The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to read external data from tools such as ci-log and ci-status to monitor the success of CI jobs. This data originates from external build environments and may contain untrusted content that could influence the agent's behavior.
Ingestion points: The agent processes output from the ci-log and ci-status tools.
Boundary markers: The instructions do not define explicit delimiters or warnings to separate external log content from the agent's internal instruction set.
Capability inventory: The skill possesses the ability to execute git commands (git_finalize, git_push) and write data to the local file system (.gm/receipts/).
Sanitization: Although the skill computes a SHA-256 hash of the log content using crypto.subtle.digest for integrity recording, there is no explicit instruction to sanitize or escape the content before it is read by the agent.
[COMMAND_EXECUTION]: The skill instructions direct the agent to utilize several command-line tools to manage repository state and CI feedback.
Evidence: The workflow involves git_status, git_finalize, git_push, git_show, ci-status, and ci-log. These tools allow the agent to read from and write to the repository and external CI interfaces.
Autonomy: The 'Finish the work' section encourages the agent to autonomously 'repair the fault where it originates' if blocked, which could lead the agent to perform unexpected system modifications depending on the available toolset.

gm-publish