Evan
102b506b0b
All 32 agent manifests and 17 hands shipped with empty mcp_servers /
skills lists, which the kernel interprets as "no filter" — every
globally-configured MCP server's tools and every installed skill get
injected into the prompt on every LLM call. On a typical instance (9
MCP servers, ~85 MCP tools + ~82 built-in tools) that's ~50k input
tokens per turn spent on definitions the agent never uses.
Changes
-------
32 agents/*/agent.toml:
- mcp_servers: 1-4 per agent. memory wherever state persists across
turns; fetch / exa-search / brave-search only where the prompt
actually calls for web; git / github / filesystem on engineering
agents; gmail / google-calendar / linear / jira on productivity
agents whose prompts mention them.
- skills: per-role allowlist driven by what the system_prompt names
(e.g. coder → rust/python/typescript/git/shell-scripting; devops-
lead → docker/kubernetes/terraform/ansible/ci-cd/helm/prometheus/
sysadmin). Generalists (assistant) keep skills = [] (see "Open
items" below).
- skills_disabled = true on the four short-conversational agents
(hello-world, recipe-assistant, health-tracker, home-automation).
Their system prompts never instruct the LLM to consult any skill,
so loading all 60 was pure waste. They also drop the explicit
max_history_messages override and inherit the kernel default (60).
- max_history_messages tiered by workload shape:
60 short conversational (hello-world, recipe, health-tracker,
home-automation) — inherits the rising kernel default
(`DEFAULT_MAX_HISTORY_MESSAGES = 60`); no override needed.
60 single-turn task agents (writer, translator, doc-writer,
email-assistant, customer-support, sales-assistant, recruit-
er, social-media, personal-finance, tutor, travel-planner,
meeting-assistant, ops, devops-lead, planner) — explicit
override at the same value to lock the cap if the kernel
default moves again.
80 multi-step / tool-heavy (coder, debugger, architect, code-
reviewer, test-engineer, security-auditor, analyst, data-
scientist, academic-researcher, researcher, legal-assistant)
120 coordinators (assistant, orchestrator) — long multi-agent
sessions where prompt-cache continuity is critical
All values sit at or above the kernel default. Pinning lower
would thrash the prompt cache (the failure mode #91 fixed for
the creator hand by *raising* the cap, not lowering it).
17 hands/*/HAND.toml:
- hand-level mcp_servers / skills now declared on every hand, so
every [agents.*] inside inherits a sensible allowlist.
- skills_disabled = true placed on each [agents.*] inside clip and
creator (pure media pipelines that don't benefit from any skill).
HandDefinitionRaw in librefang-hands does NOT have a top-level
skills_disabled field — declaring it at the hand top level would
be silently dropped by serde, so the setting must live on the
AgentManifest of each sub-agent role.
- devteam: expand existing mcp_servers = ["github"] to include
memory / git / filesystem; populate skills with the expected
dev-team expertise (replacing the placeholder skills = []).
- wiki: replace placeholder mcp_servers = [] with [memory, fetch,
filesystem]. Hand-level skills stays [].
- lead: hand-level skills was originally [email-writer, writing-
coach, interview-prep]; interview-prep is for job-interview
preparation, not lead generation. Replaced with data-analyst
(used by the qualification-scoring step in the prompt).
schema.toml: register mcp_servers / skills / max_history_messages on
the agent field schema so machine consumers (RegistrySchema in
librefang-types) see the new top-level fields. The
max_history_messages description now points at
librefang_runtime::agent_loop::DEFAULT_MAX_HISTORY_MESSAGES (60
today) by name, so the schema doesn't go stale when the constant
moves again.
agents/README.md: example block + "Adding a New Agent" checklist
mention the allowlists; max_history_messages example is shown
commented out with a prompt-cache caveat.
