The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell scripts ./scripts/release.sh and ./scripts/release-rollback.sh. These scripts are responsible for critical tasks including version bumping, code committing, and managing pull requests via the GitHub CLI.
[REMOTE_CODE_EXECUTION]: The release script connects to remote gateway servers via SSH to execute immediate update commands (gateway-auto-update.sh --force), representing a remote code execution capability on production infrastructure.
[CREDENTIALS_UNSAFE]: The operation relies on pre-configured environment credentials, including SSH keys for gateway access, GitHub CLI (gh) tokens for PR and release management, and crates.io tokens for publishing packages via cargo.
[DATA_EXFILTRATION]: The skill communicates with external messaging platforms (Matrix and River) using tools like matrix-send and riverctl to broadcast release metadata and status to external channels.
[EXTERNAL_DOWNLOADS]: The skill interacts with external services such as GitHub (via git fetch and gh) and crates.io (via cargo search) to synchronize repository state and verify artifact availability.
[PROMPT_INJECTION]: The skill processes untrusted data from git history, creating an indirect prompt injection surface.
Ingestion points: Reads git log --oneline from the repository history in SKILL.md.
Boundary markers: None; commit messages are presented to the user without delimiters.
Capability inventory: Significant capabilities including script execution, SSH access, and package publication across multiple files.
Sanitization: No evidence of sanitization or escaping for commit message content before presentation.

release