The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's installation instructions utilize npx and git clone to retrieve content from a GitHub repository owned by an untrusted user (ningzimu/codex-ppt-skill). Since this source is not part of a verified or trusted organization, it introduces a potential supply chain vulnerability.
[COMMAND_EXECUTION]: The skill operates by executing local Python scripts, specifically codex_ppt_runtime.py and assemble_ppt.py. These scripts are responsible for handling API configuration, generating images via network requests, and performing file system operations to assemble the final presentation.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, untrusted content such as articles, papers, and reports. This text is then used to generate prompts for AI-driven image creation, presenting an attack surface where malicious instructions embedded in the source documents could influence the agent's actions.
Ingestion points: Reads text from articles, papers, and reports as specified in SKILL.md.
Boundary markers: Absent. The skill does not define clear delimiters or instruct the agent to disregard instructions found within the processed documents.
Capability inventory: Network access for API-based image generation; file system writes for project directories, images, markdown files, and .pptx files.
Sanitization: Absent. There is no indication that the ingested text is sanitized or filtered before being included in prompts.
[METADATA_POISONING]: There is a notable inconsistency between the author mentioned in the skill body ('ara.so'), the GitHub repository owner ('ningzimu'), and the provided author context ('reason-machines'). This discrepancy can be deceptive regarding the true origin and security of the skill.

codex-ppt-skill