content

name

content

description

Complete content management via Korea SNS (withcenter.com) REST API. API key based authentication (Authorization: Bearer). Supports user registration/login/profile updates, post CRUD, comment CRUD, file (image) uploads, site/category lookup and management, likes/bookmarks/reactions, notifications, search, and sub-site admin banner ads (header/sidebar/forum banners with begin_at/end_at scheduling). A complete API guide for external programs, AI, and software to create, update, and delete content on Korea SNS. Use when Claude performs the following tasks: (1) Korea SNS registration/login/API key retrieval (2) User profile update/avatar upload (3) Post create/update/delete/list (4) Comment create/update/delete (5) Upload images and files and attach them to posts/comments (6) Site list/category tree lookup (7) Toggle like/bookmark/reaction (8) Notifications lookup/search (9) withcenter.com API calls (10) Automatic API documentation lookup (GET /docs) (11) Banner ads CRUD on a sub-site (header, sidebar, forum banners — requires sub-site admin API key). Keywords: Korea SNS, withcenter, post, posts, writing, post registration, post update, post deletion, comment, comments, registration, login, profile, avatar, file upload, image, category, site, like, bookmark, reaction, notification, search, API, api_key, banner, banner ad, banners, header banner, sidebar banner, forum banner, banner position, banner schedule, begin_at, end_at, ad scheduling

Korea SNS — Complete Content Management Skill

A skill that performs all content tasks such as user management, post/comment CRUD, file upload, and site/category management through the Korea SNS (withcenter.com) REST API.

🔴 Real-World Content Research — Delegate to content-research-writer

Whenever the task requires producing, gathering, or verifying real-world information — articles, news, blog posts, tutorials, case studies, thought-leadership pieces, citations, source hunting, hook writing, outlines, section-by-section drafting/feedback, or anything that needs actual facts, quotes, data, or external sources — delegate the research/writing portion to the content-research-writer skill reference located at:

• references/content-research-writer.md

Triggers for delegation (not exhaustive): "write an article about…", "find news on…", "research X and summarize…", "add citations…", "draft a blog post…", "produce content about…", "get real examples of…", "summarize recent data on…".

How to delegate:

1. Read references/content-research-writer.md and follow its workflow (outline → research → draft → citations → section feedback → polish) for anything that involves producing external/real-world content.
2. Once the draft is produced, run it through the humanizer reference below before publishing — this is a mandatory polish pass, not optional.
3. After humanization, return to this skill's API workflow (post create/update, file upload, etc.) to publish it via https://withcenter.com/api/v1.
4. Keep the writer's voice rules, citation format, and research-verification guidance from the reference intact — do not shortcut them.

🔴 AI-Pattern Removal — Delegate to humanizer

Before publishing any draft produced by content-research-writer (or any other source), always run the text through the humanizer reference to strip AI writing tells — inflated symbolism, promotional language, superficial -ing analyses, vague attributions, em-dash overuse, rule of three, AI vocabulary words, passive voice, negative parallelisms, filler phrases, curly quotes, emoji decorations, etc.

• references/humanizer.md

When to run humanizer (mandatory):

• After content-research-writer finishes a draft and before the content hits the Korea SNS API.
• When the user pastes existing copy and asks to publish, edit, or rewrite it.
• When the user explicitly says "humanize", "make this sound human", "remove AI tone", or similar.

How to delegate:

1. Read references/humanizer.md and follow its full process: identify AI patterns → rewrite → final anti-AI audit pass ("What makes the below so obviously AI generated?" → revise).
2. If the user supplied a writing sample, perform the voice calibration step described in the reference before rewriting.
3. Only after the humanized final version is approved should you publish via this skill's API workflow.

In short: content-research-writer drafts → humanizer polishes → this content skill publishes.

API Basics

Item	Value
Base URL	https://withcenter.com/api/v1
Authentication	Authorization: Bearer {API_KEY} header
API key format	{user_id}-{md5_hash} (e.g., 4-a1b2c3d4e5f6...)
Request format	JSON: Content-Type: application/json, files: multipart/form-data
Success response	Single: { "data": {...} }, list: { "data": [...], "meta": {...} }
Error response	{ "message": "error message" }
User-Agent	User-Agent: KoreaSNS-CLI/1.0 is required to bypass Cloudflare WAF blocking

Three ways to pass the API key (in priority order):

1. Authorization header (recommended): Authorization: Bearer {API_KEY}
2. api_key cookie: Cookie: api_key={API_KEY}
3. Query parameter: ?api_key={API_KEY}

Automatic API Documentation Lookup

When unsure which APIs are available, call the following endpoint to get the list of APIs:

# Fetch full API documentation
curl -s https://withcenter.com/api/v1/docs \
  -H "User-Agent: KoreaSNS-CLI/1.0"

# Filter by category (auth, user, post, comment, file, site, category, notification, search)
curl -s "https://withcenter.com/api/v1/docs?category=post" \
  -H "User-Agent: KoreaSNS-CLI/1.0"

Always check API documentation before starting work to identify available endpoints.

