BeautyPlus Skill

When to Use This Skill

Activate when the user wants any of the following on a photo / image (path, URL, or IM attachment):

• Body reshape (figure) — breast enhancement (natural / teardrop / round / outward), peach butt, O-shape butt, etc., each with strong / medium / weak tiers
• Hair — color — natural black, blonde, brown highlights, platinum, silver platinum, teddy warm brown, etc.
• Hair — style — glossy hair, layered cut, soft waves, Latino curls, etc.
• Outfits — formal — gowns, rhinestone mesh dress, feather dress, beaded dress, tartan suit, black suit, etc.
• Outfits — vacation — bunny ears, slip dress, puff dress, hoodie dress, lace corset set, tiered chiffon dress, sheer bikini overlay, etc.
• Outfits — cosplay — carnival, bunny cop, fox shirt, deer girl skirt, Grinch, Victoria Angel, Dallas Cowboy, etc.
• Outfits — party — Floral Camisole, Puff Skirt, One-Shoulder LBD, Red Latex, Moonlight Shimmer Dress, Y3K Set, etc.
• Outfits — sports — Brazilian Bikini, Tennis Set, Cozy Fit, Racing Suit, White Yoga, etc.
• Face style — Natural Beauty, Glamour Beauty, Sweet Beauty, Luminous Beauty, Youthful Beauty
• Expression — Closed Smile, Open Smile, Cool Expression, Wink
• Photo art — Tanning Filter, CCD Flash, Film Flash, Fuji Flash
• Photo restoration — denoise / repair, AI ultra-HD upscaling

Effect KEY: The CLI --task value must be the effect KEY string from the table below. The algorithm spec for each key is returned inline by POST /skill/consume.json (invoke_spec) — do not hard-code AIGC paths.

Billing and user-facing claims (MANDATORY)

• Fact: Each successful run-task (including inside a sessions_spawn worker) goes through server-side quota / credit consumption for the BP_AK tenant. This is a paid, metered commercial API, not free compute bundled with the skill or the host.
• Forbidden: Do not state or imply that the service is free, costs nothing, uses no quota, has unlimited trial, or similar. Do not invent prices, plan names, promotions, or trial rules.
• Allowed: Neutral wording — e.g. processing uses the BeautyPlus account quota tied to the configured keys; billing and plans are per your console or administrator. If the user asks about cost, point them to admin / official billing docs / console; do not guess. When the API returns quota or membership errors, follow Step 3 — MANDATORY (quota / consume failures) using server detail and pricing_url when present.
• On success too: Success summaries must stay factual (task completed, delivery). Do not add “free” or zero-cost implications.

Supported Algorithms (effect KEY → --task)

All rows use image input. Algorithm params for each key are returned as invoke_spec by POST /skill/consume.json — do not hard-code AIGC paths.

Body reshape — figure

Tiers: strong / medium / weak (append _strong / _medium / _weak to the key). If unspecified, default to _medium.

Hair — color

Hair — style

Outfits — formal

Outfits — vacation

Outfits — cosplay

Outfits — party

Outfits — sports

Face style

Expression

Photo art

Photo restoration

Video async path (reserved)

Current catalog is image-only. Use §3a (run-task) for every listed key.

Future video keys: When the server publishes video effect keys, use spawn-run-task → sessions_spawn per §3b (runTimeoutSeconds default 3600). See docs/errors-and-polling.md for polling details.

Multi-stage pipelines (chaining tasks)

When the user asks for more than one BeautyPlus step on the same media (e.g. photo restoration then outfit change), treat each step as a separate job with its own --task (effect KEY):

Rules:

1. After stage A completes (skill_status: "completed"), pass primary_result_url or output_urls[0] as --input for the next --task. That is a new job, not a retry.
2. "Do not re-run run-task" means: do not resubmit the same task_id. It does not forbid the next pipeline stage with a different effect KEY.
3. Delivery: Prefer final-stage delivery for the full pipeline. Deliver after the last stage only.
4. Video chains (reserved): One sessions_spawn = one embedded run-task. Chain = multiple spawns. Current catalog uses §3a only.

See also Step 3 success bullets and agent_instruction in the JSON.

API submission path (MANDATORY)

• New jobs: Submit only via python3 {baseDir}/scripts/beautyplus_ai.py run-task … (§3a / §3b), or the same run-task command embedded in spawn-run-task → sessions_spawn. Do not hand-craft HTTP to the skill’s wapi gateway or AIGC / invoke endpoints to replace that flow — that skips POST /skill/consume.json (quota and permission) and breaks the supported pipeline.
• Exception: query-task --task-id is only for resuming status polling on an existing full task_id (no upload, no second consume). Do not use it instead of run-task for a new submission.
• No curl replay: This skill does not emit debug curl for API calls. Do not hand-craft HTTP to wapi / AIGC to mimic requests — always use the CLI above so /skill/consume.json runs before algorithm submit.

