AGENT SKILL

mcp-tooling

Use when choosing between MCP/tools in this project, especially Serena, ai-context PREVC, Context7 docs, Playwright browser automation, Neon database, bash, read/edit/search tools, or workflow setup.

License 许可
MIT
所属仓库
actus7/cifrashub
最近更新
2026-11-14
Stars
★ 3
Forks
2

01 前言 Front Matter

name
mcp-tooling
description
Use when choosing between MCP/tools in this project, especially Serena, ai-context PREVC, Context7 docs, Playwright browser automation, Neon database, bash, read/edit/search tools, or workflow setup.

03 SKILL.md 正文

MCP Tooling

Use this skill before non-trivial engineering work in cifrashub when tool choice matters.

Default tool routing

  • Use Serena for project-aware semantic code work: activate project, read memories, inspect symbols, find references, and make symbol/regex edits.
  • Use file tools (glob, grep, read, edit, write) for exact files, small targeted reads, and normal edits.
  • Use task with explore for broad codebase discovery when the target files are unknown.
  • Use ai-context for .context scaffolding, PREVC workflow, plans, agent orchestration, and semantic context maps.
  • Use Context7 for current third-party docs. Always resolve the library ID before querying docs unless the user provided /org/project.
  • Use Playwright only for browser/UI validation, screenshots, console/network inspection, and real page interactions.
  • Use Neon only for database/project/auth/data-api operations in Neon.
  • Use bash only for terminal commands: package scripts, git inspection, builds, lint, tests, and CLIs. Do not use bash for reading/editing/searching files.

Project startup

  • Serena project: D:\\projetos\\cifrashub / cifrashub.
  • After activating Serena, read mem:core first and follow only relevant memory references.
  • For this project, prefer Serena symbolic tools over full-file reads when editing TypeScript.

Safety gates

  • Never run destructive Neon actions or destructive SQL without explicit user approval.
  • Never commit, amend, push, create PRs, or force git operations unless explicitly requested.
  • Never expose secrets from config files in responses. If a secret is found, mention only that a secret exists and recommend moving it to an env var.
  • For opencode config changes, validate shape against https://opencode.ai/config.json when unsure and tell the user to restart opencode.

Verification

  • For code changes, use npm run lint and npm run build when feasible.
  • There is no npm run test script currently; ask for the test command before claiming tests ran.
  • For UI behavior, combine build/lint with Playwright checks when a browser-observable flow changed.

04 同仓库更多 Skills