The Agent Skills Directory

COMMAND_EXECUTION (MEDIUM): The skill utilizes codex exec --full-auto to generate and apply code fixes based on an automated audit. This involves dynamic code generation and execution which could be manipulated if the remediation instructions are influenced by malicious data found in the audited project.
DATA_EXFILTRATION (LOW): The troubleshooting guide in references/troubleshooting.md suggests using webhook.site as a temporary api_host. This practice sends application event data to an external, unverified third-party domain.
PROMPT_INJECTION (LOW): The skill is susceptible to indirect prompt injection because it ingests untrusted local source code and use the results to drive the automated remediation logic.
Ingestion points: scripts/detect-environment.sh reads local files; mcp__posthog__logs-query processes log data.
Boundary markers: None identified in the prompt interpolation logic.
Capability inventory: codex exec (arbitrary code execution), vercel env add (production environment modification).
Sanitization: No explicit sanitization or validation of the findings before they are passed to the remediation tool.
EXTERNAL_DOWNLOADS (LOW): The skill invokes npx convex, which can lead to the download and execution of packages from the npm registry at runtime.
DATA_EXFILTRATION (LOW): The scripts/detect-environment.sh script accesses .env.local and queries production environment variables via the Vercel CLI to verify project configuration, exposing sensitive configuration metadata to the agent context.

posthog