The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the ExecutionProxy interface to perform various shell commands, including pnpm for building and testing, npm for registry lookups and publishing, and git for tag management and pushing updates. It also executes local management scripts such as ./scripts/release.sh and ./scripts/docker-onboard-smoke.sh.
[INDIRECT_PROMPT_INJECTION]: The skill reads from commit logs and writes to changelog files (releases/vYYYY.MDD.P.md), which are later used to create GitHub Releases. This represents an ingestion point where untrusted data (e.g., malicious commit messages) could influence the agent's context or downstream users.
Ingestion points: Commit logs and generated changelog markdown files in the releases/ directory.
Boundary markers: The skill explicitly requires a mandatory human review of the drafted changelog notes before the final publish step.
Capability inventory: The agent has capabilities for terminal command execution, file system read/write, and network access via package managers and version control tools.
Sanitization: There is no automated sanitization mentioned for the content of the changelogs; the security relies on the mandated manual review checkpoint.

release