Workflow

Step 1: Obtain an API key

If the user provides an API key directly, use it. If there is no API key but an email/password is available, log in to obtain the API key. If there is no account, register first and automatically obtain the API key.

Important: When working on a subsite (e.g., apple.withcenter.com), always set --base-url to the subsite URL.

# Register (on subsite — registration is not allowed on the main site)
python3 skills/korea/scripts/korea_api.py --api-key "" \
  --base-url "https://apple.withcenter.com/api/v1" \
  register --email "[e-mail]" --password "pass123" --display-name "NewUser"

# Login
python3 skills/korea/scripts/korea_api.py --api-key "" \
  --base-url "https://apple.withcenter.com/api/v1" \
  login --email "[e-mail]" --password "pass"

Step 2: Check site/category

Check the target site and category before writing a post.

# List sites
python3 skills/korea/scripts/korea_api.py --api-key "{KEY}" sites

# Fetch category tree (site ID required)
python3 skills/korea/scripts/korea_api.py --api-key "{KEY}" categories --site-id 1

Step 2.5: AI Pre-flight Topic Workflow (Plan → Reserve → Generate → Extract claims → Red-team → Score → Submit)

When AI is generating content, do not waste tokens by drafting first and being rejected later, and do not pollute the site with low-quality or fabricated posts. Always run this 7-step handshake with the server. Steps 1–3 are cheap API calls. Step 4 is the expensive LLM/web-search work — it runs only after a successful reservation. Steps 5–7 are verification gates that must all pass before submission.

⚠️ Why 7 steps, not 5: Early versions of this workflow had 5 steps (plan → check → reserve → generate → score). In practice, the self-evaluation score passed posts that contained fabricated citations, invented phone numbers, hallucinated anchor-dates, and outdated emergency/legal references — because an LLM scoring its own output cannot detect its own hallucinations. Steps 5 (claim extraction) and 6 (independent red-team) close that loop. See references/hallucination-prevention.md for the specific failure modes this guards against.

1. Plan — Call topic-coverage to see which topic_slugs the current user has already used.

   python3 skills/korea/scripts/korea_api.py --api-key "{KEY}" --base-url "{BASE}" \
     topic-coverage [--category-id 3] [--per-page 100]
   

2. Check (optional) — Batch-check candidate slugs before reserving. Returns available / taken_by_post / reserved_active.

   python3 skills/korea/scripts/korea_api.py --api-key "{KEY}" --base-url "{BASE}" \
     topic-check --topic-slugs "ph-cebu-diving,ph-palawan-elnido,ph-bohol-tarsier"
   

3. Reserve — Atomically lock ONE slug for up to TTL minutes (default 30, range 1–120). If 409, pick the next candidate.

   python3 skills/korea/scripts/korea_api.py --api-key "{KEY}" --base-url "{BASE}" \
     topic-reserve --topic-slug "ph-cebu-diving" [--category-id 3] [--ttl-minutes 30]
   

