# Byte Auto

> Complete a multi-step product or project request end to end and verify the result. Use when the user asks for one-click completion, all necessary work, or persistence until the outcome is genuinely handled.

- Skill: `elan6666/byte-auto` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add elan6666/byte-auto`
- Raw SKILL.md: https://api.skillmd.com/api/skills/elan6666/byte-auto/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: elan6666 (https://skillmd.com/u/elan6666)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/elan6666/byte-auto

---


# Byte Auto

Own the requested outcome end to end. Adapt the method to the task instead of
running every Byte skill or producing every Byte OS artifact.

## Outcome Contract

At the start, identify:

- the deliverable or observable result;
- important constraints and evidence already available;
- what completion can be verified now;
- actions that require new authority or user input.

Infer ordinary implementation details. Ask only when a missing choice is both
material and unsafe to infer.

## Adaptive Loop

Repeat as useful:

1. Inspect the current files, runtime, evidence, prior work, and relevant active lessons.
2. Decide the highest-value next action.
3. Research, plan, implement, or review only as needed.
4. Verify in proportion to risk.
5. Repair important failures and reassess completion.
6. Record a reusable lesson when evidence confirms a meaningful mistake or the
   user corrects a requirement misunderstanding.

Do not use a fixed iteration count. A small task may need one pass; a difficult
task may need several. Do not stop merely because a stage or plan finished.

Research is optional unless current external facts affect the result. A written
plan is optional unless it reduces coordination or execution risk. Product specs,
OKRs, role analyses, harnesses, and review documents are optional tools, not gates.

## Completion

Finish when the requested deliverable exists, relevant verification has passed or
its limits are disclosed, and no high-priority in-scope issue remains that can be
safely resolved.

Pause only when progress requires user input, credentials, payment, unavailable
external state, a consequential product decision, or authority for a destructive
or irreversible action. Report the exact blocker and the smallest next action.

## Persistent State

For long or resumable work, keep a concise `.byte-os/STATE.md` or update the
project's existing state mechanism. Record only what a future session needs:
goal, evidence, decisions, completed work, verification, blockers, and next step.
Do not create a full Byte OS document set by default.

Use `.byte-os/LESSONS.md` as the automatic error notebook. Keep only confirmed,
reusable lessons with the mistake, corrected understanding or evidence, and a
specific prevention rule. Deduplicate recurring mistakes and never store secrets
or routine trial-and-error noise.

## Handoff

Lead with the delivered result. Summarize meaningful changes, verification, and
remaining limits. Do not end with a workflow command when safe in-scope work is
still unfinished.

## Goals And Long-Running Work

Default to bounded goals for complex execution tasks when tool authority and
availability permit; use ordinary phase state otherwise. Read
[the long-running workflow](references/long-running-work.md): design monitoring
inside the preparation goal, complete that goal before launch and monitor
activation, then disable the monitor before any next goal. An active goal and
enabled monitor for the same workflow must not overlap. Overall completion still requires the
user's final deliverable. Scheduled waiting is a resumable handoff, not a blocker
or task completion.

## Source And Updates

Canonical repository: [elan6666/your-bytedance-skills](https://github.com/elan6666/your-bytedance-skills). Use its current `main` branch when checking for or installing updates.