The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill exhibits an indirect prompt injection surface because it reads progress summaries from local files like claude-progress.txt to populate handoff packages. 1. Ingestion points: scripts/state_serializer.py and scripts/context_bridge.py ingest content from text-based status files. 2. Boundary markers: Absent; the protocol transfers raw extracted text. 3. Capability inventory: Local file writing and subprocess execution for Git metadata. 4. Sanitization: Minimal; uses regex to parse specific markdown-like sections. This finding is downgraded to SAFE as the behavior is central to the skill's core functionality of context bridging.
[Command Execution] (SAFE): The skill uses subprocess.run to call Git for metadata collection. Arguments are provided as a list with no shell interpolation, preventing command injection vulnerabilities.

handoff-coordinator