# Byte Build

> Implement or change a codebase, document, prototype, or other deliverable and verify it. Use when the user asks to build, fix, integrate, or execute an existing plan.

- Skill: `elan6666/byte-build` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add elan6666/byte-build`
- Raw SKILL.md: https://api.skillmd.com/api/skills/elan6666/byte-build/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-build

---


# Byte Build

Implement the user's requested outcome with the smallest sound change set.

## Execution

Work from the relevant files, instructions, current behavior, and active lessons.
Identify the intended result and useful verification, make focused changes while
preserving unrelated work, and repair important in-scope failures. Choose the
sequence and depth that fit the task rather than treating these as fixed stages.

Use an existing plan when helpful, but do not require Byte OS plans, OKRs,
harness files, waves, or role assignments before working. Adjust a stale plan
when live evidence shows a better route.

Prefer simple implementations and narrow diffs. Broaden the change only when the
goal or discovered architecture genuinely requires it. Avoid opportunistic
refactors that do not improve the requested result.

Use persistent logs only for long-running, risky, or resumable work. When state is
useful, record concise facts and verification rather than maintaining several
parallel ledgers.

Completion means the requested change is present and relevant verification has
passed, or any remaining verification limit is clearly disclosed.

When user correction or direct evidence confirms a reusable mistake, create or
update one deduplicated `.byte-os/LESSONS.md` entry with the correction, evidence,
and prevention rule. Exclude routine debugging, trivial slips, and sensitive data.

## Long-Running Execution

Default to [the bounded-goal workflow](references/long-running-work.md)
for complex implementation. When a long external job is needed, finish the
preparation goal with the monitoring plan designed, then launch, verify the job
identity, and enable monitoring subject to the workflow's tool-authority rules.
Disable and verify the monitor after terminal evidence, before any new goal;
a job submission does not establish the requested result.

## 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.