context-engineering
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is a set of best-practice instructions for 'Context Engineering.' It provides structural advice for organizing agent memory and project rules without including executable scripts or dangerous directives.
- [INDIRECT_PROMPT_INJECTION]: The skill explicitly recognizes the threat of indirect prompt injection. It provides a trust-level framework that classifies user-submitted content, third-party API responses, and external documentation as 'untrusted.' It further instructs agents to treat instruction-like content in these sources as data to be surfaced to the user, rather than directives to be followed.
- Ingestion points: The skill recommends reading project source files, specs, and external documentation (SKILL.md).
- Boundary markers: Recommends using delimiters or structured blocks (e.g., 'PROJECT CONTEXT') to separate information.
- Capability inventory: No direct tool or subprocess capabilities are defined in the skill's own manifest.
- Sanitization: Recommends manual verification of external content and treats instruction-like patterns in data as non-executable.
- [CREDENTIALS_SAFE]: The documentation includes a template for project rules (CLAUDE.md) that explicitly advises users 'Never commit .env files or secrets,' which is a standard security best practice.
Audit Metadata