The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from requirement files (digest, audit, and pm-checklist) to generate design specifications. This creates an attack surface where instructions embedded in these requirements could influence the agent's output during the technical design phase.
Ingestion points: External files located in sdlc/<demand_name>/intake/ (specifically digest-*.md, audit-*.md, and pm-checklist-*.md).
Boundary markers: The skill does not explicitly define delimiters or provide instructions to the agent to disregard potentially malicious commands embedded in the ingested content.
Capability inventory: The skill possesses file-writing capabilities (design.md) and utilizes codegraph and mysql MCP tools to explore the local environment.
Sanitization: There is no evidence of sanitization, filtering, or validation of the content read from the external intake files.
[COMMAND_EXECUTION]: The skill's handoff documentation (references/trellis-handoff.md) instructs the user or system to execute a local Python script for development gatekeeping.
Evidence: The documentation explicitly describes running python3 <sdlc-gate 安装目录>/scripts/guard_dev.py <项目根> <需求名> to verify development readiness. This involves shell execution of a script with variable arguments based on the project context.

sdlc-design