Back to Details

agency-client-handoff

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Agency client handoff

代理机构客户交接

Package a completed engagement for transfer to the client's in-house team, DAM, or next agency — with the asset set, prompt library, model pins, rights documentation, and everything else the receiving team needs to regenerate consistent work without you in the loop.
Engagement done → reproducible handoff bundle. Knowledge transfer over lock-in: the receiving team should be able to keep the brand on-model without calling the agency back.
将已完成的项目成果打包，交付给客户内部团队、DAM、下一合作代理机构，其中包含资产集、提示词库、固定模型、权限文档，以及接收团队无需依赖你即可重新生成一致成果所需的全部内容。
项目完成 → 可复现的交接包。 注重知识传递而非锁定：接收团队应能在无需再次联系代理机构的情况下，保持品牌输出的一致性。

When to Use

使用场景

  • End of an engagement, client taking work in-house
  • Transition to a new agency of record — outgoing handoff
  • Completed pitch / campaign — asset package with provenance for the client's DAM
  • Contract-mandated "source file" delivery with regeneration rights
  • Year-end portfolio snapshot for a long-term retainer client
Do not use for work-in-progress reviews (use Drive links), approvals (use proof PDFs), or internal archival (use your agency's backup). Handoff is the formal, client-owned, reproducible bundle.
  • 项目结束，客户将工作收回内部处理
  • 更换合作代理机构时的转出交接
  • 完成提案/营销活动后，为客户DAM提供带来源证明的资产包
  • 合同要求交付带有再生权限的“源文件”
  • 为长期续约客户提供年终项目成果快照
请勿用于 进行中工作评审（请使用Drive链接）、审批（请使用样稿PDF）或内部存档（请使用代理机构的备份系统）。交接包是正式的、归客户所有的可复现成果包。

Prerequisites

前置条件

Ask the user (one message):
  1. Client slug + engagement scope — which client, which project or retainer period is being handed off
  2. Receiving team — in-house creative? Another agency? Which format and tools do they use (Figma, Adobe, CapCut, Notion)?
  3. Rights status — are all inputs licensed? Any stock, fonts, or reference faces that need redacting or citing?
  4. Regeneration expectation — does the client need to re-generate assets on their own, or is this a static archive?
  5. Confidentiality — any internal prompts, competitor references, or tagged metadata that must be stripped before delivery?
  6. Delivery format — zip + Drive link? Direct to client S3 / DAM? Physical drive?
If the client needs regeneration capability, they'll need the gen-ai CLI themselves — add an install guide to the bundle.
向用户询问以下信息（一次发送）：
  1. 客户标识（slug）+ 项目范围 — 涉及哪个客户，交接的是哪个项目或续约周期的成果
  2. 接收团队 — 是客户内部创意团队？还是其他代理机构？他们使用哪些格式和工具（Figma、Adobe、CapCut、Notion）？
  3. 权限状态 — 所有输入素材是否已获得授权？是否有需要编辑或标注来源的素材、字体或参考人脸？
  4. 再生需求 — 客户是否需要自行重新生成资产，还是仅需静态存档？
  5. 保密要求 — 是否有必须在交付前移除的内部提示词、竞品参考或标记元数据？
  6. 交付格式 — ZIP文件+Drive链接？直接上传至客户S3/DAM？还是物理存储介质？
如果客户需要再生能力，他们需自行安装gen-ai CLI — 请在交接包中添加安装指南。