4. Generate — ONLY after a successful reservation, do the expensive work:
   • Gather ≥ 20 distinct web sources. While researching, maintain a working claim ledger ({claim, source_url, source_quote, date_accessed}) — every specific fact you plan to use must get an entry as you discover it, not retroactively after drafting.
   • Download ≥ 3 images locally (curl with realistic User-Agent), then upload each to the server via upload, capture the returned id and server URL — never hotlink external images.
   • Draft the article in Markdown, targeting the length band for the topic class (see Length tiers below) — not a blanket 5,000-word floor. Use the full Markdown syntax (headings, bullet + numbered lists, table, bold + italic, blockquote, inline link, horizontal rule, inline code, image).
   • Sprinkle ≥ 15 emojis through the body; every ## heading gets a leading topic-matching emoji.
   • End the body with the mandatory 📍 방문 정보 metadata table containing country, region, city, full address, lat/lon (≥ 4 decimals), name, contact (price + hours optional).
   • 🔴 Anti-hallucination rules (absolute) — detailed catalog in references/hallucination-prevention.md:
     • Never invent citations. If you do not have a specific URL for a specific claim, omit the specific detail or hedge with a vague phrase ("여러 연구", "some reports") — never invent a journal name, author, or year to sound authoritative.
     • Never invent anchor numbers. Phrases like "1,300년 인류사에서", "연간 45,000명" must be sourced; otherwise replace with vague-but-true ("현대 해양 관찰 기록에서", "매년 수만 명").
     • Specifics (phones, addresses, lat/lon, law numbers, emergency numbers, embassy contacts) require a primary source — the owning entity's own website, the official statute text, the government hotline page. Secondary-source mentions are not sufficient for these categories.
     • When uncertain, drop the specific and keep the general. Readers forgive vagueness; they don't forgive wrong phone numbers, outdated emergency codes, or fabricated laws.
5. Extract claims — Produce a structured claims.json listing every testable factual claim from the draft (numbers, names, dates, phones, addresses, coordinates, laws, citations, absolute statements like "no recorded cases"). Each entry records source_url, confidence, and risk_class (safety / legal / financial / contact / descriptive). Any claim with source_url: null is either removed or replaced with a verified one. High-risk-class claims (safety/legal/financial/contact) may not ship with confidence < "high". Full spec in references/content-quality-score.md §10.
6. Red-team pass — Spawn an independent fact-check agent (fresh context, ideally a different model invocation) with the draft + claim ledger. The agent returns {definite_errors, likely_errors, plausible_unverified}. Any definite_errors.length > 0 → block submission, revise, repeat. This closes the self-grading loop that a single-model rubric cannot. Prompt template in references/hallucination-prevention.md.
7. Score & Submit — Apply the Content Quality Score rubric in references/content-quality-score.md. Submit only if the total score ≥ 90/100 AND every one of the 38 gates passes (G1–G38 — including the new anti-hallucination gates G33–G38). Otherwise revise (max 2 retries) or abandon the reservation.

   # only after claims.json passes, red-team returns zero definite errors, and rubric ≥ 90:
   python3 skills/korea/scripts/korea_api.py --api-key "{KEY}" --base-url "{BASE}" \
     create --title "..." --content "$BODY_MARKDOWN" --category-id 3 \
     --topic-slug "ph-cebu-yakimix-buffet" --reservation-id 889 \
     --upload-ids "124,125,126,127,128"
   

The 90/100 quality gate, 38 hard gates, AND clean red-team pass are all mandatory. A reserved-but-low-quality or fabricated post is worse than no post — it consumes the slug permanently for this user and damages site credibility. If three drafts (1 original + 2 revisions) cannot clear all three bars, do not submit; let the reservation expire and pick a different topic.

Length tiers — target range by topic class (replaces the old ≥ 5,000-word floor)

The old ≥ 9,000 Korean-char floor forced every topic — even small ones — into exhaustive-guide territory, producing bloated posts no one reads to the end. New policy: length follows topic depth, within a target range with a cap so posts don't become encyclopedias.

Topic class	Target Korean chars	Typical read time	Examples
Place / destination guide	4,500 – 7,000	8–12 min	Oslob whale shark, Boracay beaches, Palawan El Nido
How-to / visa / logistics	2,500 – 4,500	5–8 min	9G visa 연장, GCash 외국인 가입, SIM 비교
Restaurant / single venue review	1,500 – 3,000	3–5 min	세부 야키믹스 뷔페, BGC 한식당 리뷰
Cultural deep-dive / long-form essay	5,000 – 8,000	10–15 min	필리핀 가톨릭 문화, 필리피노 결혼식 관습

Both undershooting (shallow) and overshooting (bloated) fail the new G2. Rewards hitting the right size for the topic, not chasing maximum word count.

⏱️ TTL hint: Long-form posts (5,000+ words + image upload) often exceed the default 30-minute TTL. Use topic-reserve --ttl-minutes 90 (max 120) when reserving for full place guides.

Rules of the system:

• Uniqueness is enforced per user: user A and user B can each post the same topic once; user A cannot post the same topic_slug (or the same normalized title+content) twice.
• Posts without topic_slug are treated as normal human posts and are not subject to duplicate prevention or the score gate (e.g. a user re-listing the same sale post every day is allowed).
• content_hash is computed server-side (SHA-256 of NFC-normalized title+content) and is only stored/checked when topic_slug is present.
• Reservations expire automatically after TTL; the next topic-reserve call for the same slug cleans up expired rows before inserting a new one.
• 409 Conflict means the slug or the content hash is already in use for this user — catch it, pick a different topic, retry.

