coding-prefs
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- Indirect Prompt Injection Surface: The skill is designed to read and adhere to instructions stored in a persistent file (
/memory/coding-prefs.md), which could allow content from that file to influence the agent's future logic. - Ingestion points: The agent reads user-provided preferences from
/memory/coding-prefs.mdas specified inSKILL.md. - Capability inventory: The instructions in
SKILL.mddirect the agent to follow these rules when making coding and style decisions, effectively granting the file's content control over the agent's output. - Boundary markers: The skill does not implement specific delimiters or instructions to ignore potentially conflicting or malicious directives within the preferences file.
- Sanitization: There are no explicit mechanisms described for validating or sanitizing the content before it is stored or acted upon.
- Instruction Persistence: By design, the skill creates a mechanism where user feedback is converted into persistent rules. This allows instructions provided in one context to maintain influence across future sessions, which is a standard characteristic of personalization skills but warrants awareness regarding the types of instructions stored.
Audit Metadata