{"slug": "readable-ad-hoc-commands-agent-skill", "title": "Readable ad-hoc commands agent skill", "summary": "A new open-source project, readable-ad-hoc-commands, formats AI coding agents' shell commands into readable scripts with named values, comments, phase banners, and plain-English summaries, aiming to reduce blind approvals of destructive commands. The project, hosted on GitHub under a-b/readable-ad-hoc-commands, works with any agent and command language, and includes rules in AGENTS.md and a Claude Code skill adapter.", "body_md": "**Stop approving shell commands you can't read.** When an AI coding agent fires\na dense one-liner at your terminal, you either reverse-engineer it or rubber-stamp\nit blind. This makes the agent format those commands so you can see their intent,\nsteps, and **blast radius** *before* they run — and preview anything destructive\nfirst.\n\nWorks with any agent, in any command language (shell, `python -c`\n\n, `node -e`\n\n,\n`jq`\n\n, SQL).\n\n**Before** — what an agent normally emits:\n\n```\nfind . -name \"*.log\" -mtime +7 -exec gzip {} \\; && find . -name \"*.gz\" -mtime +30 -delete && du -sh .\n```\n\n**After** — what this produces:\n\n``` bash\n#!/usr/bin/env bash\nset -euo pipefail\n\n# Housekeeping for log files in the current tree.\nroot=\".\"\ncompress_after_days=7      # gzip logs older than this\ndelete_after_days=30       # remove archives older than this\n\necho \"==> Compressing .log files older than ${compress_after_days}d\"\nfind \"$root\" -name \"*.log\" -mtime \"+${compress_after_days}\" -exec gzip {} \\;\n\necho \"==> Deleting .gz archives older than ${delete_after_days}d\"\nfind \"$root\" -name \"*.gz\" -mtime \"+${delete_after_days}\" -delete\n\necho \"==> Disk usage after cleanup\"\ndu -sh \"$root\"\n```\n\n…plus a one-line plain-English summary in chat before it runs. Seven rules do\nthe work — name values, comment the *why*, banner each phase, fail loud, preview\ndeletions, prose first — and the agent knows to skip all of it for a bare `ls`\n\n.\nFull rules and per-language mechanics: [ AGENTS.md](/a-b/readable-ad-hoc-commands/blob/main/AGENTS.md). More transforms\n(\n\n`python -c`\n\n, SQL): [.](/a-b/readable-ad-hoc-commands/blob/main/EXAMPLES.md)\n\n`EXAMPLES.md`\n\n** skills.sh** (recommended — works for every supported agent):\n\n```\nnpx skills add a-b/readable-ad-hoc-commands       # this project\nnpx skills add a-b/readable-ad-hoc-commands -g    # global: all projects\n```\n\n**Claude Code, by hand** (bundled skill, auto-discovered on startup):\n\n```\ngit clone https://github.com/a-b/readable-ad-hoc-commands.git\ncp -r readable-ad-hoc-commands/readable-ad-hoc-commands ~/.claude/skills/   # all projects\n```\n\n**Any other agent** — the rules are plain Markdown in [ AGENTS.md](/a-b/readable-ad-hoc-commands/blob/main/AGENTS.md);\ndrop them where your tool reads instructions:\n\n| Agent | Where |\n|---|---|\n| Codex, Amp, Zed, Jules (\n|\n\n`AGENTS.md`\n\nat repo root`.cursor/rules/readable-commands.mdc`\n\n`.windsurf/rules/`\n\n`.github/copilot-instructions.md`\n\n`AGENTS.md`\n\nto the system prompt\n\n```\nAGENTS.md                       # harness-neutral rules (portable source of truth)\nEXAMPLES.md                     # worked transforms for python -c and SQL\nreadable-ad-hoc-commands/\n└── SKILL.md                    # Claude Code skill adapter (same rules + frontmatter)\n```\n\n", "url": "https://wpnews.pro/news/readable-ad-hoc-commands-agent-skill", "canonical_source": "https://github.com/a-b/readable-ad-hoc-commands", "published_at": "2026-08-01 00:25:22+00:00", "updated_at": "2026-08-01 00:52:24.740355+00:00", "lang": "en", "topics": ["ai-agents", "developer-tools", "ai-safety"], "entities": ["a-b/readable-ad-hoc-commands", "Claude Code", "Codex", "Amp", "Zed", "Jules", "Cursor", "Windsurf"], "alternates": {"html": "https://wpnews.pro/news/readable-ad-hoc-commands-agent-skill", "markdown": "https://wpnews.pro/news/readable-ad-hoc-commands-agent-skill.md", "text": "https://wpnews.pro/news/readable-ad-hoc-commands-agent-skill.txt", "jsonld": "https://wpnews.pro/news/readable-ad-hoc-commands-agent-skill.jsonld"}}