The Agent Skills Directory

[DYNAMIC_EXECUTION]: The Preamble section in SKILL.md executes a shell script using a relative path traversal: bash \"$(dirname \"${BASH_SOURCE[0]}\")/../../check-update.sh\". This pattern attempts to run an external script located two directories above the skill's source, which is an unverifiable execution of environment-local code that could be hijacked if a malicious file exists at that path.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data from project-specific documents such as PRD and MVP files to generate product metrics.\n
Ingestion points: Step 2 explicitly reads content from docs/02-方案设计/PRD产品需求文档.md and docs/01-需求调研/MVP方案.md.\n
Boundary markers: The instructions lack any explicit delimiters or 'ignore instructions' warnings to prevent the agent from following malicious commands hidden within the processed documentation.\n
Capability inventory: The skill utilizes high-impact tools including Bash and Write, which significantly increases the risk if an injection occurs.\n
Sanitization: No validation or sanitization is performed on the data read from the external documents before it is processed by the agent.

pm-data