How to Run

操作步骤

1. INVENTORY   → pull every asset + results.json from clients/<slug>/ across the engagement
2. FILTER      → exclude drafts, rejected variants, internal-only WIP
3. STRIP       → white-label: remove agency tags, internal prompt notes, competitor refs
4. PIN MODELS  → lock every prompt's model to an exact ID + version (not "latest")
5. DOCUMENT    → generate README, RIGHTS.md, CHANGELOG, prompt library
6. PACKAGE     → zip with consistent folder structure + naming
7. VERIFY      → extract and cold-test: can a fresh machine regenerate one asset?
8. DELIVER     → upload to client destination, share link, brief the receiving team
Rules:
  • Pin every model ID. 
    recraftv4
    → 
    recraftv4@2026-03-15
    (or the exact ID returned by 
    gen-ai models info
    ). "Latest" will drift and break reproducibility.
  • Strip internal metadata. Tags like 
    retainer-week-2026-04-22
    or 
    pitch-concept-b
    are internal — not for client eyes.
  • Include prompts, not just outputs. A prompt library lets the client's in-house team iterate without re-inventing the system.
  • Never ship a handoff you haven't cold-tested. Unzip, pick one prompt, regenerate — if the output drifts, something isn't pinned.
1. INVENTORY   → pull every asset + results.json from clients/<slug>/ across the engagement
2. FILTER      → exclude drafts, rejected variants, internal-only WIP
3. STRIP       → white-label: remove agency tags, internal prompt notes, competitor refs
4. PIN MODELS  → lock every prompt's model to an exact ID + version (not "latest")
5. DOCUMENT    → generate README, RIGHTS.md, CHANGELOG, prompt library
6. PACKAGE     → zip with consistent folder structure + naming
7. VERIFY      → extract and cold-test: can a fresh machine regenerate one asset?
8. DELIVER     → upload to client destination, share link, brief the receiving team
规则：
  • 固定每个模型ID。 
    recraftv4
    → 
    recraftv4@2026-03-15
    （或
    gen-ai models info
    返回的精确ID）。使用“latest”会导致输出偏差，破坏可复现性。
  • 移除内部元数据。 诸如
    retainer-week-2026-04-22
    或
    pitch-concept-b
    这类标签属于内部内容，不得提供给客户。
  • 包含提示词，而非仅输出成果。 提示词库可让客户内部团队无需重新搭建系统即可进行迭代。
  • 切勿交付未经过冷测试的交接包。 解压后选取一个提示词进行再生测试，如果输出出现偏差，说明有内容未被固定。

Quick Reference

快速参考

json
{
  "handoff_kind": "client-final",
  "client_slug": "acme-fintech",
  "engagement": "2025-Q3 to 2026-Q2 retainer",
  "delivered_at": "2026-04-22",
  "brand_system": "docs/brand-system.json",
  "brand_rules": "docs/brand.md",
  "pinned_models": "docs/model-pins.json",
  "assets_count": 247,
  "prompts_count": 54,
  "rights_status": "all generated, no licensed stock",
  "regeneration_supported": true,
  "support_window_days": 30
}
Include this as 
docs/handoff-manifest.json
— a single file that tells the receiving team what they got and what they can do with it.
json
{
  "handoff_kind": "client-final",
  "client_slug": "acme-fintech",
  "engagement": "2025-Q3 to 2026-Q2 retainer",
  "delivered_at": "2026-04-22",
  "brand_system": "docs/brand-system.json",
  "brand_rules": "docs/brand.md",
  "pinned_models": "docs/model-pins.json",
  "assets_count": 247,
  "prompts_count": 54,
  "rights_status": "all generated, no licensed stock",
  "regeneration_supported": true,
  "support_window_days": 30
}
将此文件保存为
docs/handoff-manifest.json
— 该文件可向接收团队说明交接包内容及使用权限。

Quick Reference

快速参考

Handoffs don't generate new assets — they package existing ones. But for reproducibility checks:
TaskModelNotes
Dry-run regeneration test (cheapest)
gemini-3.1-flash-image
Use to verify payload validity, not output quality
Full regeneration proof (one asset)Same as original, pinnedMatches the original bundle; compare outputs visually
Re-export at higher resolution
topaz-upscale-image
If the client DAM wants 4K versions
Never substitute a newer model for the pinned one during handoff — the whole point is that the client can reproduce the agency's output exactly.
交接包不会生成新资产，仅对现有资产进行打包。但为了验证可复现性，可执行以下操作：
任务模型说明
再生测试空跑（成本最低）
gemini-3.1-flash-image
用于验证请求有效性，而非输出质量
完整再生验证（单个资产）与原始固定模型一致需与原始包输出匹配，进行视觉对比
高分辨率重新导出
topaz-upscale-image
当客户DAM需要4K版本时使用
交接过程中切勿用新版本模型替代固定模型 — 核心目标是让客户能精确复现代理机构的输出成果。

Procedure

流程规范

  • Reproducibility is the deliverable. If the client can't regenerate without you, the handoff failed — even if the asset folder is beautiful.
  • Strip, don't redact. Remove internal tags entirely; the bundle should look native to the client.
  • Version the handoff — 
    handoff-v1/
    , 
    handoff-v2/
    . Don't overwrite if an updated bundle is requested later.
  • Include a sunset date for agency support — clarifies when the client self-serves or re-engages.
  • Pin exact model IDs with 
    gen-ai models info <id> --json
    — preserves capabilities at handoff time.
  • Test on a clean machine. Your laptop has tokens, fonts, caches. The bundle has to work without those.
  • No lock-in flags. Everything in the bundle runs on a stock 
    gen-ai
    install.
  • Ship prompts, brand, and rationale. Context beats artifacts for knowledge transfer.
  • 可复现性是核心交付标准。 如果客户无法在不依赖你的情况下再生资产，即便资产文件夹整理得再规范，交接也视为失败。
  • 彻底移除，而非编辑。 完全删除内部标签，交接包应看起来完全适配客户的使用习惯。
  • 为交接包版本化 — 如
    handoff-v1/
    、
    handoff-v2/
    。若后续客户要求更新交接包，请勿覆盖原有版本。
  • 注明代理机构支持的终止日期 — 明确客户何时需自行服务或重新合作。
  • 使用
    gen-ai models info <id> --json
    固定精确模型ID — 保留交接时的模型功能。
  • 在干净环境的机器上测试。 你的笔记本电脑可能存有令牌、字体、缓存，交接包需在无这些内容的环境下正常运行。
  • 无锁定限制。交接包内所有内容均可在标准
    gen-ai
    安装环境下运行。
  • 交付提示词、品牌规范及设计思路。 知识传递中，背景信息比成果本身更重要。

