The Agent Skills Directory

[SAFE]: The skill follows standard agentic workflow patterns for software engineering tasks. It emphasizes local code analysis and evidence-based planning before implementation.
[SAFE]: The skill processes project-specific data which presents a surface for indirect prompt injection, but it implements basic identification of injected content.
Ingestion points: Reads task instructions from plan.md, docs/project-reference/domain-entities-reference.md, and .ai/workspace/analysis/ files.
Boundary markers: Includes specific logic to check for [Injected: ...] headers in project documentation files to identify auto-generated content.
Capability inventory: Orchestrates sub-agents for coding (fullstack-developer), testing (tester), and version control (git-manager). It also executes a local utility script at .claude/scripts/code_graph.
Sanitization: Relies on the host agent's internal safety filters and the specific 'Understand Code First' hard-gates to validate implementation steps against the existing codebase.

code-parallel