code-review-graph

Fail

Audited by Gen Agent Trust Hub on Sep 12, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The troubleshooting documentation provides a command to install the 'uv' utility by piping a remote script directly to the shell.
  • Evidence: curl -LsSf https://astral.sh/uv/install.sh | sh in SKILL.md.
  • [EXTERNAL_DOWNLOADS]: The skill instructs the user to download and install several external components, including the 'code-review-graph' package and the 'uv' tool from astral.sh.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes the content of local codebases to build a knowledge graph, exposing the agent to instructions that might be embedded within source code.
  • Ingestion points: The skill parses source files in the project directory using Tree-sitter to build a persistent SQLite graph.
  • Boundary markers: There are no explicit markers or 'ignore' instructions provided to the agent to differentiate between code structure and embedded natural language instructions.
  • Capability inventory: The skill provides multiple tools for the agent to query the graph and retrieve structural summaries (e.g., get_review_context_tool, query_graph_tool).
  • Sanitization: No sanitization or filtering of code content is specified before it is processed by the graph builder or presented to the agent.
  • [METADATA_POISONING]: There is an inconsistency in the identity of the skill's author and repository; the description attributes the skill to 'ara.so', while the installation instructions reference a different repository path ('tirth8205/code-review-graph').
Recommendations
  • HIGH: Downloads and executes remote code from: https://astral.sh/uv/install.sh - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
Sep 12, 2026, 07:16 PM
Security Audit — agent-trust-hub — code-review-graph