0. Pre-Flight Check (MANDATORY — run before anything else)

Verify AK/SK are configured (only run this command; do not read other Python sources first):

python3 {baseDir}/scripts/beautyplus_ai.py preflight

• Output ok → continue to Step 1
• Output missing → stop and send the user the configuration message below

Feishu — send an interactive card via the Feishu API (do not use the message tool for this):

import json, urllib.request
cfg = json.loads(open("/home/ec2-user/.openclaw/openclaw.json").read())
feishu = cfg["channels"]["feishu"]["accounts"]["default"]
token = json.loads(urllib.request.urlopen(urllib.request.Request(
    "https://open.feishu.cn/open-apis/auth/v3/tenant_access_token/internal",
    data=json.dumps({"app_id": feishu["appId"], "app_secret": feishu["appSecret"]}).encode(),
    headers={"Content-Type": "application/json"}
)).read())["tenant_access_token"]
card = {
    "config": {"wide_screen_mode": True},
    "header": {"title": {"tag": "plain_text", "content": "🖼️ BeautyPlus — credentials required"}, "template": "blue"},
    "elements": [{"tag": "div", "text": {"tag": "lark_md", "content": "1. Apply for **Access Key** and **Secret Key** at [BeautyPlus Developers](https://beautyplus.com/developers).\
2. Set **BP_AK** and **BP_SK** in `scripts/.env` (see `scripts/.env.example`), then reload env:\
```\
source scripts/.env\
```\
If keys are issued by your organization, ask your administrator."}}],
}
urllib.request.urlopen(urllib.request.Request(
    "https://open.feishu.cn/open-apis/im/v1/messages?receive_id_type=open_id",
    data=json.dumps({"receive_id": "\x3CUSER_OPEN_ID>", "msg_type": "interactive", "content": json.dumps(card)}).encode(),
    headers={"Authorization": f"Bearer {token}", "Content-Type": "application/json"}
))

Telegram / Discord / other channels — use the message tool with plain text:

🖼️ BeautyPlus — credentials required

1. Get Access Key and Secret Key (apply here if needed):
   https://beautyplus.com/developers

2. Set BP_AK and BP_SK in scripts/.env (see scripts/.env.example), then run:
   source scripts/.env

If keys are issued by your organization, ask your administrator.

Step 1 — Pick effect KEY and input

Choose --task = effect KEY from Supported Algorithms (must exist in server algorithm.invoke after config fetch). Confirm the input file location.

Intent → effect KEY (MANDATORY checklist):

1. Map user intent to one row — Match feature / effect name / scene to an effect KEY in the table (English snake_case). If the user only states a broad category (e.g. “change my hairstyle”), ask one clarifying question (e.g. soft waves vs. Latino curls) or offer 2–3 KEYs to pick from.
2. Body reshape tiers — Natural breast, teardrop, round, outward, peach butt, and O-shape butt each have strong / medium / weak (*_strong / *_medium / *_weak). If unspecified, default to medium or confirm briefly before submit.
3. Input medium — All keys are still images (.jpg / .jpeg / .png / .webp / .gif / .bmp). Use §3a run-task. If video keys appear later, follow §3b.
4. Ambiguous intent — With no attachment and vague intent, ask one question (which effect / any reference image) or pull media from IM per docs/im-attachments.md; do not guess the wrong KEY.
5. Video + still in one message — While no video KEYs exist, use the user-specified image as --input. If video keys exist later, apply the video path to the video only.

Getting media from IM messages (full detail: docs/im-attachments.md):

python3 {baseDir}/scripts/beautyplus_ai.py resolve-input --file /tmp/saved.jpg --output-dir /tmp
# or: --url, --telegram-file-id, --feishu-image-key + --feishu-message-id

Use the JSON path field as --input.

--input as a URL: Quote the full URL in the shell (avoids & splitting). Defaults: 120 s read timeout, 100 MB max — override with MT_AI_URL_READ_TIMEOUT, MT_AI_URL_CONNECT_TIMEOUT, MT_AI_URL_MAX_BYTES. For flaky or large links, use resolve-input --url first and pass the local path.

If the user already gave a path or URL when triggering the skill, go to Step 2 without asking again.

Reply immediately to acknowledge the task, for example:

"🖼️ Processing — please wait a moment…"

Step 2 — Install dependencies

python3 {baseDir}/scripts/beautyplus_ai.py install-deps

If dependencies are already installed this step is quick; then continue to Step 3.

