pm-interview
Pass
Audited by Gen Agent Trust Hub on Aug 3, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: Local Maintenance Script Dependency. The preamble section attempts to execute a shell script named
check-update.shlocated two directories above the skill's location (../../check-update.sh). This demonstrates a dependency on the local environment's file structure and the execution of external scripts not contained within the skill itself.- [PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill ingests data from local markdown files to guide its planning process. Findings include: - Ingestion points: It checks for and reads
docs/01-需求调研/需求调研报告.mdto extract user personas and hypotheses. - Boundary markers: The skill lacks delimiters or instructions to the agent to disregard embedded instructions within the ingested report file.
- Capability inventory: The skill possesses significant capabilities including
Bashfor command execution,Writefor file modification, and theTasktool for spawning sub-agents. - Sanitization: There is no evidence of sanitization or validation performed on the content retrieved from the local report file before it is used to influence agent logic.
Audit Metadata