The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the command bun run format as part of its finalization step. This is a standard development workflow used to maintain file consistency and is considered a legitimate operation within the scope of documentation management.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection through its handling of external release data.
Ingestion points: The BODY input variable in SKILL.md receives raw markdown release notes from potentially untrusted contributors.
Boundary markers: There are no explicit delimiters or system instructions provided to the agent to treat the BODY content strictly as data rather than instructions.
Capability inventory: The agent has the ability to overwrite documentation files in the docs/changelogs/ directory and execute shell commands via bun.
Sanitization: The skill performs structural modifications (reformatting links, deleting sections) but does not include sanitization to filter out or neutralize embedded natural language instructions.

docs-changelog