grill-then-plan

Pass

Audited by Gen Agent Trust Hub on Aug 30, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill incorporates utility commands and local git queries within its reference files to automate documentation management tasks such as ADR numbering. These commands (e.g., git rev-parse, git ls-tree) are scoped to repository metadata and do not involve network exfiltration or sensitive file access.
  • [INDIRECT_PROMPT_INJECTION]: The design workflow involves reading codebase content, commit history, and existing documentation to drive design grilling. Evidence: 1. Ingestion points: SKILL.md Step 1 (codebase, commits, documentation). 2. Boundary markers: Absent. 3. Capability inventory: File writing (ADRs, specs) and local git metadata querying. 4. Sanitization: Verification of root cause (Step 1a) and glossary-based domain validation (Step 3).
  • [DYNAMIC_EXECUTION]: The skill provides pre-defined Bash and PowerShell scripts within ADR-FORMAT.md that the agent is instructed to execute locally to determine the next sequential ID for architecture decisions. This ensures global uniqueness of decision records across worktrees and branches.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 30, 2026, 02:26 AM
Security Audit — agent-trust-hub — grill-then-plan