thememory_smart_searchtool (SKILL.md, EXAMPLES.md). - Boundary markers: Absent. The skill lacks instructions to wrap retrieved memories in delimiters or warn the agent about potential instructions embedded in the memory content.
- Capability inventory: The skill uses
memory_smart_searchfor reading and troubleshooting steps suggestcurl/HTTP requests for REST fallbacks. - Sanitization: Absent. There is no instruction to sanitize or filter the retrieved memory content before presenting it to the user or processing it further.
Audit Metadata