Pitfalls

常见陷阱

  • Unpinned models — client regenerates 6 months later, output drifts, blames the agency.
  • Internal metadata leakage — competitor refs, pitch-concept tags, reviewer names surviving in prompt files.
  • Missing rights documentation — client legal can't sign off on re-use, bundle is DOA.
  • Proprietary tools in the regen path — strip to stock 
    gen-ai
    CLI only.
  • No cold test — bundle ships, step 1 of the README fails, urgent Slack follows.
  • One-and-done mentality — promise 30 days of Q&A post-handoff, price it in.
  • 未固定模型 — 客户6个月后再生资产时输出偏差，归咎于代理机构。
  • 内部元数据泄露 — 提示词文件中残留竞品参考、提案概念标签或审核人员姓名。
  • 缺失权限文档 — 客户法务无法批准资产复用，交接包直接失效。
  • 再生流程中使用专有工具 — 需仅保留标准
    gen-ai
    CLI。
  • 未进行冷测试 — 交接包交付后，README中的第一步操作失败，引发紧急沟通。
  • 一劳永逸的心态 — 应承诺交接后30天的答疑服务，并将其纳入定价。

Verification

验证方法

Run 
gen-ai whoami
to confirm authentication, then re-run the failed command with 
--debug
.
运行
gen-ai whoami
确认身份验证，然后添加
--debug
参数重新运行失败的命令。

Step 1: Inventory and filter

步骤1：盘点与筛选

bash
CLIENT="acme-fintech"
HANDOFF="handoff/$CLIENT-$(date +%Y-%m-%d)"
mkdir -p "$HANDOFF"/{assets,prompts,source,docs}

find "clients/$CLIENT" -name "results.json" -exec cp {} "$HANDOFF/source/" \;
cp -r "clients/$CLIENT/deliverables/"* "$HANDOFF/assets/"
node scripts/extract-prompts.js "$HANDOFF/source/" > "$HANDOFF/prompts/prompt-library.json"
bash
CLIENT="acme-fintech"
HANDOFF="handoff/$CLIENT-$(date +%Y-%m-%d)"
mkdir -p "$HANDOFF"/{assets,prompts,source,docs}

find "clients/$CLIENT" -name "results.json" -exec cp {} "$HANDOFF/source/" \;
cp -r "clients/$CLIENT/deliverables/"* "$HANDOFF/assets/"
node scripts/extract-prompts.js "$HANDOFF/source/" > "$HANDOFF/prompts/prompt-library.json"

Step 2: Pin models, strip internal data

步骤2：固定模型，移除内部数据

bash
undefined
bash
undefined

Get pinned model IDs for every model the engagement used

Get pinned model IDs for every model the engagement used

for model in $(jq -r '.jobs[].model' "$HANDOFF/source/"*.json | sort -u); do gen-ai models info "$model" --json >> "$HANDOFF/docs/model-pins.json" done
for model in $(jq -r '.jobs[].model' "$HANDOFF/source/"*.json | sort -u); do gen-ai models info "$model" --json >> "$HANDOFF/docs/model-pins.json" done

Strip internal tags from the prompt library

Strip internal tags from the prompt library

jq 'del(.jobs[].tags[] | select(startswith("internal-") or startswith("retainer-")))'
"$HANDOFF/prompts/prompt-library.json" > "$HANDOFF/prompts/prompt-library.clean.json" mv "$HANDOFF/prompts/prompt-library.clean.json" "$HANDOFF/prompts/prompt-library.json"
undefined
jq 'del(.jobs[].tags[] | select(startswith("internal-") or startswith("retainer-")))'
"$HANDOFF/prompts/prompt-library.json" > "$HANDOFF/prompts/prompt-library.clean.json" mv "$HANDOFF/prompts/prompt-library.clean.json" "$HANDOFF/prompts/prompt-library.json"
undefined

