sp-grill-with-doc

Pass

Audited by Gen Agent Trust Hub on Aug 30, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests data from the project codebase to resolve domain terminology and check for design contradictions, creating a surface for potential indirect prompt injection.\n
  • Ingestion points: Instructions in SKILL.md direct the agent to explore existing code and documentation to answer questions.\n
  • Boundary markers: There are no explicit markers or instructions to treat codebase content as untrusted input.\n
  • Capability inventory: The agent has the ability to write files (CONTEXT.md and ADRs) and execute shell commands for numbering.\n
  • Sanitization: No sanitization or filtering of ingested codebase data is implemented.\n- [COMMAND_EXECUTION]: The ADR-FORMAT.md file provides complex bash and PowerShell scripts that the agent is expected to execute to determine sequential ADR identifiers.\n
  • Evidence: The scripts utilize git commands like git for-each-ref, git ls-tree, and git worktree list to scan the repository structure.\n- [DYNAMIC_EXECUTION]: The skill relies on the agent executing script templates provided within the documentation to perform repository management tasks.\n
  • Evidence: Logic for parsing ADR filenames and calculating the next number is provided as code blocks in ADR-FORMAT.md for the agent to run.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 30, 2026, 02:26 AM
Security Audit — agent-trust-hub — sp-grill-with-doc