메뉴
Access Sapiens AI's multimodal models (Agnes AI platform) via OpenAI-compatible API. Covers agnes-2.0-flash (text), agnes-image-2.0-flash (image generation/editing), agnes-video-v2.0 (video generation), and agnes-image-2.1-flash.
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/QianJinGuo/wiki --skill agnes-ai명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
SKILL.md 표시 중
| name | agnes-ai |
| description | Access Sapiens AI's multimodal models (Agnes AI platform) via OpenAI-compatible API. Covers agnes-2.0-flash (text), agnes-image-2.0-flash (image generation/editing), agnes-video-v2.0 (video generation), and agnes-image-2.1-flash. |
| version | 1.0.0 |
| author | ["hermes"] |
| license | MIT |
| platforms | ["macos","linux","windows"] |
| compatibility | Requires AGNES_IMAGE_API_KEY env var set to your API key from https://agnes-ai.com. This single key works for ALL model types (text, image, video). On macOS the key is typically exported from ~/.zshenv — it is NOT in ~/.hermes/.env. |
| prerequisites | {"commands":["curl"]} |
| metadata | {"hermes":{"tags":["image-generation","video-generation","agnes","openai-compat","creative","generative-ai","multimodal"],"category":"creative"}} |
Unified access to Sapiens AI's multimodal models via the Agnes AI platform. All models share a single API key.
| Model | Type | Model ID | Endpoint |
|---|---|---|---|
| Agnes 2.0 Flash | Text | agnes-2.0-flash | POST /v1/chat/completions |
| Agnes Image 2.0 Flash | Image | agnes-image-2.0-flash | POST /v1/images/generations |
| Agnes Image 2.1 Flash | Image | agnes-image-2.1-flash | POST /v1/images/generations |
| Agnes Video V2.0 | Video | agnes-video-v2.0 | POST /v1/videos |
https://apihub.agnes-ai.com/v1Authorization: Bearer $AGNES_IMAGE_API_KEYexport AGNES_IMAGE_API_KEY="sk-YourApiKeyHere"
One key for everything — a single API key authenticates all model types (text, image, video). No separate key needed per modality.
See references/ for session-specific detail from skill creation/debugging:
style param not supported
→ Reference: references/api-probing-results.mdextra_body pattern
→ Reference: references/api-probing-results.mdreferences/live-probe-2026-06-02.mdreferences/hand-drawn-marker-style.mdcurl -X POST https://apihub.agnes-ai.com/v1/images/generations \
-H "Authorization: Bearer $AGNES_IMAGE_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"agnes-image-2.0-flash","prompt":"a cat","size":"1024x1024"}'
# → Returns immediately with URL in data[].url
# Step 1: Submit
TASK_ID=$(curl -s -X POST https://apihub.agnes-ai.com/v1/videos \
-H "Authorization: Bearer $AGNES_IMAGE_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"agnes-video-v2.0","prompt":"a cat walking"}' \
| python3 -c "import sys,json; print(json.load(sys.stdin)['task_id'])")
# Step 2: Poll until completed
curl -s "https://apihub.agnes-ai.com/v1/videos/${TASK_ID}" \
-H "Authorization: Bearer $AGNES_IMAGE_API_KEY"
# → Completed response: remixed_from_video_id = MP4 URL
style param not supported on images — Causes 400 UnsupportedParamsError. Do not send it.n parameter ignored on images — Even n: 2 returns only 1 image.b64_json always null — Both image and video APIs only return URLs, never base64.queued task_id, must poll GET /videos/{id}.remixed_from_video_id — Not data[] like images. This field holds the MP4 URL.extra_body — Image arrays go inside extra_body.image, not at top level. Keyframes mode requires ≥2 images.style parameter name collision — Different from OpenAI's style param which this API rejects. Confusing but important.width+height.num_frames/frame_rate.agnes-image-2.0-flash not agnes-image-2. agnes-video-v2.0 not agnes-video-v2.GET /v1/models/agnes-image-2.0-flash returns "model does not exist". Use GET /v1/models to list.revised_prompt always null — No prompt auto-refinement on images or video.POST /v1/chat/completions returns 500 division by zero and POST /v1/videos returns 500 fail_to_fetch_task. These are upstream backend bugs, not auth/credential problems. Image endpoints were stable as of 2026-06-02 but are now ALSO intermittent (see #19).POST /v1/images/generations call succeeds, but ALL subsequent calls return 无效的令牌 (invalid token). Applies to both agnes-image-2.0-flash and agnes-image-2.1-flash. GET /v1/models auth probe still works. Interpretation: token-based rate limiting or one-shot auth — token temporarily invalidated after one use. Impact: NOT viable for batch image generation (e.g. OpenMAIC classrooms). Treat ALL endpoints as unreliable. See references/live-probe-2026-06-18.md. ("Authentication