V.A.R. Protocol

Enforce Versioning, Archiving, and Rollback discipline for long-lived assets and multi-step delivery work. Use when editing codebases, websites, documents, decks, brand assets, prompts, configs, or any project artifact that needs safe iteration, backup-first handling, semantic version naming, changelogs, rollback anchors, or non-destructive updates. Trigger on requests about refactoring, major revisions, migration, redesign, rework, release packaging, version-control hygiene, or safe multi-agent collaboration. 中文简介：用于长期资产与多阶段交付的安全协作协议，核心是版本清晰、备份优先、遇错先回退，避免 destructive overwrite 导致资产丢失。

Apply V.A.R. whenever the cost of overwriting the wrong thing is higher than the cost of being slightly more disciplined.

中文简介

V.A.R. Protocol 是一个面向长期项目、重要资产和多阶段交付的协作规范。 它的核心原则是：版本清晰（Versioning）、备份优先（Archiving）、遇错先回退（Rollback）。 适用于网站改版、代码重构、文档/Deck 迭代、Prompt/Config 升级、多 Agent 协作等场景，目标是在推进迭代速度的同时，避免 destructive overwrite、内容丢失与不可逆回归。

V.A.R. stands for:

• Versioning
• Archiving
• Rollback

What this skill is for

Use this skill to turn risky editing work into a safer delivery flow. It is designed for code, content, design, and multi-agent execution where recoverability matters as much as speed.

Typical triggers:

• “Refactor this site without breaking the current version.”
• “Rewrite this proposal / deck, but keep a safe fallback.”
• “Migrate these prompts or configs and preserve rollback options.”
• “Have multiple agents work on the same long-lived asset safely.”
• “Package this release with a clear version, changelog, and rollback point.”

Core operating sequence

Follow this order:

1. Stabilize the baseline — identify the current working version or last known good artifact.
2. Back up before structural change — create a restorable copy before large edits, refactors, migrations, or agent handoffs.
3. Use explicit versions — name outputs with clear semantic versions instead of vague labels like final, latest, or new2.
4. Change incrementally — prefer small, reversible modifications over full replacement.
5. Attach a changelog — state what changed, why, and what the rollback anchor is.
6. Rollback first when broken — if the new version regresses, restore the last stable version before attempting another merge.

When to apply strict mode

Use the strictest form of V.A.R. when any of the following are true:

• The asset is long-lived or business-critical.
• Multiple agents or contributors are touching the same artifact.
• The task involves refactor, redesign, migration, or restructuring.
• The user may need to compare versions or recover prior work quickly.
• The output is a website, source tree, proposal, deck, prompt system, knowledge base, automation config, or brand asset set.

Minimum operating requirements

Before making a high-impact change:

• Identify the artifact owner and the current stable version.
• Create a backup or duplicate that can be restored quickly.
• Decide the next semantic version.
• Preserve or document the rollback anchor.

During the change:

• Avoid destructive overwrite when a parallel version is feasible.
• Prefer additive or surgical edits over wholesale replacement.
• Keep filenames and folders readable and comparable.

At delivery:

• State the new version.
• Provide a concise change summary.
• Name the rollback anchor explicitly.
• Call out any known risks or incomplete areas.

Default naming guidance

Prefer patterns like:

• [project]-[artifact]-v1.0.ext
• [project]-[artifact]-v1.1.ext
• [project]-[artifact]-v1.1.1.ext

Use version levels consistently:

• Major: architecture, structure, or strategy changed.
• Minor: meaningful feature/module expansion without core rewrite.
• Patch: small fixes, copy edits, image swaps, bug fixes.

Delivery contract

When delivering work under V.A.R., include:

• Version
• Change summary
• Rollback anchor
• Known risks if any remain

Failure handling

If the new version loses functionality, quality, or content:

1. Stop stacking more changes onto the broken version.
2. Restore or switch back to the last stable artifact.
3. Diagnose the failure from the stable baseline.
4. Re-apply only the needed changes in smaller slices.

Multi-agent rule

The lead agent owns backup and rollback safety. Do not assume a sub-agent already created a safe restore point unless it is explicitly confirmed.

Read next

For the full protocol text, delivery checklist, and example delivery block, read references/protocol.md.