The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external story, task, and epic markdown files to automate branch naming and determine branching models, creating an indirect prompt injection surface.
Ingestion points: Metadata and frontmatter are read from user-provided file paths such as story.*.md, epic.*.md, and task.*.md (via the doc-id.js utility and frontmatterField logic).
Boundary markers: The skill relies on standard YAML frontmatter blocks but lacks explicit instructions for the agent to ignore or isolate potential Natural Language instructions that might be embedded within those data fields.
Capability inventory: The skill possesses the capability to execute Git operations (git checkout, git branch, git push), perform file system reads/writes, and update session state files in the .claude/state/ directory.
Sanitization: The references/doc-id.js utility provides robust sanitization by enforcing strict regex patterns on IDs and slugs, which mitigates the risk of malicious payloads in branch names or identifiers.

create-branch