Error, Client is not connected to the query engine"). These are upstream backend bugs, not auth/credential problems. Image endpoints (/v1/images/generations) are stable. If chat/video is critical, retry once and fall back to image-2.1 for text-adjacent work, or treat the skill as image-only until upstream recovers.GET /v1/models lists 5 models — agnes-1.5-flash, agnes-2.0-flash, agnes-image-2.0-flash, agnes-image-2.1-flash, agnes-video-v2.0. The agnes-1.5-flash text model is undocumented by the platform but is a valid model ID.agnes-image-2.0-flash: a "system architecture diagram" prompt with labeled nodes produced structurally correct composition but garbled text labels (Atlsse and Hedrorct for "AWS and Bedrock", entiriss/concepts/comparisons for entities/concepts/comparisons, agent-nemss-architecture, imlf-systems, self-provemag). Short words ("mcp", "rag", "vibe-coding") and brand-style strings ("harness-engineering", "Multi-Agent") often survive. Use cases that need correct text (labeled diagrams, infographics, flow charts with readable node names, anything where labels carry the message) → use a structured diagram tool instead: Mermaid (markdown-native, edits inline), Excalidraw (hand-drawn JSON), or hand-rendered SVG/ASCII. Agnes / diffusion image models are only safe for: hero/cover images, mood imagery, decorative illustrations, and structural compositions where label legibility is non-critical. The right-tool-for-right-job pattern: generate a visual mood/cover image with Agnes for emotional appeal, then ship a Mermaid/Excalidraw/SVG version for the labeled/structural content that the user will actually read. Do not promise a "diagram" from a diffusion model and call it done.references/hand-drawn-marker-style.md for the prompt recipe + dual-track pattern. Do NOT improvise — the reference has the magic style words, palette cheat sheet, and a known-good template prompt. Without the reference, you'll get generic "clean infographic" output and waste 2-3 regenerations.references/hand-drawn-marker-style.md § "dual-track pattern (USER-VALIDATED)" for the full pattern.source ~/.zshenv # key lives here on macOS, not in ~/.hermes/.env
KEY=${AGNES_IMAGE_API_KEY:?must be set}
# 1. Confirm base URL reachable + auth works
curl -s -w "\nHTTP %{http_code}\n" "https://apihub.agnes-ai.com/v1/models" \
-H "Authorization: Bearer $KEY" | head -20
# Expect: HTTP 200, JSON `data` array with 5 model IDs
# 2. Confirm image endpoint is healthy (the stable path)
curl -s -X POST "https://apihub.agnes-ai.com/v1/images/generations" \
-H "Authorization: Bearer $KEY" -H "Content-Type: application/json" \
-d '{"model":"agnes-image-2.0-flash","prompt":"a tiny red apple","size":"512x512"}' \
| python3 -c "import sys,json; d=json.load(sys.stdin); print('OK' if d.get('data',[{}])[0].get('url') else d)"
# Expect: OK (or HTTP 200 with a GCS URL)
# 3. Probe chat endpoint — currently 500 upstream
curl -s --max-time 10 -X POST "https://apihub.agnes-ai.com/v1/chat/completions" \
-H "Authorization: Bearer $KEY" -H "Content-Type: application/json" \
-d '{"model":"agnes-2.0-flash","messages":[{"role":"user","content":"hi"}],"max_tokens":5}'
# If HTTP 500 "division by zero" → text path down, use image-only
If the image probe fails: stop and treat the key as broken — do not retry indefinitely. The platform has no SLA on these endpoints.
| Feature | Image 2.0 | Video V2.0 |
|---|---|---|
| Sync/Async | Sync | Async (poll) |
| Response field | data[].url | remixed_from_video_id |
| Input image | input_image (string) or image_urls (array) | image (string) or extra_body.image (array) |
| Multi-image | image_urls: [...] | extra_body: {image: [...]} |
| Style param | ❌ rejected | N/A |
| Size | 1024x1024 default | 1280x768 default |
| Negative prompt | ✅ | ✅ |
| Seed | ✅ | ✅ |
精选高价值 RSS feeds 扫描，输出到 raw/rss-inbox/ 暂存区。只保留有独立知识深度的 feed（非 digest 类），不再自动入库。包含 rss-inbox-curl-recovery.py 绕过 blogwatcher read-state 漏抓的兜底。
Meta-orchestrator that wires web-content-reviewer, llm-wiki, and wiki-evolver into a single four-phase knowledge pipeline (Triage → Gate → Store → Evolve). Use as the single entry point for all knowledge base operations. Includes a 6-URL-validated user-pasted WeChat URL fast path, three-axis dedup decision matrix (NEW/MERGE/DEDUP), orphan-raw detection protocol, and sibling-subagent race V7 evidence, and a two-variant V6 mid-write fix for matching vs different-slug duplicates.
从 Gmail 中提取 TLDR AI 等 newsletter 的链接，写入 raw/email-inbox/candidates.md，供后续 inbox-screener 评分。只做链接不做评分，与 inbox-screener 配合使用。
SOC 직업 분류 기준