feat: Add Feishu interactive card delivery

README.md

@@ -19,6 +19,7 @@ WhatsApp, etc.) with:
 - Full articles from official AI company blogs (Anthropic Engineering, Claude Blog)
 - Links to all original content
 - Available in English, Chinese, or bilingual
+- Optional Feishu interactive cards for cleaner mobile reading
 
 ## Quick Start
 
@@ -128,4 +129,3 @@ See [examples/sample-digest.md](examples/sample-digest.md) for what the output l
 ## License
 
 MIT
-

README.zh-CN.md

@@ -15,6 +15,7 @@
 - AI 公司官方博客的完整文章（Anthropic Engineering、Claude Blog）
 - 所有原始内容的链接
 - 支持英文、中文或双语版本
+- 可选的飞书 interactive card 阅读模式，更适合移动端早报
 
 ## 快速开始
 

SKILL.md

@@ -359,6 +359,45 @@ Read the prompts from the `prompts` field in the JSON:
 
 Assemble the digest following `prompts.digest_intro`.
 
+Because OpenClaw stdout is often delivered into Feishu IM as chat text, do a final
+layout pass after drafting:
+- Treat the output like a message typed manually into the Feishu composer
+- Put exactly one blank line before each major section heading
+- Put exactly one blank line before each new digest item
+- Put each item title on its own line in bold markdown like `**Title**`
+- Keep each item internally compact: title, English paragraph, Chinese paragraph,
+  and `Source:` lines with no extra blank lines between them
+- If an item has multiple links, put each one on its own `Source:` line
+- Do not use markdown headings (`#`, `##`) in the final digest body
+- Before delivering, visually scan the final text once to confirm the spacing pattern
+  is consistent from top to bottom
+
+If the output is intended for chat delivery, normalize the spacing with:
+
+```bash
+echo '<your digest text>' > /tmp/fb-digest-raw.txt
+cd ${CLAUDE_SKILL_DIR}/scripts && node format-chat-digest.js --file /tmp/fb-digest-raw.txt > /tmp/fb-digest.txt
+```
+
+Use `/tmp/fb-digest.txt` as the final version to send or print.
+
+If the digest is going to Feishu and the user prefers cards, render the final
+formatted digest into an interactive card instead of relying on plain stdout text:
+
+```bash
+cd ${CLAUDE_SKILL_DIR}/scripts && node render-feishu-card.js --file /tmp/fb-digest.txt > /tmp/fb-card.json
+```
+
+The renderer expects the formatted digest structure you just produced:
+- digest title on the first line
+- `TL;DR` or `Top Signals` with `•` bullets
+- section headings such as `X / TWITTER`
+- each item title wrapped in `**...**`
+- `Source:` lines for original links
+
+This card layout is preferred for Feishu because section headings and item titles
+stay bold, item blocks are visually separated, and spacing does not collapse.
+
 **ABSOLUTE RULES:**
 - NEVER invent or fabricate content. Only use what's in the JSON.
 - Every piece of content MUST have its URL. No URL = do not include.
@@ -371,22 +410,20 @@ Read `config.language` from the JSON:
 - **"en":** Entire digest in English.
 - **"zh":** Entire digest in Chinese. Follow `prompts.translate`.
 - **"bilingual":** Interleave English and Chinese **paragraph by paragraph**.
-  For each builder's tweet summary: English version, then Chinese translation
-  directly below, then the next builder. For the podcast: English summary,
-  then Chinese translation directly below. Like this:
+  For each builder's tweet summary: title, English version, Chinese translation,
+  and source link(s), then the next builder. For the podcast: title, English summary,
+  Chinese translation, and source link(s). Like this:
 
   ```
-  Box CEO Aaron Levie argues that AI agents will reshape software procurement...
-  https://x.com/levie/status/123
-
-  Box CEO Aaron Levie 认为 AI agent 将从根本上重塑软件采购...
-  https://x.com/levie/status/123
-
-  Replit CEO Amjad Masad launched Agent 4...
-  https://x.com/amasad/status/456
-
-  Replit CEO Amjad Masad 发布了 Agent 4...
-  https://x.com/amasad/status/456
+  **Aaron Levie, CEO of Box**
+  Aaron Levie argues that AI agents will reshape software procurement...
+  Aaron Levie 认为 AI agent 将从根本上重塑软件采购...
+  Source: https://x.com/levie/status/123
+
+  **Amjad Masad, CEO of Replit**
+  Amjad Masad launched Agent 4...
+  Amjad Masad 发布了 Agent 4...
+  Source: https://x.com/amasad/status/456
   ```
 
