dev1-start

Pass

Audited by Gen Agent Trust Hub on Aug 14, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple sources including the $ARGUMENTS variable (user-provided feature description) and project-level files such as CLAUDE.md or .claude/CLAUDE.md. These files are explicitly read to gather "project-specific instructions and context," which could be used to influence or override agent behavior if they contain malicious instructions.
  • Ingestion points: SKILL.md (via $ARGUMENTS), project root (via CLAUDE.md, package.json, Cargo.toml, go.mod).
  • Boundary markers: None present; external content is written directly into .llm/<feature_name>/tmp_context.md and processed as context.
  • Capability inventory: Directory creation, file read, and file write operations within the local workspace.
  • Sanitization: No sanitization or validation of the input content is performed before writing to the context files.
  • [DATA_EXPOSURE]: The skill performs high-level exploration of the project codebase, which includes reading dependency manifests (package.json, Cargo.toml, go.mod) and mapping the directory structure. While this is the intended purpose of the tool, it involves systematic access to project metadata.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 14, 2026, 12:37 AM
Security Audit — agent-trust-hub — dev1-start