The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary workflow involves reading untrusted content from a file named sucai.md to extract information and opinions for article generation. This presents a surface where adversarial text could attempt to manipulate the agent's subsequent logic.
Ingestion points: Untrusted content is loaded from sucai.md (and inbox.md) during the initial '选题质检' (Topic Quality Check) phase.
Boundary markers: The instructions do not specify the use of XML tags, delimiters, or clear warnings to the agent to ignore any instructions embedded within the source material.
Capability inventory: The skill instructions authorize the agent to execute shell commands (date), run local Python validation scripts (scripts/check_prose.py, scripts/check_body_length.py), and perform recursive file reading within local project directories (/Users/itwanger/Documents/GitHub/).
Sanitization: There is no defined process for sanitizing or filtering the source content before processing.
[COMMAND_EXECUTION]: The skill uses local system commands and distributed Python scripts to maintain formatting and quality standards.
Evidence: SKILL.md contains explicit instructions to run date for context and to invoke the scripts/check_prose.py and scripts/check_body_length.py tools during the self-check phase (Step 7). This is a functional requirement of the skill.

ai-article