Only when topic-reserve returns 201 should the AI spend real tokens on web search and drafting. Only when the rubric returns total ≥ 90 should the AI call POST /posts.

Quality Score — quick summary

100 points across 6 quality categories, gated by 38 hard requirements (G1–G38). Full rubric, image/metadata workflow, claim-extraction spec, red-team prompt, and revision guidance: references/content-quality-score.md.

Category	Max	What it measures
Originality & Uniqueness	20	Genuinely new framing/analysis vs. rehashed search results
Depth & Substance	20	Concrete numbers, comparisons, trade-offs vs. surface definitions
Accuracy & Verifiability	15	Sourced, current facts (incl. coordinates, address, phone) vs. fabrication
Structure, Markdown & Readability	15	Sections + full Markdown syntax + short paragraphs + visual rhythm
Reader Value	20	Tailored to Korean expat audience, KRW conversions, Maps link, actionable
Polish	10	No typos, natural Korean, no machine-translation artifacts, valid Markdown

Pass requirement: total ≥ 90 AND all 38 gates pass AND the independent red-team pass returns definite_errors: []. The 38 gates cover:

Group	Gates	Summary
Content	G1–G7	≥ 20 sources · length in target band for topic class (see Length tiers) · ≥ 5 sections · slug match · slug available · coherent language · valid Markdown with full syntax
Metadata (REQUIRED)	G8–G14	🏳️ Country · 🗺️ Region · 🏙️ City · 📮 Address · 📌 Lat/Lon (≥ 4 decimals) · 🏢 Name · 📞 Contact
Images	G15–G19	≥ 3 images · download → upload → server URL (no hotlinking) · descriptive alt text · ≥ 1 photo of the place · --upload-ids matches body URLs
Markdown syntax	G20–G28	Headings ##/### · bullet and numbered list · table · bold + italic · blockquote · inline link · --- rule · inline code · Markdown image syntax
Emoji & readability	G29–G32	≥ 15 emojis spread out · every ## heading has a topic emoji · paragraphs ≤ 6 sentences (avg ≤ 4) · lists/tables/callouts every ~600 words
Anti-hallucination (NEW)	G33–G38	Every academic citation has a DOI or direct URL · every phone/emergency number links to the owning entity's official page · every law/regulation number links to official statute text · every "since Year" or "N-year history" claim cites a source · every absolute claim ("no recorded cases", "only operation in the world", "largest ever") has a source · no category-mismatch (safety/legal/financial/contact) claim with confidence < "high"

Optional metadata (do not gate, include if known): 💰 Pricing, ⏰ Hours, 🌐 Website.

On failure:

1. Fix gate failures first — they are absolute (e.g. G2 word count, G15 images, G12 coordinates).
2. Identify weak_areas (quality categories below the "Good" band) and revise only those — do not rewrite from scratch on round 1.
3. Re-score.
4. Max 2 revisions; if still failing, abandon the reservation (let it expire) and pick another topic.

The AI must keep an internal scorecard JSON like:

{ "word_count": 5421, "korean_char_count": 6200, "topic_class": "place_guide",
  "length_band_hit": true,
  "image_count": 5,
  "metadata": { "country":"Philippines","region":"Cebu Province","city":"Cebu City",
                "address":"123 Mango Ave, Lahug, Cebu City 6000",
                "lat":10.3157, "lon":123.8854,
                "name":"Sample Restaurant", "contact":"[phone]" },
  "gates_passed": ["G1","G2","...","G38"],
  "red_team": { "definite_errors":[], "likely_errors":[], "plausible_unverified":[] },
  "scores": { "originality":18, "depth":17, "accuracy":14, "structure":13, "reader_value":19, "polish":9 },
  "total": 90, "pass": true, "weak_areas": [], "revision_round": 0 }

Inflating scores or fabricating metadata defeats the purpose. Never invent addresses, coordinates, phone numbers, academic citations, law numbers, or emergency codes — abandon the topic instead.

Step 3: Execute the content task

Include --base-url "https://<domain>/api/v1" in every command. In the examples below, {BASE} is a subsite URL such as https://apple.withcenter.com/api/v1.

