context-anchoring
Pass
Audited by Gen Agent Trust Hub on May 26, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Vulnerability to indirect prompt injection through context documents.
- Ingestion points: The skill reads Markdown files located in
.lattice/context/or a custom path defined in.lattice/config.yaml, as well as optionalrequirement_docfiles specified in the document frontmatter. - Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat the content of these files as untrusted or to ignore embedded instructions.
- Capability inventory: The skill allows the agent to read and write files within the repository. The instructions explicitly command the agent to "Honor all logged decisions" as "active commitments" and to treat "constraints as non-negotiable," which establishes a high-trust relationship with potentially untrusted data.
- Sanitization: No sanitization, validation, or filtering of the file content is performed before it is integrated into the agent's operating context.
Audit Metadata