pave-contribute

Pass

Audited by Gen Agent Trust Hub on Jun 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes the "Bash" tool to perform git operations, including cloning the "tonone-ai/tonone" repository, branching, committing, and pushing code. It also uses the GitHub CLI ("gh") to fork repositories and create Pull Requests. These actions are performed to facilitate upstream contributions to the vendor's own repository.
  • [DATA_EXFILTRATION]: Extracted insights from the user's conversation are transmitted to a remote GitHub repository. To mitigate the risk of unintended data exposure, the skill implements a sanitization step to remove credentials and specific identifiers, and it mandates user approval via the "AskUserQuestion" tool before the Pull Request is created and data is sent.
  • [PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks as it processes untrusted conversation data to generate contribution content.
  • Ingestion points: The agent reads and analyzes the full session history to identify valuable patterns or gaps (SKILL.md, Step 1).
  • Boundary markers: No specific delimiters or safety instructions are used to separate conversation content from the agent's internal logic during the extraction process.
  • Capability inventory: The skill possesses the capability to modify repository files and interact with remote servers using "Bash", "Write", and "Edit" tools (SKILL.md, Step 5).
  • Sanitization: A sanitization phase (Step 3) is included to automatically identify and redact sensitive information such as project names, paths, and credentials before any external transmission occurs.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 17, 2026, 02:58 AM
Security Audit — agent-trust-hub — pave-contribute