The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements a feedback loop that processes external user input to drive automated design and build tasks, creating an indirect prompt injection surface.
Ingestion points: External user feedback is ingested in Phase 2 (Step 6) and Phase 3 (Step 12) through the feedback-analyst agent.
Boundary markers: The instructions do not specify explicit delimiters or boundary markers to isolate user feedback from the agent's instructions, nor do they instruct the agent to ignore potential commands embedded in the feedback.
Capability inventory: The skill has capabilities to write files to the local workspace (_workspace/), execute Python scripts (compose.py), and perform network operations via git push to a specific repository.
Sanitization: User feedback is transformed into "change specifications" by the feedback-analyst agent, providing a layer of interpretation, but no explicit sanitization for adversarial text is defined.
[DATA_EXFILTRATION]: The skill contains logic for pushing build artifacts to a remote location.
Evidence: Phase 4 (Step 17) describes a procedure to push font releases to the repository github.com/revfactory/font-harness.
Context: The target repository belongs to the skill's author (revfactory). This is a legitimate vendor resource used for its intended purpose of hosting the font project's output.

hangul-font-studio