Open items
----------
`assistant` (the default user-facing agent) keeps `skills = []`
deliberately. It is the generalist entry point — capping its skill
surface at a small allowlist would defeat its "delegate to any
specialist" job. The trade-off is that this single agent still pays
the full skill-definition load on every turn; operators who want a
strict allowlist for `assistant` can override it after install.
Why not adopt PR #89's approach
-------------------------------
#89 covers similar ground but with three issues this PR avoids:
1. mcp_servers = ["_none"] sentinel. #89's body explicitly notes
it's pending upstream librefang#4808 (mcp_disabled). Shipping a
magic-string today means coming back later to clean it up. This
PR uses real allowlists.
2. max_history_messages = 8 / 12 / 15 / 20. Far below today's
kernel default (60) and #91's direction for long-workflow hands
(80–120). Every turn that hits the cap invalidates the cached
prompt prefix; the cost of cache misses exceeds the saving from
shorter history. This PR uses 60–120.
3. Doubling max_llm_tokens_per_hour (coder 200k→500k, assistant
300k→500k) widens the per-agent budget — the opposite direction
from #87's "reduce per-call cost" goal. Left to the operator's
instance-specific tuning.
Refs librefang/librefang-registry#87, librefang/librefang-registry#89
452 lines
14 KiB
TOML
452 lines
14 KiB
TOML
id = "creator"
|
|
version = "1.0.0"
|
|
name = "Creator Hand"
|
|
description = "AI media studio — generates images, videos, music, and speech from text prompts"
|
|
|
|
category = "content"
|
|
icon = "lucide:palette"
|
|
tools = [
|
|
"image_generate",
|
|
"video_generate",
|
|
"video_status",
|
|
"music_generate",
|
|
"text_to_speech",
|
|
"file_read",
|
|
"file_write",
|
|
"file_list",
|
|
"web_fetch",
|
|
"web_search",
|
|
"memory_store",
|
|
"memory_recall",
|
|
]
|
|
|
|
# Per-hand resource allowlists (refs librefang/librefang-registry#87).
|
|
# Inherited by every [agents.*] in this hand unless overridden.
|
|
mcp_servers = ["memory", "filesystem"]
|
|
skills = []
|
|
# `skills_disabled = true` lives on each [agents.*] below — HandDefinitionRaw
|
|
# in librefang-hands doesn't carry the field, so a hand-level entry would be
|
|
# silently dropped by serde.
|
|
|
|
|
|
[routing]
|
|
aliases = [
|
|
"generate image",
|
|
"create image",
|
|
"make a picture",
|
|
"generate video",
|
|
"create video",
|
|
"make a video",
|
|
"generate music",
|
|
"create music",
|
|
"compose music",
|
|
"text to speech",
|
|
"generate speech",
|
|
"voice over",
|
|
"media generation",
|
|
]
|
|
weak_aliases = [
|
|
"illustration",
|
|
"artwork",
|
|
"album cover",
|
|
"background music",
|
|
"jingle",
|
|
"narration",
|
|
"audio",
|
|
"thumbnail",
|
|
"poster",
|
|
"banner",
|
|
]
|
|
|
|
# ---- Requirements ----------------------------------------------------------------
|
|
|
|
[[requires]]
|
|
key = "media_provider"
|
|
label = "At least one media provider API key must be set"
|
|
requirement_type = "any_env_var"
|
|
check_value = "OPENAI_API_KEY,MINIMAX_API_KEY"
|
|
description = "Creator Hand needs at least one configured media provider. OpenAI supports image and TTS; MiniMax supports image, TTS, video, and music."
|
|
optional = false
|
|
|
|
# ---- Settings --------------------------------------------------------------------
|
|
|
|
[[settings]]
|
|
key = "default_provider"
|
|
label = "Preferred Provider"
|
|
description = "Which provider to use by default. Auto will pick the first configured provider for each capability."
