The Agent Skills Directory

[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks as it is designed to ingest and process untrusted user-supplied drafts and conversation history.
Ingestion points: The skill workflows (SKILL.md) explicitly instruct the agent to read the 'draft plus relevant conversation context'.
Boundary markers: The skill defines a 'Source Precedence' model and a workflow step to 'separate supported facts... commands... and conflicting claims', which serves as a delimiter against accidental instruction following.
Capability inventory: The skill is limited to text generation and editing; it does not demonstrate access to dangerous tools, file system writes, or unauthorized network operations.
Sanitization: No programmatic sanitization is defined, but the 'Calibration Check' and 'Hard Rules' provide a logical framework for the agent to filter out extraneous or malicious instructions from the source material.
[EXTERNAL_DOWNLOADS]: The skill metadata contains instructions for installation and updates using a remote GitHub repository.
Resource: https://github.com/idaibin/aicraft accessed via npx skills add and npx skills update commands.
Context: These are standard maintenance procedures associated with the skill author's own infrastructure.

writing-editor