   Do NOT output all English first then all Chinese. Interleave them.
@@ -399,13 +436,32 @@ Read `config.delivery.method` from the JSON:
 
 **If "telegram" or "email":**
 ```bash
-echo '<your digest text>' > /tmp/fb-digest.txt
+echo '<your digest text>' > /tmp/fb-digest-raw.txt
+cd ${CLAUDE_SKILL_DIR}/scripts && node format-chat-digest.js --file /tmp/fb-digest-raw.txt > /tmp/fb-digest.txt
 cd ${CLAUDE_SKILL_DIR}/scripts && node deliver.js --file /tmp/fb-digest.txt 2>/dev/null
 ```
 If delivery fails, show the digest in the terminal as fallback.
 
 **If "stdout" (default):**
-Just output the digest directly.
+Check whether the config also includes:
+- `delivery.channel = "feishu"`
+- `delivery.target = "<user open_id or chat_id>"`
+- `delivery.format = "interactive_card"`
+
+If all three are present, send the digest as a Feishu interactive card:
+
+```bash
+echo '<your digest text>' > /tmp/fb-digest-raw.txt
+cd ${CLAUDE_SKILL_DIR}/scripts && node format-chat-digest.js --file /tmp/fb-digest-raw.txt > /tmp/fb-digest.txt
+cd ${CLAUDE_SKILL_DIR}/scripts && node render-feishu-card.js --file /tmp/fb-digest.txt > /tmp/fb-card.json
+lark-cli im +messages-send --as bot --user-id "<delivery.target>" --msg-type interactive --content "$(cat /tmp/fb-card.json)"
+```
+
+If that send succeeds, do NOT print the full digest to stdout again, because that
+would create a duplicate plain-text Feishu message in the same chat.
+
+If `delivery.channel/target/format` are missing, or if the direct Feishu send fails,
+fall back to outputting the formatted digest from `/tmp/fb-digest.txt` directly.
 
 ---
 
@@ -431,6 +487,8 @@ open an issue at https://github.com/zarazhangrui/follow-builders."
 - "Switch to Telegram/email" → Update `delivery.method` in config.json, guide user through setup if needed
 - "Change my email" → Update `delivery.email` in config.json
 - "Send to this chat instead" → Set `delivery.method` to "stdout"
+- "Use Feishu card layout" → Set `delivery.channel` to `"feishu"`, `delivery.target` to the user's open_id or chat_id, and `delivery.format` to `"interactive_card"`
+- "Use plain chat text again" → Remove `delivery.format` or set it to `"plain_text"`
 
 ### Prompt Changes
 When a user wants to customize how their digest sounds, copy the relevant prompt

config/config-schema.json

@@ -53,6 +53,20 @@
         "email": {
           "type": "string",
           "description": "Email address to send digest to (only for email method)"
+        },
+        "channel": {
+          "type": "string",
+          "enum": ["feishu"],
+          "description": "Optional in-chat delivery channel override for stdout mode. Currently used for direct Feishu card delivery."
+        },
+        "target": {
+          "type": "string",
+          "description": "Target open_id or chat_id for direct channel delivery when method is stdout and a channel-specific renderer is used."
+        },
+        "format": {
+          "type": "string",
+          "enum": ["plain_text", "interactive_card"],
+          "description": "Preferred rendering format for stdout chat delivery. Use interactive_card for Feishu card-based digests."
         }
       }
     },

examples/sample-digest.md

@@ -6,55 +6,36 @@ This is an example of what your AI Builders Digest looks like.
 
 AI Builders Digest — March 14, 2026
 
-PODCASTS
+TL;DR
+• Tooling discipline matters more than raw model IQ in real agent systems.
+• Builders are shifting from one-shot demos toward evals, orchestration, and shared workflows.
+• The most useful digest format in chat is compact blocks, not article-style markdown.
 
