The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided thesis content in Markdown format, creating a surface for indirect instructions to influence agent behavior.
Ingestion points: User-provided thesis files or text snippets are read into the agent's context using the Read tool in Step 1 of the workflow.
Boundary markers: The instructions do not define specific delimiters or "ignore embedded instructions" warnings for the ingested thesis text.
Capability inventory: The skill utilizes Read and Write tools to perform file system operations on local thesis documents.
Sanitization: No explicit sanitization, filtering, or validation of the external thesis content is specified before it is processed for diagnosis and polishing.
[COMMAND_EXECUTION]: The workflow involves automated file system interaction via tools to read draft documents and write polished output, following a specific filename template (品牌方-mba-thesis-polished-{YYYY-MM-DD}-v0.1.md).
[PERSISTENCE_MECHANISMS]: The "General Feedback Loop" section instructs the agent to analyze user feedback to determine if changes should be applied to the "canonical Skill" or "shared specifications." This mechanism allows the agent to update its own rules and cross-skill configurations based on interaction history, representing a managed form of behavioral persistence.

lov-thesis-polish