write-skill
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's primary functionality involves reading and editing other skill files, which creates an attack surface for indirect prompt injection.\n
- Ingestion points: The skill ingests untrusted data when reading existing skill files via the
Readtool or thegstack-skill-billutility.\n - Boundary markers: The
sections/preamble.mdfile establishes a 'User-origin gate' to differentiate instructions in processed data from trusted user input.\n - Capability inventory: The skill possesses file system access (
Read,Write,Edit,Bash) and executes framework-specific binaries in~/.claude/skills/gstack/bin/.\n - Sanitization: The framework mandates session ID validation for dynamic instruction blocks and keyword-origin verification for preference tuning to sanitize inputs from ingested files.\n- [SAFE]: The skill utilizes local binaries and resources provided by the vendor, garrytan, stored in the
~/.claude/skills/gstack/directory.\n- [SAFE]: The preamble includes explicit privacy controls, such as 'Ground Rules' that prohibit the agent from displaying environment variables, hardcoded secrets, or process tables.\n- [SAFE]: The skill implements a security gate for preference management, ensuring that configuration changes only occur when the user explicitly includes a keyword in their message, protecting against automated profile poisoning.
Audit Metadata