The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingest logs from GitHub Actions, which are external and could be manipulated by an attacker to include malicious instructions. These instructions could mislead the agent during its automated fix-and-push workflow.\n
Ingestion points: The run.mts script fetches CI failure logs and writes them to temporary files (e.g., in /tmp/greening-ci.../), which the agent is then directed to read for failure classification in SKILL.md.\n
Boundary markers: The skill does not employ specific delimiters or 'ignore embedded instructions' warnings when presenting log content to the agent.\n
Capability inventory: The skill provides the agent with extensive capabilities including Bash (git, gh, node, pnpm), Edit, and Write tools, which are used to modify code and push changes to the repository.\n
Sanitization: No sanitization or filtering is applied to the log data before it is processed by the agent.\n- [COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to execute commands for gh, git, node, pnpm, and other utilities. While these are necessary for the skill's purpose of managing CI and fixing code, and are restricted by prefixes in allowed-tools, they represent a high-privilege toolset that could be abused if the agent is influenced by malicious input.

greening-ci