|
|
setting_type = "select"
|
|
default = "auto"
|
|
|
|
[[settings.options]]
|
|
value = "auto"
|
|
label = "Auto-detect (best available)"
|
|
|
|
[[settings.options]]
|
|
value = "openai"
|
|
label = "OpenAI (image, TTS)"
|
|
provider_env = "OPENAI_API_KEY"
|
|
|
|
[[settings.options]]
|
|
value = "minimax"
|
|
label = "MiniMax (image, TTS, video, music)"
|
|
provider_env = "MINIMAX_API_KEY"
|
|
|
|
[[settings]]
|
|
key = "image_model"
|
|
label = "Image Model"
|
|
description = "Model to use for image generation"
|
|
setting_type = "select"
|
|
default = "auto"
|
|
|
|
[[settings.options]]
|
|
value = "auto"
|
|
label = "Provider default"
|
|
|
|
[[settings.options]]
|
|
value = "gpt-image-1"
|
|
label = "GPT Image 1 (OpenAI)"
|
|
|
|
[[settings.options]]
|
|
value = "dall-e-3"
|
|
label = "DALL-E 3 (OpenAI)"
|
|
|
|
[[settings.options]]
|
|
value = "image-01"
|
|
label = "Image-01 (MiniMax)"
|
|
|
|
[[settings]]
|
|
key = "image_size"
|
|
label = "Default Image Size"
|
|
description = "Default resolution for generated images"
|
|
setting_type = "select"
|
|
default = "1024x1024"
|
|
|
|
[[settings.options]]
|
|
value = "1024x1024"
|
|
label = "1024x1024 (square)"
|
|
|
|
[[settings.options]]
|
|
value = "1792x1024"
|
|
label = "1792x1024 (landscape)"
|
|
|
|
[[settings.options]]
|
|
value = "1024x1792"
|
|
label = "1024x1792 (portrait)"
|
|
|
|
[[settings]]
|
|
key = "tts_voice"
|
|
label = "TTS Voice"
|
|
description = "Default voice for text-to-speech"
|
|
setting_type = "select"
|
|
default = "alloy"
|
|
|
|
[[settings.options]]
|
|
value = "alloy"
|
|
label = "Alloy (neutral)"
|
|
|
|
[[settings.options]]
|
|
value = "echo"
|
|
label = "Echo (male)"
|
|
|
|
[[settings.options]]
|
|
value = "fable"
|
|
label = "Fable (storytelling)"
|
|
|
|
[[settings.options]]
|
|
value = "nova"
|
|
label = "Nova (female)"
|
|
|
|
[[settings.options]]
|
|
value = "onyx"
|
|
label = "Onyx (deep male)"
|
|
|
|
[[settings.options]]
|
|
value = "shimmer"
|
|
label = "Shimmer (warm female)"
|
|
|
|
[[settings]]
|
|
key = "minimax_api_key"
|
|
label = "MiniMax API Key"
|
|
description = "API key from platform.minimaxi.com for video and music generation"
|
|
setting_type = "text"
|
|
env_var = "MINIMAX_API_KEY"
|
|
default = ""
|
|
|
|
# ---- Agent configuration ---------------------------------------------------------
|
|
|
|
[agents.main]
|
|
coordinator = true
|
|
name = "creator-hand"
|
|
description = "AI media studio — generates images, videos, music, and speech from natural language"
|
|
module = "builtin:chat"
|
|
provider = "default"
|
|
model = "default"
|
|
skills_disabled = true
|
|
max_tokens = 8192
|
|
temperature = 0.5
|
|
max_iterations = 30
|
|
# Raise the history cap above the kernel default. Creator Hand's
|
|
# async video_generate flow polls video_status every 15-20s
|
|
# until completion (1-3 min typical), so a single video request
|
|
# consumes 5-15 turns. Combined workflows (video + TTS + music) plus
|
|
# normal back-and-forth easily exceed 60 messages, which triggered
|
|
# every-turn history trim and 0% prompt-cache hit-ratio in production.
|
|
# 80 covers ~30 polls plus a comfortable pre-context window.
|
|
max_history_messages = 80
|
|
system_prompt = """You are Creator Hand — an AI media studio that generates images, videos, music, and speech from natural language requests.
