The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references a package named 'cf' to be installed and executed via 'pnpm dlx' in both 'scripts/measure-row.sh' and 'references/measure-worker.md'. The package 'cf' on the public npm registry is a wrapper for the Cloud Foundry CLI, which is unrelated to the Cloudflare Workers observability tasks described. This use of 'pnpm dlx' to fetch an ambiguous package name for sensitive operations involving API tokens is a security risk.
[REMOTE_CODE_EXECUTION]: The provided bash script 'scripts/measure-row.sh' executes external code by downloading the 'cf' package at runtime. This allows for the execution of unverifiable code that is not pinned to a specific version or a known trusted source, potentially leading to the execution of malicious logic if the package name is typosquatted or hijacked.
[COMMAND_EXECUTION]: The skill includes a functional bash script 'scripts/measure-row.sh' that performs shell operations, including piping data between 'pnpm', 'jq', and 'grep'. While primarily intended for telemetry, this provides a mechanism for arbitrary command execution within the environment.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process project-specific files such as 'wrangler.jsonc', 'package.json', and backend source code to perform architectural analysis and edits. This represents a vulnerability surface where malicious content within a processed repository could influence the agent's actions.
Ingestion points: The agent is instructed to read project architecture, configuration files, and source code in the 'apps/' and 'packages/core/' directories.
Boundary markers: The skill uses '[detect: ...]' markers to guide the agent's context, but lacks explicit delimiters or instructions to treat project data as untrusted.
Capability inventory: The skill possesses the ability to execute provided bash scripts and modify project source code and database migrations.
Sanitization: No explicit sanitization, validation, or escaping of the ingested project data is performed before it is used to inform code generation or script execution.

cloudflare-backend