agent-memory
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted project data to generate or update instruction files, establishing an indirect prompt injection surface.\n
- Ingestion points: The skill reads project configuration files (package.json, go.mod), README.md, git logs, and documentation files within the docs/devdocs/ directory (SKILL.md).\n
- Boundary markers: Managed sections are delimited by tags and YAML frontmatter blocks in AGENTS.md.\n
- Capability inventory: The skill utilizes Read, Write, Glob, Grep, Edit, Bash, and AskUserQuestion tools to process and update files.\n
- Sanitization: Content is filtered using standard quality rules in best-practices.md and a list of forbidden content types in devdocs-state-template.md, prohibiting the inclusion of commit hashes, LOC counts, and test results in memory files.\n- [COMMAND_EXECUTION]: The skill uses shell commands via the Bash tool to facilitate project analysis.\n
- Evidence: The skill executes wc -c for file size checks (SKILL.md) and shell pipelines involving grep, sed, sort, and tail to determine the maximum entity IDs from documentation files (devdocs-state-template.md). These commands are based on static internal templates and are used to provide analytical metadata rather than executing arbitrary external code.
Audit Metadata