Step 3: Documentation

步骤3：文档生成

Generate a 
README.md
covering: folder walkthrough, regeneration steps (CLI install + one worked example), pinned model note, rights status, support contact + sunset date. Include 
docs/brand-system.json
, 
docs/brand.md
, 
docs/model-pins.json
, 
docs/RIGHTS.md
, 
docs/CHANGELOG.md
.
Minimum regeneration steps to include in the README:
1. Install CLI: curl -fsSL https://picsart.com/gen-ai-cli/install.sh | bash
2. gen-ai login
3. Pick a prompt from prompts/prompt-library.json
4. gen-ai generate -m <pinned-model-id> -p "<prompt>"
生成
README.md
文件，内容包括：文件夹说明、再生步骤（CLI安装+示例）、固定模型说明、权限状态、支持联系人及终止日期。同时包含
docs/brand-system.json
、
docs/brand.md
、
docs/model-pins.json
、
docs/RIGHTS.md
、
docs/CHANGELOG.md
文件。
README中需包含的最低限度再生步骤：
1. Install CLI: curl -fsSL https://picsart.com/gen-ai-cli/install.sh | bash
2. gen-ai login
3. Pick a prompt from prompts/prompt-library.json
4. gen-ai generate -m <pinned-model-id> -p "<prompt>"

Step 4: Package + verify

步骤4：打包与验证

bash
cd handoff
zip -r "$CLIENT-handoff-$(date +%Y-%m-%d).zip" "$CLIENT-$(date +%Y-%m-%d)/" \
  -x "*.DS_Store" "*/.git/*"
bash
cd handoff
zip -r "$CLIENT-handoff-$(date +%Y-%m-%d).zip" "$CLIENT-$(date +%Y-%m-%d)/" \
  -x "*.DS_Store" "*/.git/*"

Cold test — extract to a clean dir, regenerate one asset

Cold test — extract to a clean dir, regenerate one asset

tmpdir=$(mktemp -d) && unzip -q "$CLIENT-handoff-$(date +%Y-%m-%d).zip" -d "$tmpdir" cd "$tmpdir/$CLIENT-$(date +%Y-%m-%d)" SAMPLE_PROMPT=$(jq -r '.jobs[0].prompt' prompts/prompt-library.json) SAMPLE_MODEL=$(jq -r '.jobs[0].model' prompts/prompt-library.json) gen-ai generate -m "$SAMPLE_MODEL" -p "$SAMPLE_PROMPT" --dry-run --debug
tmpdir=$(mktemp -d) && unzip -q "$CLIENT-handoff-$(date +%Y-%m-%d).zip" -d "$tmpdir" cd "$tmpdir/$CLIENT-$(date +%Y-%m-%d)" SAMPLE_PROMPT=$(jq -r '.jobs[0].prompt' prompts/prompt-library.json) SAMPLE_MODEL=$(jq -r '.jobs[0].model' prompts/prompt-library.json) gen-ai generate -m "$SAMPLE_MODEL" -p "$SAMPLE_PROMPT" --dry-run --debug

Dry-run validates = bundle is regeneration-ready.

Dry-run validates = bundle is regeneration-ready.

undefined
undefined

Cost & time

成本与时间

PhaseSpendTime
Inventory + filter + strip$01-2 hrs
Documentation (README, RIGHTS, CHANGELOG)$01-2 hrs
Cold-test regeneration (1-2 assets)~$0.5015 min
Total handoff<$1~4 hrs
The real cost is time, not credits. Budget a half-day per handoff; cutting corners creates support calls for months after.
阶段花费时间
盘点+筛选+移除$01-2小时
文档生成（README、RIGHTS、CHANGELOG）$01-2小时
再生冷测试（1-2个资产）~$0.5015分钟
总交接成本<$1约4小时
真正的成本是时间，而非算力credits。每个交接包需预留半天时间；偷工减料会导致后续数月的支持请求。

See also

相关链接

  • workflows/agency-brand-scoping/
    — 
    brand-system.json
    + 
    brand.md
    that ship in the handoff
  • workflows/agency-multi-brand-pack/
    — source of per-client results.json archive
  • workflows/agency-pitch-mockups/
    — pitch bundles that become part of the final handoff
  • gen-ai-use.md
    — regeneration reference for the client
  • workflows/agency-brand-scoping/
    — 交接包中包含的
    brand-system.json
    和
    brand.md
    文件来源
  • workflows/agency-multi-brand-pack/
    — 各客户results.json归档文件的来源
  • workflows/agency-pitch-mockups/
    — 最终交接包中包含的提案样稿包
  • gen-ai-use.md
    — 供客户使用的再生操作参考文档