The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands as fallbacks if primary MCP tools are unavailable, including the execution of local and remote scripts.
Evidence: node .gitnexus/run.cjs impact "X" --direction upstream --repo . and npx gitnexus.
[DYNAMIC_EXECUTION]: The skill recommends executing a JavaScript file found at a relative path within the workspace of the repository being analyzed.
Evidence: node .gitnexus/run.cjs
Analysis: This pattern allows for arbitrary code execution if a malicious repository contains a script at the specified path. The agent is instructed to run this script as a gate for repository edits.
[EXTERNAL_DOWNLOADS]: The skill suggests running a package from the npm registry using npx.
Evidence: npx gitnexus in the CLI fallback instructions.
Analysis: This involves fetching and running external code from a public registry at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the repository being analyzed to provide analysis results, which could potentially be manipulated.
Ingestion points: Code symbols and process metadata retrieved via the impact tool and gitnexus://repo/{name}/processes (SKILL.md).
Boundary markers: None mentioned; the instructions rely on the tool's output for risk assessment.
Capability inventory: The agent has shell command execution capabilities via node and npx.
Sanitization: No sanitization or validation of the repository-sourced metadata is described.
Analysis: Maliciously crafted code or metadata in the analyzed repository could potentially influence the agent's risk report or mislead it regarding the safety of changes.

gitnexus-impact-analysis