Step 3 — Run the task

Default: All listed effect KEYs are image jobs — use §3a (run-task).

Reserved: If a future video effect key appears that the CLI treats as spawn-only, use §3b (spawn-run-task + sessions_spawn). No current keys hit this branch.

3a — Inline (blocking, default for all catalog effect KEYs)

Blocking call — use for all listed effect KEYs.

python3 {baseDir}/scripts/beautyplus_ai.py run-task \
  --task "\x3Ceffect_key>" \
  --input "\x3Cimage_url_or_path>"

Replace \x3Ceffect_key> (e.g. hair_soft_waves) and \x3Cimage_url_or_path> with real values. If the server returns Unknown invoke preset, the key is not in the tenant's invoke map — do not invent params; check config or admin.

Default params include rsp_media_type: url. For custom JSON params:

python3 {baseDir}/scripts/beautyplus_ai.py run-task \
  --task "\x3Ceffect_key>" \
  --input "\x3Curl_or_path>" \
  --params '{"parameter":{"rsp_media_type":"url"}}'

When run-task exits 0, stdout is JSON that includes:

• skill_status: "completed" — job done; result is ready.
  • Single stage → proceed to Step 4.
  • Multi-stage pipeline → pass primary_result_url as --input for the next --task; deliver after the last stage.
  • Do not resubmit run-task for the same task_id; use query-task to resume polling.
• output_urls — ordered http(s) links (same extraction as before: data.result.urls, images, media_info_list, etc.).
• primary_result_url — same as output_urls[0] when present; convenient for delivery scripts.
• task_id — full task id as a top-level string when known (from data.result.id or the polling session). Keep it for manual status recovery or support handoff; do not truncate. Some synchronous completions may omit it if the API does not return an id.
• agent_instruction — short reminder for the model.
• meta / data — full API payload for debugging.

On skill_status: "completed" you must:

1. Send the user a short natural-language summary of what was done.
2. Complete Step 4 delivery using primary_result_url or output_urls[0] — unless the user explicitly asked for the URL only.

Do not end the turn with raw JSON alone.

When run-task exits non-zero, stdout has skill_status: "failed" — explain the error to the user; do not deliver or treat as success.

Quota / consume failures — when failure_stage: "consume_quota":

Never dump raw JSON to the user. Never retry run-task for credit/membership errors.

Video (reserved): When §3b applies, see docs/errors-and-polling.md and §3c–§3d for polling, timeouts, and recovery.

3b — Async worker (sessions_spawn, reserved for video effect keys)

Current catalog is image-only. spawn-run-task is not used — CLI rejects it for image keys. Use §3a for all listed effect KEYs.

When video keys return, the main agent does not block; a sub-session runs run-task and delivers the result.

1. Build the payload (\x3Ceffect_key> must be a video task name accepted by spawn-run-task — historically e.g. videoscreenclear / hdvideoallinone when server + CLI expose them):

python3 {baseDir}/scripts/beautyplus_ai.py spawn-run-task \
  --task "\x3Ceffect_key>" \
  --input "\x3Cvideo_url_or_path>" \
  --deliver-to "\x3Coc_xxx_or_ou_xxx_or_chat_id>" \
  --deliver-channel "feishu"

Optional: --params '\x3Cjson>' (same as run-task), --deliver-channel telegram|discord|..., --run-timeout-seconds (default 3600, aligned with extended poll budget). Do not reduce runTimeoutSeconds below the payload default unless you accept timeout risk — wall time varies (often minutes to tens of minutes).

2. Call OpenClaw sessions_spawn with the printed sessions_spawn_args (task, label, runTimeoutSeconds) without reducing runTimeoutSeconds unless you intentionally accept timeout risk.

3. Reply immediately to the user that processing has started. The sub-agent runs install-deps (if needed), run-task, then Step 4 per the embedded task text.

Multi-stage + spawn: One embed = one run-task. Image chains (current): §3a only. Video chains (reserved): see Multi-stage pipelines, rule 4.

3c — Resume polling (query-task)

When you already have a full task_id (from a previous stdout JSON, e.g. success, poll_timeout, or poll_aborted, or from stderr task_id=... lines) and the job may still be running on the server — do not run run-task again for that id; resume polling only:

python3 {baseDir}/scripts/beautyplus_ai.py query-task \
  --task-id "\x3Cfull_task_id>"

Optional --task sets the task_name field in the success JSON for your logs (default labels as query_task). Uses the same BP_AK / BP_SK and remote config as the original submit. Stdout JSON and exit codes match run-task: exit 0 with skill_status: "completed" when the task finishes successfully; exit non-zero with skill_status: "failed" / error on timeout, query errors, or API-reported failure.

