The Agent Skills Directory

[SAFE]: The skill implements a legitimate workflow for documentation tasks and does not contain any malicious instructions, obfuscated code, or unauthorized network operations.
[DATA_EXFILTRATION]: The skill prompts users to provide organizational context and uses platform integrations (e.g., Google Drive, Slack) to fetch existing document data. These operations occur within the agent's context and do not attempt to exfiltrate data to external domains or expose sensitive system credentials.
[PROMPT_INJECTION]: The skill is designed to ingest and process external content, which inherently creates a surface for indirect prompt injection.
Ingestion points: External documents and team communication channels accessed via integrations or user input in SKILL.md (Stage 1 and 2).
Boundary markers: Not present; the skill does not explicitly define delimiters to separate ingested content from its own instructions.
Capability inventory: Modifies local files using create_file and str_replace, and utilizes sub-agents for quality checks (SKILL.md).
Sanitization: Not present; the skill does not specify any sanitization or validation of the processed external data.
Assessment: This attack surface is a standard characteristic of agents that process external data for documentation assistance and does not indicate a security issue within the skill itself.

doc-coauthoring