|
|
|
|
## Available Tools
|
|
|
|
You have access to these media generation tools:
|
|
|
|
### image_generate
|
|
Generate images from text prompts.
|
|
- Parameters: `prompt` (required), `provider`, `model`, `width`, `height`, `count`, `quality`, `seed`
|
|
- Returns: URLs of generated images (served at /api/uploads/...)
|
|
- Providers: OpenAI (gpt-image-1, dall-e-3), MiniMax (image-01)
|
|
|
|
### text_to_speech
|
|
Convert text to spoken audio.
|
|
- Parameters: `text` (required), `provider`, `model`, `voice`, `speed`, `format`
|
|
- Returns: URL to the audio file
|
|
- Providers: OpenAI (tts-1, tts-1-hd), MiniMax (speech-2.8-hd)
|
|
|
|
### video_generate
|
|
Submit a video generation task (asynchronous).
|
|
- Parameters: `prompt` (required), `provider`, `model`, `duration_secs`, `resolution`
|
|
- Returns: `task_id` and `provider` — use video_status to poll for completion
|
|
- Providers: MiniMax (T2V-01, video-01)
|
|
|
|
### video_status
|
|
Check the status of a video generation task.
|
|
- Parameters: `task_id` (required), `provider` (required)
|
|
- Returns: status ("pending", "processing", "completed", "failed") and result URL when done
|
|
|
|
### music_generate
|
|
Generate music from a text prompt and/or lyrics.
|
|
- Parameters: `prompt`, `lyrics`, `provider`, `model`, `instrumental` (bool), `format`
|
|
- At least one of `prompt` or `lyrics` is required
|
|
- Returns: URL to the audio file
|
|
- Providers: MiniMax (music-2.5)
|
|
|
|
## Workflow Guidelines
|
|
|
|
1. **Clarify intent**: If the user's request is vague, ask what type of media they want and suggest options.
|
|
|
|
2. **Image generation**:
|
|
- Be detailed in prompts — describe style, mood, composition, lighting
|
|
- For multi-image requests, vary the prompts meaningfully
|
|
- Default to 1024x1024 unless the user specifies otherwise
|
|
|
|
3. **Video generation** (async):
|
|
- Always explain that video takes time (typically 1-3 minutes)
|
|
- After calling video_generate, immediately poll with video_status
|
|
- If status is "processing", wait 15-20 seconds and poll again
|
|
- Keep the user informed of progress
|
|
|
|
4. **Music generation**:
|
|
- For instrumental, set `instrumental: true`
|
|
- For songs with vocals, provide both `prompt` (style description) and `lyrics`
|
|
- Suggest genres and moods if the user doesn't specify
|
|
|
|
5. **TTS**:
|
|
- Choose a voice that matches the content tone
|
|
- For long text, break into paragraphs and generate separately if needed
|
|
|
|
6. **Combined workflows** — these are where you shine:
|
|
- "Make a podcast intro" → generate music + TTS narration
|
|
- "Create a social media post" → generate image + caption
|
|
- "Make a video with narration" → generate video + TTS voice-over
|
|
- Always present results together with all URLs
|
|
|
|
## User Configuration
|
|
|
|
Check the User Configuration section for:
|
|
- `default_provider` — use this provider unless the user overrides
|
|
- `image_model` — preferred image model (use if not "auto")
|
|
- `image_size` — default image dimensions
|
|
- `tts_voice` — default TTS voice
|
|
|
|
Apply these defaults but allow the user to override in any request.
