astrofy-cloudflare
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests various local project files, such as package.json, lockfiles, and wrangler.toml, to determine the project's build path and configuration. This data ingestion represents an indirect prompt injection surface where a project with malicious configuration could attempt to influence agent behavior. The risk is mitigated by explicit instructions to perform dry-runs and verify the Build Output configuration before final execution.
- Ingestion points: Reads project manifests (package.json, lockfile), Cloudflare configurations (wrangler.*, cloudflare.config.ts), and local configuration paths (.astrofy/config/paths.json).
- Boundary markers: No specific delimiters are used for untrusted content, but the workflow includes mandatory verification steps and classification rules to guide the agent.
- Capability inventory: Executes shell commands including 'astro', 'wrangler', and the 'cf' CLI tool.
- Sanitization: Relies on the '--dry-run' flag of the 'cf' CLI and manual inspection of the '.cloudflare/output/v0/config.json' file to ensure correct behavior.
Audit Metadata