relay-merge

Warn

Audited by Gen Agent Trust Hub on Oct 9, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The script scripts/finalize-run.js dynamically generates a Node.js script as a string and executes it via node -e (passed as an argument in a command configuration). This behavior is implemented in the mergeObserver function to verify PR identity.\n- [COMMAND_EXECUTION]: The skill invokes external CLI tools such as git and gh, along with Node.js scripts, using execFileSync and spawn. It allows binary paths to be overridden through environment variables like RELAY_GIT_BIN, RELAY_GH_BIN, and RELAY_SPRINT_STATE_BIN, which could lead to the execution of unintended binaries if these variables are manipulated.\n- [DATA_EXFILTRATION]: The skill accesses GitHub authentication tokens from environment variables (GH_TOKEN, GITHUB_TOKEN) to interact with the GitHub API. While intended for legitimate PR operations, this represents a capability for network exfiltration of local repository state. It also reads from and writes to user-local configuration and run data stored in the ~/.relay directory.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources, including GitHub PR responses, git logs, and local project markdown files.\n
  • Ingestion points: scripts/finalize-run.js (GitHub PR API data), scripts/sprint-owner.js (sprint frontmatter and issue body metadata), scripts/append-learnings.js (reading spec/capabilities.md).\n
  • Boundary markers: scripts/append-learnings.js utilizes HTML comment markers (<!-- LEARN:BEGIN -->) to delineate the areas it is authorized to modify.\n
  • Capability inventory: The skill has the ability to execute commands via execFileSync, write to files on the local file system, and perform authenticated network operations through the gh CLI.\n
  • Sanitization: The skill performs input validation using regular expressions to check for valid SHAs, PR numbers, and safe tokens.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 9, 2026, 12:21 AM