|
|
|
|
## Important Rules
|
|
|
|
- Always show the result URLs to the user so they can access the generated media
|
|
- For video tasks, ALWAYS poll until completion or failure — don't leave the user hanging
|
|
- Track generation stats via memory_store:
|
|
- `creator_hand_images_generated` — count
|
|
- `creator_hand_videos_generated` — count
|
|
- `creator_hand_music_generated` — count
|
|
- `creator_hand_tts_generated` — count
|
|
- If a provider is not configured, suggest the user set up the API key
|
|
- Never fabricate URLs or results — only return actual tool output
|
|
"""
|
|
|
|
[agents.prompt_writer]
|
|
invoke_hint = "Creative prompt engineering — writing detailed, effective prompts for image, video, and music generation"
|
|
name = "prompt-writer"
|
|
description = "Prompt engineer that crafts detailed, effective prompts for media generation"
|
|
module = "builtin:chat"
|
|
provider = "default"
|
|
model = "default"
|
|
skills_disabled = true
|
|
max_tokens = 4096
|
|
temperature = 0.8
|
|
system_prompt = """You are Prompt Writer, a creative prompt engineer within the Creator Hand.
|
|
|
|
Your job is to transform simple user requests into detailed, effective prompts for media generation models.
|
|
|
|
IMAGE PROMPTS:
|
|
- Specify art style (photorealistic, watercolor, digital art, anime, oil painting, etc.)
|
|
- Include composition details (wide shot, close-up, bird's eye view, etc.)
|
|
- Describe lighting (golden hour, studio lighting, dramatic shadows, etc.)
|
|
- Add mood/atmosphere (serene, chaotic, mysterious, vibrant, etc.)
|
|
- Include technical details when relevant (depth of field, lens type, etc.)
|
|
|
|
VIDEO PROMPTS:
|
|
- Describe the scene and action clearly
|
|
- Specify camera movement if desired (pan, zoom, dolly, tracking shot)
|
|
- Keep descriptions concise but vivid — video models work best with clear, focused prompts
|
|
|
|
MUSIC PROMPTS:
|
|
- Specify genre, tempo (BPM), and mood
|
|
- Describe instrumentation (piano, synth, acoustic guitar, etc.)
|
|
- Include structure hints (intro, verse, chorus, bridge)
|
|
- For songs with vocals, write lyrics with clear verse/chorus structure
|
|
|
|
Always present multiple prompt variations for the user to choose from."""
|
|
|
|
# ---- Dashboard metrics -----------------------------------------------------------
|
|
|
|
[dashboard]
|
|
[[dashboard.metrics]]
|
|
label = "Images Generated"
|
|
memory_key = "creator_hand_images_generated"
|
|
format = "number"
|
|
|
|
[[dashboard.metrics]]
|
|
label = "Videos Generated"
|
|
memory_key = "creator_hand_videos_generated"
|
|
format = "number"
|
|
|
|
[[dashboard.metrics]]
|
|
label = "Music Tracks"
|
|
memory_key = "creator_hand_music_generated"
|
|
format = "number"
|
|
|
|
[[dashboard.metrics]]
|
|
label = "TTS Audio"
|
|
memory_key = "creator_hand_tts_generated"
|
|
format = "number"
|
|
|
|
# ---- Metadata --------------------------------------------------------------------
|
|
|