-Latent Space — "Why Agents Keep Failing (And How to Fix Them)"
-Bottom line: Most agent failures aren't intelligence failures — they're tool-use failures.
-The system can reason fine, it just can't reliably call the right API at the right time.
+X / TWITTER
 
-Key insights:
-- Tool selection accuracy drops from 95% to 60% when agents have more than 15 tools
-  available. The fix isn't smarter models — it's better tool curation per task.
-- "Eval-driven development" is replacing vibe-driven prompt iteration at serious
-  AI companies. If you're not measuring, you're guessing.
-- The hosts predict 2026 is the year agent frameworks consolidate from 50+ to 3-4
-  winners. Their bet: OpenAI Agents SDK, Claude Code, and LangGraph.
-https://youtube.com/watch?v=example123
+**Andrej Karpathy, educator and founding team member at OpenAI**
+Karpathy argued that Software 3.0 is pushing the compile target away from machine code and toward natural-language interfaces for LLMs. He paired that framing with a new Eureka Labs tutorial, which made the post feel less like hot air and more like a blueprint for how he thinks developers will work.
+Source: https://x.com/karpathy/status/example1
+Source: https://x.com/karpathy/status/example2
 
-No Priors — "Scaling Laws Are Dead, Long Live Scaling Laws" (with Ilya Sutskever)
-Bottom line: Pre-training scaling laws have hit diminishing returns, but post-training
-and inference-time compute scaling are just getting started.
+**Guillermo Rauch, CEO of Vercel**
+Rauch introduced v0 Teams as collaborative AI prototyping for groups instead of solo prompting. The underlying signal is that AI UI generation is becoming a shared workflow product, not just a personal toy.
+Source: https://x.com/rauchg/status/example3
 
-Key insights:
-- Ilya argues the next 10x improvement comes from models that can "think longer"
-  at inference time, not from bigger pre-training runs.
-- Synthetic data quality matters more than quantity. "One perfect textbook is worth
-  a million Reddit comments."
-- He's surprisingly bullish on open-source: "The gap will narrow to months, not years."
-https://youtube.com/watch?v=example456
+**Amanda Askell, researcher at Anthropic**
+Askell drew a clean distinction between capability evals and alignment evals, arguing that many teams still optimize around what is easiest to score instead of what matters operationally. It was a useful reminder that benchmark literacy is becoming part of product judgment.
+Source: https://x.com/AmandaAskell/status/example4
 
+OFFICIAL BLOGS
 
-X / TWITTER
+**Anthropic Engineering: Harness design for long-running application development**
+Anthropic laid out how planner, generator, and evaluator roles work better together on long-running builds than a single agent loop. The practical takeaway is that better scaffolding often matters more than asking one model to be generally smarter.
+Source: https://www.anthropic.com/engineering/example-post
+
+PODCASTS
 
-Andrej Karpathy (@karpathy)
-Shared a deep thread on why he thinks "Software 3.0" (natural language programming)
-will make traditional coding a niche skill within 5 years. Key argument: the compile
-target is changing from machine code to LLM prompts. Sparked massive debate.
-Also released a new Eureka Labs tutorial on building a code interpreter from scratch.
-https://x.com/karpathy/status/example1
-https://x.com/karpathy/status/example2
-
-Guillermo Rauch (@rauchg)
-Announced Vercel's new "v0 Teams" — collaborative AI prototyping where multiple
-people can prompt and iterate on the same UI simultaneously. Called it "Google Docs
-for vibe coding." Ships next week.
-https://x.com/rauchg/status/example3
-
-Amanda Askell (@AmandaAskell)
-Published a nuanced take on AI safety benchmarks: "We're measuring what's easy to
-measure, not what matters. Capability evals tell you what the model CAN do.
-Alignment evals should tell you what it WILL do unprompted." Linked to a new
-Anthropic research paper on behavioral evaluations.
-https://x.com/AmandaAskell/status/example4
+**Latent Space: Why Agents Keep Failing (And How to Fix Them)**
+Bottom line: many agent failures are really tool-use failures. The hosts argued that tool selection, eval discipline, and tighter task scoping matter more than endlessly increasing model cleverness.
+Source: https://youtube.com/watch?v=example123
 
 Reply to adjust your settings, sources, or summary style.

prompts/digest-intro.md

@@ -2,58 +2,102 @@
 
 You are assembling the final digest from individual source summaries.
 