3d — Last task and history (user-visible)

Local state under ~/.openclaw/workspace/beautyplus-ai/ (last_task.json, history/task_*.json, last 50 records). For async run-task, last_task.json may briefly show skill_status: "polling" with task_id while the client is still polling (checkpoint so query-task can resume if the process is killed mid-poll):

python3 {baseDir}/scripts/beautyplus_ai.py last-task
python3 {baseDir}/scripts/beautyplus_ai.py history

Use when the user asks whether a recent job finished, or for a short history summary. Do not expose raw secrets.

Step 4 — Deliver result to the channel

Required after success: When skill_status is completed, deliver here — the CLI does not post to IM by itself. Send the processed image or video back on the user’s platform (and keep the Step 3 MANDATORY summary in the same turn).

Resolve deliver-to target

Feishu — image tasks

python3 {baseDir}/scripts/feishu_send_image.py \
  --image "\x3Cresult_url>" \
  --to "\x3Coc_xxx or ou_xxx>"

Feishu — video tasks (reserved; e.g. legacy videoscreenclear / hdvideoallinone when video keys exist)

curl -sL -o /tmp/beautyplus_result.mp4 "\x3Cprimary_result_url_or_output_urls[0]>"
python3 {baseDir}/scripts/feishu_send_video.py \
  --video /tmp/beautyplus_result.mp4 \
  --to "\x3Coc_xxx or ou_xxx>" \
  --video-url "\x3Cprimary_result_url_or_output_urls[0]>" \
  [--cover-url "\x3Coptional_thumb_url>"] \
  [--duration \x3Cmilliseconds_if_known>]

--video-url adds a second message with the download link. Optional cover/duration; details: docs/feishu-send-video.md.

Telegram — image tasks

TELEGRAM_BOT_TOKEN="$TELEGRAM_BOT_TOKEN" python3 {baseDir}/scripts/telegram_send_image.py \
  --image "\x3Cresult_url>" \
  --to "\x3Cchat_id>" \
  --caption "✅ Done"

Telegram — video tasks (reserved; long async video jobs)

curl -sL -o /tmp/beautyplus_result.mp4 "\x3Cprimary_result_url_or_output_urls[0]>"
TELEGRAM_BOT_TOKEN="$TELEGRAM_BOT_TOKEN" python3 {baseDir}/scripts/telegram_send_video.py \
  --video /tmp/beautyplus_result.mp4 \
  --to "\x3Cchat_id>" \
  --video-url "\x3Cprimary_result_url_or_output_urls[0]>" \
  [--cover-url "\x3Coptional_thumb_url>"] \
  [--duration \x3Cseconds>] \
  --caption "✅ Done"

--video-url sends a follow-up text message with the download link. Max ~50 MB for Bot API video; larger files rely on the link line.

Discord

Download the result, then send with the message tool (use .mp4 for video, .jpg / .png for image):

curl -L "\x3Cresult_url>" -o /tmp/result_image.jpg

Then:

message(action="send", channel="discord", target="\x3Cchannel_id>", filePath="/tmp/result_image.jpg")

For files over ~25MB, send the result URL as a link instead.

WhatsApp / Signal / others

Use the message tool with media, or send the result URL directly.

Quick commands reference (agent)

Notes

• Single business entrypoint: algorithm runs and config fetch go through beautyplus_ai.py; agents do not need to open client.py / ai/api.py. Must not bypass this with direct HTTP to AIGC/wapi for new jobs — see API submission path (MANDATORY) above. query-task is the supported way to resume polling when a task_id is already known.
• Video tasks (reserved): When the CLI again accepts video-only --task values for spawn-run-task, use spawn-run-task + sessions_spawn in the main session; the worker runs run-task and delivery. Today: all catalog keys are image — use run-task (§3a) only; run-task in the main session is also for recovery (query-task). Polling and env tuning: docs/errors-and-polling.md.
• AK/SK loading: environment variables BP_AK / BP_SK first; if unset, scripts/.env is read automatically (same as SkillClient).
• Client init pulls the latest algorithm config from the server; no manual INVOKE setup.
• Bot token safety: pass TELEGRAM_BOT_TOKEN and similar only via environment variables — never as CLI arguments.
• On failure: stdout JSON has skill_status: "failed" / error, exit code ≠ 0 — explain to the user; check AK/SK, network, quotas; timeouts / SIGKILL / no final JSON: docs/errors-and-polling.md. URL input errors may mention HTTP 403 (expired signed URL) or timeout — see MT_AI_URL_* env vars above.
• More docs: README.md, docs/multi-platform.md, docs/im-attachments.md, docs/feishu-send-video.md.