The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection surface. The skill instructions direct the agent to load and parse an external project.md file to establish a session context, which can influence the behavior of subsequent operations.
Ingestion points: The agent reads project-specific metadata from ~/.config/finance-skills/projects/{project-code}/project.md (referenced in SKILL.md).
Boundary markers: There are no instructions defining delimiters or explicit 'ignore embedded instructions' warnings for the data read from the project files.
Capability inventory: The resulting current_project context variable controls the output paths and operational scope for several other tools and skills, including /internal-audit:control-testing and /month-end-close:* (as listed in the '协作' section).
Sanitization: While the skill mentions YAML frontmatter parsing, it does not specify any validation or sanitization for the values retrieved from the external file before they are injected into the session context.

project-switch