# Post CRUD
python3 skills/korea/scripts/korea_api.py --api-key "{KEY}" --base-url "{BASE}" create --title "Title" --content "Content" [--category-id 3]
python3 skills/korea/scripts/korea_api.py --api-key "{KEY}" --base-url "{BASE}" update --id {ID} --title "New title" --content "New content"
python3 skills/korea/scripts/korea_api.py --api-key "{KEY}" --base-url "{BASE}" delete --id {ID}
python3 skills/korea/scripts/korea_api.py --api-key "{KEY}" --base-url "{BASE}" get --id {ID}
python3 skills/korea/scripts/korea_api.py --api-key "{KEY}" --base-url "{BASE}" list [--page 1] [--per-page 10] [--category free]

# Comment CRUD
python3 skills/korea/scripts/korea_api.py --api-key "{KEY}" --base-url "{BASE}" comment-create --post-id {ID} --content "Comment"
python3 skills/korea/scripts/korea_api.py --api-key "{KEY}" --base-url "{BASE}" comment-update --comment-id {ID} --content "Updated"
python3 skills/korea/scripts/korea_api.py --api-key "{KEY}" --base-url "{BASE}" comment-delete --comment-id {ID}

# Upload a file and attach it to a post
python3 skills/korea/scripts/korea_api.py --api-key "{KEY}" --base-url "{BASE}" upload --file "/path/to/image.jpg"
# → Pass the returned upload ID through --upload-ids when creating a post/comment
python3 skills/korea/scripts/korea_api.py --api-key "{KEY}" --base-url "{BASE}" create --title "Photo post" --content "Content" --upload-ids "10,11"

# Test/seed images with no user-supplied file:
#   1. Default → picsum.photos (curl -L to a local temp file, then upload the file)
#        curl -sL "https://picsum.photos/400/300?random=$RANDOM" -o /tmp/img.jpg
#   2. Opt-in for "interesting/fun" imagery → WebSearch a royalty-free source
#      (Unsplash / Pexels / Wikimedia Commons), curl -L to a local file, then upload.
#   Always download locally first — the /files/upload endpoint only accepts multipart, never a remote URL.
#   Full workflow + size table: commands/create-banner.md Step 5.

# Profile update
python3 skills/korea/scripts/korea_api.py --api-key "{KEY}" --base-url "{BASE}" update-profile --display-name "New name" --bio "About me"

# Avatar upload
python3 skills/korea/scripts/korea_api.py --api-key "{KEY}" --base-url "{BASE}" upload-avatar --file "/path/to/avatar.jpg"

Using curl directly (alternative)

Always include the User-Agent header to avoid Cloudflare WAF blocking.

# Create a post
curl -s -X POST https://withcenter.com/api/v1/posts \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer {API_KEY}" \
  -H "User-Agent: KoreaSNS-CLI/1.0" \
  -d '{"title": "Title", "content": "Content", "category_id": 3}'

# Upload a file
curl -s -X POST https://withcenter.com/api/v1/files/upload \
  -H "Authorization: Bearer {API_KEY}" \
  -H "User-Agent: KoreaSNS-CLI/1.0" \
  -F "file=@/path/to/image.jpg"

# Attach the uploaded file to a post
curl -s -X POST https://withcenter.com/api/v1/posts \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer {API_KEY}" \
  -H "User-Agent: KoreaSNS-CLI/1.0" \
  -d '{"title": "Photo post", "content": "Content", "upload_ids": [10, 11]}'

Data / Information Collection

When the user's prompt requires collecting information, research, facts, news, references, or background material before creating/updating content, you MUST gather data from a minimum of 20 different sites (web pages) before writing.

Rules

1. Minimum 20 sources: Do not produce content until you have collected information from at least 20 distinct web pages. Fewer sources is not acceptable — expand the search until this threshold is met.
2. Primary tools: Use the WebSearch tool to discover candidate sources, and use the WebFetch tool to retrieve the full content of each page.
3. Fallback to curl: If WebFetch is unavailable, blocked, or fails, use curl via the Bash tool to fetch the page. Always include a realistic User-Agent header (e.g., -H "User-Agent: Mozilla/5.0 (compatible; KoreaSNS-CLI/1.0)").
4. Source diversity: Prefer different domains over multiple pages from the same site. Aim for a mix of official/primary sources, news outlets, community discussions, and reference material relevant to the user's topic.
5. Recency: Prioritize recent sources when the topic is time-sensitive (news, prices, events, releases, etc.).
6. De-duplicate: If two pages contain substantially the same content (syndicated articles, mirrors), count them as a single source and keep searching.
7. Record sources: Keep track of the URLs you consulted so they can be cited or referenced in the final post/comment when appropriate.
8. Synthesize, don't copy: Combine and paraphrase findings into original content. Do not paste copyrighted text verbatim.

