skills/ory/lumen/add-benchmark/Gen Agent Trust Hub

add-benchmark

Pass

Audited by Gen Agent Trust Hub on Sep 1, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub repositories (issues, PRs, and source code) to curate benchmark tasks. This creates a surface for indirect prompt injection if the repository content contains malicious instructions intended to influence the agent's behavior during extraction or validation.
  • Ingestion points: GitHub issue or PR URLs provided as arguments in SKILL.md.
  • Boundary markers: None observed in the instructions to separate untrusted repo content from agent instructions.
  • Capability inventory: The agent clones repositories, writes JSON and patch files to the local file system (bench-swe/tasks/, bench-swe/patches/), and determines/executes test commands.
  • Sanitization: The skill implements resource limits (size < 50MB, < 800 files, < 50 dependencies) but does not explicitly mention sanitization of natural language content for prompt injection patterns.
  • [COMMAND_EXECUTION]: The skill instructions state that the agent will "Determine the test command from repo conventions." This implies the execution of test commands (such as npm test or go test) found within untrusted, cloned repositories, which represents a potential command execution risk if the repository is malicious.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 1, 2026, 02:27 AM