The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external repositories, such as commit messages, descriptions, and file diffs, using commands like jj --no-pager log and jj --no-pager show. These ingestion points could potentially contain malicious instructions embedded by external contributors.
Ingestion points: Data enters the context through jj --no-pager log, jj --no-pager show <id>, and jj --no-pager diff --git (found in SKILL.md).
Boundary markers: The instructions do not specify explicit delimiters or warnings to ignore instructions embedded within commit messages or file content.
Capability inventory: The skill is permitted to execute jj commands via the Bash tool, which includes network operations like jj git clone, jj git fetch, and jj git push.
Sanitization: No specific sanitization or filtering of repository metadata or content is mentioned. However, this is a standard risk for any version control tool and the skill provides appropriate guidance for reviewing atomic changes, which facilitates safer operations.

jujutsu