|
[metadata]
|
|
frequency = "on-demand"
|
|
token_consumption = "low"
|
|
default_active = false
|
|
|
|
# ---- Internationalization --------------------------------------------------------
|
|
|
|
[i18n.zh]
|
|
name = "创作 Hand"
|
|
description = "AI 媒体工作室——根据文本提示生成图片、视频、音乐和语音"
|
|
category = "内容"
|
|
|
|
[i18n.zh.agents.main]
|
|
name = "创作工作室"
|
|
description = "AI 媒体工作室——通过自然语言生成图片、视频、音乐和语音"
|
|
|
|
[i18n.zh.agents.prompt_writer]
|
|
name = "提示词工程师"
|
|
description = "提示词工程师，为媒体生成模型编写详细、高效的提示词。"
|
|
|
|
[i18n.zh.settings.default_provider]
|
|
label = "首选提供商"
|
|
description = "默认使用哪个提供商。自动模式会为每种能力选择首个可用的提供商。"
|
|
|
|
[i18n.zh.settings.image_model]
|
|
label = "图像模型"
|
|
description = "用于图像生成的模型"
|
|
|
|
[i18n.zh.settings.image_size]
|
|
label = "默认图像尺寸"
|
|
description = "生成图像的默认分辨率"
|
|
|
|
[i18n.zh.settings.tts_voice]
|
|
label = "TTS 语音"
|
|
description = "文字转语音的默认声音"
|
|
|
|
[i18n.zh.settings.minimax_api_key]
|
|
label = "MiniMax API 密钥"
|
|
description = "来自 platform.minimaxi.com 的 API 密钥，用于视频和音乐生成"
|
|
|
|
[i18n.zh-TW]
|
|
name = "創作者 Hand"
|
|
description = "AI 媒體工作室——根據文字提示生成圖片、影片、音樂和語音"
|
|
|
|
[i18n.ja]
|
|
name = "クリエイター Hand"
|
|
description = "AIメディアスタジオ——テキストから画像、動画、音楽、音声を生成"
|
|
category = "コンテンツ"
|
|
|
|
[i18n.ja.settings.default_provider]
|
|
label = "優先プロバイダー"
|
|
description = "デフォルトで使用するプロバイダー。自動は各機能で最初に設定されたプロバイダーを選択。"
|
|
|
|
[i18n.ja.settings.image_model]
|
|
label = "画像モデル"
|
|
description = "画像生成に使用するモデル"
|
|
|
|
[i18n.ja.settings.image_size]
|
|
label = "デフォルト画像サイズ"
|
|
description = "生成画像のデフォルト解像度"
|
|
|
|
[i18n.ja.settings.tts_voice]
|
|
label = "TTS音声"
|
|
description = "テキスト読み上げのデフォルト音声"
|
|
|
|
[i18n.ja.settings.minimax_api_key]
|
|
label = "MiniMax APIキー"
|
|
description = "platform.minimaxi.comからのAPIキー。動画と音楽生成に使用。"
|
|
|
|
[i18n.ko]
|
|
name = "크리에이터 Hand"
|
|
description = "AI 미디어 스튜디오 — 텍스트 프롬프트로 이미지, 비디오, 음악, 음성 생성"
|
|
category = "콘텐츠"
|
|
|
|
[i18n.ko.settings.default_provider]
|
|
label = "기본 공급자"
|
|
description = "기본으로 사용할 공급자. 자동은 각 기능에 대해 첫 번째 구성된 공급자를 선택."
|
|
|
|
[i18n.ko.settings.image_model]
|
|
label = "이미지 모델"
|
|
description = "이미지 생성에 사용할 모델"
|
|
|
|
[i18n.ko.settings.image_size]
|
|
label = "기본 이미지 크기"
|
|
description = "생성 이미지의 기본 해상도"
|
|
|
|
[i18n.ko.settings.tts_voice]
|
|
label = "TTS 음성"
|
|
description = "텍스트-음성 변환의 기본 음성"
|
|
|
|
[i18n.ko.settings.minimax_api_key]
|
|
label = "MiniMax API 키"
|
|
description = "platform.minimaxi.com에서 발급한 API 키. 동영상 및 음악 생성에 사용."
|
|
|
|
[i18n.es]
|
|
name = "Hand Creador"
|
|
description = "Estudio de medios IA — genera imágenes, videos, música y voz a partir de texto"
|
|
category = "Contenido"
|
|
|
|
[i18n.fr]
|
|
name = "Hand Créateur"
|
|
description = "Studio média IA — génère images, vidéos, musique et voix à partir de texte"
|
|
category = "Contenu"
|
|
|
|
[i18n.de]
|
|
name = "Kreator-Hand"
|
|
description = "AI-Medienstudio — erzeugt Bilder, Videos, Musik und Sprache aus Textprompts"
|
|
category = "Inhalt"
|