The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external data sources, which presents a surface for indirect prompt injection if the ingested data contains instructions designed to manipulate the agent.
Ingestion points: The skill reads production logs, traces, and browser data (console outputs, network requests, snapshots) via Cloudflare Observability and Playwright (documented in SKILL.md under '計測結果を読む' and 'ブラウザ').
Boundary markers: No specific delimiters or safety instructions are defined to separate ingested log/trace data from the agent's internal reasoning.
Capability inventory: The skill can execute shell commands via vp (e.g., vp exec alchemy logs, vp run dev), query Cloudflare APIs via MCP servers, and control browsers via Playwright.
Sanitization: There is no evidence of sanitization or filtering for the external content retrieved from logs or browser sessions before it is processed by the agent.
[COMMAND_EXECUTION]: The skill instructions direct the agent to execute various shell commands to facilitate debugging and observation.
Evidence: Use of vp exec alchemy logs --tail for real-time log monitoring and vp run dev for local development (found in '計測結果を読む').
Evidence: Use of MCP servers (cloudflare-observability and cloudflare-api) to programmatically interact with external services.
[EXTERNAL_DOWNLOADS]: The skill utilizes external tools and libraries to perform its tasks.
Evidence: Fetches and executes @playwright/cli using vp dlx (a wrapper for npx/dlx) to perform browser automation (found in 'ブラウザ'). Playwright is a well-known automation service.
[DYNAMIC_EXECUTION]: The skill uses dynamic tool execution at runtime.
Evidence: The use of vp dlx @playwright/cli involves downloading and executing the latest version of the Playwright CLI dynamically when browser observation is required.

measure-dont-guess