Workflow

1. Parse the user's prompt → extract key topics / keywords / questions.
2. Run WebSearch with multiple query variations to build a candidate URL list (target ≥ 30 URLs so that after filtering you still have ≥ 20 usable pages).
3. For each URL, call WebFetch (or curl as fallback) and extract the relevant facts.
4. Continue until at least 20 distinct, useful pages have been successfully read.
5. Synthesize the collected information into the post/comment body.
6. Call the Korea SNS content APIs (create/update) with the finished content.

curl fallback example

curl -sL "https://example.com/article" \
  -H "User-Agent: Mozilla/5.0 (compatible; KoreaSNS-CLI/1.0)" \
  -H "Accept: text/html,application/xhtml+xml"

Notes

1. User-Agent required: Cloudflare WAF blocks requests without a User-Agent. When using curl, -H "User-Agent: KoreaSNS-CLI/1.0" is required.
2. API key security: Do not expose the API key in logs, files, or output.
3. Error handling: If the response contains a message field, it is an error. Relay it to the user.
4. Permissions: Updates/deletions are allowed only for the author or site administrators.
5. File upload order: Upload the file first (POST /files/upload) to receive an ID, then link it when creating the post/comment via the upload_ids array.
6. Multitenant: When working on a subsite, set the subsite URL with the --base-url "https://<domain>/api/v1" option. Registration/posting is not allowed on the main site (withcenter.com).
7. Check API docs: When unsure how to perform a task, call GET /docs to see the available APIs.

Quick reference for all API routes

GET    /docs                           — API documentation (JSON, filter with ?category=)

POST   /auth/register                  — Register
POST   /auth/login                     — Login
POST   /auth/logout                    — Logout

GET    /me                             — Get my info
PATCH  /me                             — Update my info
POST   /me/avatar                      — Upload avatar
POST   /me/cover                       — Upload cover image

GET    /posts                          — List posts
POST   /posts                          — Create post
GET    /posts/{id}                     — Get post details
PUT    /posts/{id}                     — Update post
DELETE /posts/{id}                     — Delete post

GET    /posts/{id}/comments            — List comments
POST   /posts/{id}/comments            — Create comment
PATCH  /comments/{id}                  — Update comment
DELETE /comments/{id}                  — Delete comment

POST   /files/upload                   — Upload file (multipart/form-data)
DELETE /files/{id}                     — Delete file

GET    /sites                          — List sites
GET    /sites/{id}/categories/tree     — Category tree

POST   /posts/{id}/like                — Toggle post like
POST   /comments/{id}/like             — Toggle comment like
POST   /posts/{id}/bookmark            — Toggle bookmark
POST   /posts/{id}/reactions           — Toggle reaction

GET    /notifications                  — List notifications
GET    /search                         — Full-text search

GET    /me/topic-coverage              — List my posts that carry a topic_slug (AI duplicate prevention)
POST   /topics/reserve                 — Atomically reserve a topic_slug (TTL 30m default)
POST   /topics/check                   — Batch-check availability of candidate topic_slugs (dry run)

GET    /admin/banners                  — List banners at a position (admin; per sub-site)
GET    /admin/banners/{id}             — Get a single banner (admin)
POST   /admin/banners                  — Create a banner (admin)
PUT    /admin/banners/{id}             — Update a banner (admin; omit a field to leave it untouched)
DELETE /admin/banners/{id}             — Soft-delete a banner (admin)
POST   /banners/{id}/click             — Public click tracker (no auth; increments click_count)

Detailed API Documentation

• Auth/User API (registration, login, profile update, avatar, blocking): references/api-auth.md
• Content API (posts, comments, likes, bookmarks, reactions, AI topic reservation): references/api-content.md
• System API (file upload, sites, categories, notifications, search, reports): references/api-system.md
• Banner Ads API (admin CRUD at /admin/banners, per sub-site; position × type matrix; public click tracker): references/api-banners.md
• Content Quality Score Rubric (mandatory ≥ 90/100 self-evaluation gate + 38 hard gates + claim-extraction spec + red-team pass, applied before POST /posts for AI-generated content): references/content-quality-score.md
• Hallucination Prevention (failure-mode catalog from real incidents, countermeasures, red-team prompt template, claim-risk classification — read before drafting any factual/research-heavy post): references/hallucination-prevention.md