The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes processes for an agent to ingest and analyze data from external Git repositories via the Cloudflare Artifacts service.
Ingestion points: Methods like repo.readFile, repo.readTree, and repo.readBlob in references/workers-binding.md and references/patterns.md allow the agent to read repository content into its context.
Boundary markers: The skill instructions advocate for a "repo-per-agent" pattern in SKILL.md and references/patterns.md to establish clear data boundaries between agents and sessions.
Capability inventory: The skill documents capabilities for reading/writing repository files, creating commits, and minting scoped access tokens.
Sanitization: Content retrieved from repositories is treated as data for the agent's work; the skill relies on repository-level isolation rather than content sanitization as the primary security control.
[COMMAND_EXECUTION]: The skill provides patterns for interacting with the git CLI tool to perform repository management tasks.
Evidence: Shell scripts in SKILL.md and a Go-based wrapper in references/git-and-tokens.md using exec.CommandContext demonstrate standard ways to execute git operations.
These examples follow security best practices by passing credentials through environment variables rather than command-line arguments.
[EXTERNAL_DOWNLOADS]: The skill references official Cloudflare API endpoints and development tools required for the service's operation.
References include api.cloudflare.com for the REST API, *.artifacts.cloudflare.net for the Git server, and the wrangler CLI.
All external references target a trusted vendor (Cloudflare) and are necessary for the skill's documented functionality.
[SAFE]: No obfuscation, credential exfiltration, or prompt injection patterns were detected. The skill provides clear guidance on secure authentication and data handling practices within the Cloudflare ecosystem.

cloudflare-artifacts