The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources that could contain embedded instructions intended to influence the agent's behavior.
Ingestion points: The agent reads third-party documentation via Context7 and repository content (source code, issues, and pull requests) from GitHub in SKILL.md.
Boundary markers: The skill instructs the agent to "Reconcile instead of blending" and explicitly compare public contracts with implementation evidence, which encourages critical evaluation of the ingested data.
Capability inventory: The agent's capabilities are limited to read-only tool access for GitHub and documentation queries for Context7 as defined in agents/openai.yaml.
Sanitization: The instructions do not specify automated sanitization or escaping methods for external content retrieved during the workflow.
[DATA_EXFILTRATION]: The skill includes explicit security guidelines in SKILL.md to prevent data leaks. It instructs the agent to "Never send credentials, private source, proprietary identifiers, or personal data in a Context7 query," which mitigates the risk of exfiltrating sensitive project information to external documentation services.
[EXTERNAL_DOWNLOADS]: The agent is configured in agents/openai.yaml to utilize external MCP tool services hosted at api.githubcopilot.com and mcp.context7.com. These connections are functional requirements for retrieving the documentation and source code evidence described in the skill's purpose.

github-context7-research