AgentSkillsCN

test-runbook

验证 .plan/runbook.md 中的命令是否可执行——检查可执行文件、配置文件、环境变量，以及 TODO 占位符

SKILL.md
--- frontmatter
name: test-runbook
description: Validate that commands in .plan/runbook.md are executable — checks executables, config files, environment variables, and TODO placeholders
allowed-tools:
  - Read
  - Grep
  - Glob
  - Bash(which *)
  - Bash(command -v *)
  - Bash(*--version*)
  - Bash(test -f *)
  - Bash(test -d *)
  - Bash(printenv *)

Runbook Command Validator

Step 1: Read Runbook

Read .plan/runbook.md. Missing → FAIL, stop. Empty → FAIL, stop.

Step 2: TODO/TBD/FIXME Check

Case-insensitive scan. These block validate_planning_exit_gate(). Record line number, section, text. Flag as blocking at top of report.

Step 3: Extract Commands

Parse fenced code blocks. For each: note section (Build/Test/Deploy/Rollback), extract command lines (skip comments/blanks), split multi-line (&&/||/|) into individual executables, note $VAR/${VAR} references.

Step 4: Validate

4a. Executable: command -v <base-executable> (ignore sudo/env/nohup prefix). PASS/FAIL.

4b. Config files: Check implicit configs by tool: npm→package.json, make→Makefile, docker build→Dockerfile, docker-compose→compose.yml, cargo→Cargo.toml, go→go.mod, pip→requirements.txt, poetry→pyproject.toml. Also check explicit paths in commands.

4c. Env vars: printenv for each $VAR. PASS (set) / WARN (not set, may be CI-injected).

4d. Version (optional): <executable> --version. PASS/WARN.

Step 5: Report

code
Blocking: TODO/TBD/FIXME with line numbers
| Section | Command | Executable | Status | Notes |
Env vars: variable → set/not set
Fixes: concrete remediation per FAIL/WARN
Summary: "X commands, all passed" / "Z failures, Y warnings"

Safety

NEVER execute deploy/rollback/destructive commands. NEVER run rm/delete/destroy/drop/kill. NEVER pipe to sh/bash/eval. Only: which, command -v, --version, test -f, printenv.