The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted user-supplied repository files, Wrangler configurations, and Infrastructure-as-Code (Terraform) to perform audits. This architecture is susceptible to indirect prompt injection, where malicious instructions embedded in the audited files could attempt to influence the agent's behavior during the audit process.
Ingestion points: The skill inventories and reads source code, config files, and environment exports from the user's workspace.
Boundary markers: The instructions mandate an "Evidence first" approach and require grounding all findings in official documentation to mitigate hallucinations or adversarial influence.
Capability inventory: The agent is capable of running shell commands (Python scripts, Wrangler, npx) and fetching external URLs.
Sanitization: The provided cfdoctor_static_scan.py script includes logic to redact discovered secret values before they are presented in the analysis.
[COMMAND_EXECUTION]: The skill's standard workflow involves executing a local Python script (cfdoctor_static_scan.py) for static analysis. It also suggests running third-party utility tools, such as the @acoyfellow/deadlint package via npx, to perform specific reachability checks.
[EXTERNAL_DOWNLOADS]: The skill routinely fetches live documentation and index files from Cloudflare's official developer portal (developers.cloudflare.com) using curl. This is used to ensure that audits are based on the most current platform limits, pricing, and best practices.

cloudflare-doctor