AgentSkillsCN

pal-mcporter

通过 `mcporter` CLI，面向 PAL 的标准输入输出 MCP 服务器（`uvx … pal-mcp-server`）运行 PAL 的预提交、代码审查、安全审计、调试、API 查找、模型列表等工具，而无需在 `~/.codex/config.toml` 中将 PAL 配置为 Codex MCP 服务器。当您希望以更贴近 Shell 或脚本的方式运行 PAL 工具、避免 MCP 工具模式带来的上下文开销，或需要在未接入 Codex MCP 网络的环境中使用 PAL 时，这一技能尤为适用。

SKILL.md
--- frontmatter
name: pal-mcporter
description: Run PAL MCP tools (precommit, codereview, secaudit, debug, apilookup, listmodels, etc.) via the `mcporter` CLI against the PAL stdio MCP server (`uvx … pal-mcp-server`) instead of configuring PAL as a Codex MCP server in `~/.codex/config.toml`. Use when you want shell/script-friendly PAL runs, want to avoid MCP tool-schema context overhead, or need PAL access in an environment without Codex MCP wiring.

PAL via MCPorter

Quick start

  • •See which tools are available (from the bundled generated CLI help):
    • •bash "<path-to-skill>/scripts/pal"
  • •Print copy/pastable examples for every tool currently exposed by your PAL server:
    • •bash "<path-to-skill>/scripts/pal" examples
  • •Run a fast self-test (validates bundled CLI + PAL server startup; no live model call):
    • •bash "<path-to-skill>/scripts/pal" selftest
  • •Run self-test including live model calls (costs tokens; requires an API key):
    • •bash "<path-to-skill>/scripts/pal" selftest --live
  • •Call tools:
    • •bash "<path-to-skill>/scripts/pal" -o json version
    • •bash "<path-to-skill>/scripts/pal" -o json listmodels
    • •bash "<path-to-skill>/scripts/pal" -o markdown apilookup --prompt "OpenAI Responses API streaming"
    • •bash "<path-to-skill>/scripts/pal" -o markdown chat --prompt "ping" --working-directory-absolute-path "$PWD" --model auto

Workflow

  1. •Read the embedded tool list.
    • •bash "<path-to-skill>/scripts/pal"
  2. •Use examples to find minimal calls / argument shapes (recommended).
    • •bash "<path-to-skill>/scripts/pal" examples
  3. •Call the tool via the bundled CLI.
    • •Prefer --raw '{...}' for complex payloads and optional parameters.
    • •Global flags (-o/--output, -t/--timeout) must come before the tool name.
  4. •Treat stdout as the tool result.
    • •Use -o markdown for readability or -o json when you want structured output.

Does this match pal-mcp-server?

Mostly. This skill runs PAL via uvx … pal-mcp-server and bundles a CLI generated from PAL’s tool schemas via mcporter generate-cli. If upstream adds/removes tools, the bundled CLI may need to be regenerated.

  • •Check which PAL version you’re running: bash "<path-to-skill>/scripts/pal" -o json version
  • •Source-of-truth tool surface (live): bash "<path-to-skill>/scripts/pal" examples

Notes (from upstream + schemas)

  • •Tool surface can be reduced with DISABLED_TOOLS (helpful for clients that ingest tool schemas). Upstream often recommends: analyze,refactor,testgen,secaudit,docgen,tracer to save context.
  • •Upstream PAL supports multi-turn threads via continuation_id, but this skill runs PAL in ad-hoc STDIO mode (fresh server per invocation), so continuations generally won’t resume across separate pal commands.
  • •Prefer passing files via absolute_file_paths (and images via images) rather than pasting large blobs into prompts.

Selftest knobs

  • •PAL_SELFTEST_CLI_NAME: which CLI clink should call in selftest --live (auto-detects codex, claude, gemini if present in PATH)
  • •PAL_SELFTEST_MODEL: optionally override the model used during selftest --live (when supported by the tool)
  • •PAL_SELFTEST_THINKING_MODE: defaults to minimal

Configuration

This skill uses ad-hoc stdio mode (no mcporter.json required) and ships a bundled CLI at scripts/pal-cli.js.

Set these env vars when needed:

  • •OPENAI_API_KEY (required for OpenAI-backed PAL usage; alternatives include GEMINI_API_KEY, OPENROUTER_API_KEY, etc.)
  • •PAL_DEFAULT_MODEL (default: auto; forwarded as DEFAULT_MODEL when DEFAULT_MODEL is unset)
  • •PAL_MCP_FROM (default: git+https://github.com/BeehiveInnovations/pal-mcp-server.git)
  • •MCPORTER_VERSION (default pinned in "<path-to-skill>/scripts/pal")
  • •PAL_MCPORTER_TIMEOUT_MS (default: 120000)
  • •DISABLED_TOOLS (PAL setting; comma-separated tool names to hide/disable in the server; unset/empty enables all)

Notes

  • •bash "<path-to-skill>/scripts/pal" is the quick “what’s available?” view; use examples when you need parameter shapes.
  • •STDIO mode inherits your shell environment automatically; avoid passing secrets via --env KEY=value unless you have to.
  • •If PAL startup is slow (first uvx run), increase timeouts with PAL_MCPORTER_TIMEOUT_MS or -t <ms>.