The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes story content within sub-agent prompts, creating a surface for indirect prompt injection.
Ingestion points: Story text is interpolated into the {待审查内容} placeholder in SKILL.md (Phase 2).
Boundary markers: Lacks strong delimiters or specific instructions to ignore embedded commands within the analyzed story text.
Capability inventory: The main agent can read local files and spawn sub-agents; the consistency-checker and story-researcher can perform searches.
Sanitization: No filtering or validation is applied to the analyzed text before it is sent to sub-agents.
[COMMAND_EXECUTION]: The skill uses local repository tools to facilitate the review process.
Evidence: Employs git diff to determine the review scope and grep for scanning project files during consistency checks.

story-review