agent-memory
Pass
Audited by Gen Agent Trust Hub on Oct 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from local project documentation to build and maintain its memory structure, which presents a surface for indirect prompt injection if those files are attacker-controlled.
- Ingestion points: The
Operation: Build(found inSKILL.md) scans project files including*.md,package.json,pyproject.toml, and other configuration files. - Boundary markers: The skill employs YAML frontmatter blocks and specific directory structures within
.agent-memory/to encapsulate data. - Capability inventory: The skill performs extensive file-system operations, including creating directories, writing markdown files, and modifying instructions in
AGENTS.md. - Sanitization: The instructions mandate that the agent "Distill, don't transcribe" and "Summarize don't copy," utilizing the LLM's summarization logic to filter raw input content.
Audit Metadata