Autonomous execution

SkillWeb & browsing

Persistent PLAN→EXECUTE→VERIFY→RECOVER loop for computer tasks. USE WHEN: multi-step desktop/browser/file work, anything that can fail mid-way. DO NOT USE FOR: pure chat questions with no side effects.

Available today. Use it from your connected AI after setup.

Add ahel to your AI once: Claude, ChatGPT, Cursor, Claude Code or Codex. Then ask it to use this.

Then ask your AI: use the Autonomous execution skill

What this skill tells your AI

The instructions your AI receives, as published by safiyev/heyagent in skills/autonomous-execution/SKILL.md and read by ahel’s review.

Protocol

  1. PLAN — numbered steps with expected outcome each.
  2. EXECUTE — one meaningful tool action (prefer API/file/shell over UI).
  3. VERIFY — screenshot / file_exists / URL / tool output must match intent.
  4. RECOVER — on fail, diagnose; up to 3 alternatives; then escalate.
  5. FINISH — report only verified paths/URLs; never invent DONE.

Examples

  • input: «открой блокнот и напиши рассказ о мальчике» tools: notepad.write content=проза-рассказ title=rasskaz_o_malchike destination=desktop never: short rhyme / файл «стишок»

  • input: «открой мне дым сигарет с ментолом» tools: youtube.open.lesson mode=quick → first organic match never: scroll past the correct top video

  • input: «пройди тест на вкладке» tools: browser.tabs.list → focus → answer → verify → Next (never Next blind)

Never stop after the first tool failure. Ask the owner only for CAPTCHA, login, payment, or irreversible choice.

Signals

GitHub stars
52
Forks
6
Last commit
Sep 2026
Advanced
Item type
skill
Key
autonomous-execution
Source
github.com/safiyev/heyagent