The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes Git and GitHub CLI commands through the Bash tool. It follows security best practices by using absolute paths (e.g., /usr/bin/env, /bin/bash, /usr/bin/printf) and unsetting critical environment variables (BASH_ENV, ENV, GIT_EXEC_PATH, GIT_CONFIG_*) to protect against command hijacking and environment-based attacks.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the local environment, such as Git branch names and remote URLs. This surface is mitigated by a multi-phase validation process where the agent generates a 'Push Plan' and requires explicit user confirmation via AskUserQuestion. The plan is cryptographically bound to the final execution phase using SHA-256 digests of the repository state.
[DATA_EXPOSURE]: The skill contains robust logic to identify and redact sensitive credentials (tokens, passwords) embedded in Git remote URLs. This ensures that secrets are not exposed in terminal output, logs, or user prompts during the approval process.
[DYNAMIC_EXECUTION]: The skill uses complex, multi-line shell scripts embedded in the markdown to perform its validation logic. While these scripts involve advanced shell features like parameter expansion and subshells, they are static parts of the skill's instructions and do not facilitate the execution of arbitrary, untrusted remote code.

push-ci