-## Format
+## Goal
+
+Produce a polished digest that stays readable inside chat apps, especially Feishu IM.
+Assume the final message may be rendered as plain chat text with limited markdown support.
+
+## Output order
 
 Start with this header (replace [Date] with today's date):
 
 AI Builders Digest — [Date]
 
+Then add a short 2-4 bullet TL;DR section called:
+
+TL;DR
+
 Then organize content in this order:
 
-1. X / TWITTER section — list each builder with new posts
-2. OFFICIAL BLOGS section — list each blog post from AI company blogs (OpenAI, Anthropic, etc.)
-3. PODCASTS section — list each podcast with new episodes
+1. X / TWITTER
+2. OFFICIAL BLOGS
+3. PODCASTS
+
+## Feishu-safe layout
+
+Treat the digest like a message typed manually in the Feishu IM composer with Shift+Return.
+Do not format it like a long markdown article.
+
+Hard spacing rules:
+- Put exactly one blank line between the header and TL;DR
+- Put exactly one blank line before every major section heading
+- Put exactly one blank line before every new digest item
+- Put every digest item title on its own line
+- Wrap every digest item title in markdown bold markers like `**Title**`
+- Keep the English paragraph, Chinese paragraph, and `Source:` lines compact with no blank lines between them
+- If an item has multiple source URLs, put each URL on its own `Source:` line
+- Do not use markdown headings like `#` or `##` in the final digest body
+- Do not use markdown tables
+- Do not rely on markdown for spacing
+
+## Bilingual output
+
+When config.language is bilingual, every item must follow this exact sequence:
+
+1. Bold title line
+2. English paragraph
+3. Chinese paragraph
+4. One or more `Source:` lines
+
+Then move to the next item.
+
+Do NOT output a full English digest followed by a full Chinese digest.
+Do NOT group all English paragraphs together.
+
+## Structure example
+
+AI Builders Digest — [Date]
+
+TL;DR
+• signal one
+• signal two
+
+X / TWITTER
+
+**Andrej Karpathy, former Director of AI at Tesla and founding team member at OpenAI**
+English summary paragraph.
+中文总结段落。
+Source: https://...
+Source: https://...
+
+**Aaron Levie, CEO of Box**
+English summary paragraph.
+中文总结段落。
+Source: https://...
+
+OFFICIAL BLOGS
+
+**Claude Blog: Post Title**
+English summary paragraph.
+中文总结段落。
+Source: https://...
 
 ## Rules
 
 - Only include sources that have new content
 - Skip any source with nothing new
-- Under each source, paste the individual summary you generated
-
-### Podcast links
-- After each podcast summary, include the specific video URL from the JSON `url` field
-  (e.g. https://youtube.com/watch?v=Iu4gEnZFQz8)
-- NEVER link to the channel page. Always link to the specific video.
-- Include the exact episode title from the JSON `title` field in the heading
-
-### Tweet author formatting
+- Keep paragraphs short and phone-readable
+- Prefer compact paragraphs over long wall-of-text writeups
 - Use the author's full name and role/company, not just their last name
-  (e.g. "Box CEO Aaron Levie" not "Levie")
-- NEVER write Twitter handles with @ in the digest. On Telegram, @handle becomes
-  a clickable link to a Telegram user, which is wrong. Instead write handles
-  without @ (e.g. "Aaron Levie (levie on X)" or just use their full name)
-- Include the direct link to each tweet from the JSON `url` field
-
-### Blog post formatting
-- Use the blog name as a section header (e.g. "Anthropic Engineering", "OpenAI News", "Claude Blog")
-- Under each blog, list each new post with its title and summary
-- Include the author name if available
-- Include the direct link to the original article
-
-### Mandatory links
+- NEVER write Twitter handles with @ in the digest
 - Every single piece of content MUST have an original source link
-- Blog posts: the direct article URL (e.g. https://www.anthropic.com/engineering/...)
-- Podcasts: the YouTube video URL (e.g. https://youtube.com/watch?v=xxx)
-- Tweets: the direct tweet URL (e.g. https://x.com/levie/status/xxx)
-- If you don't have a link for something, do NOT include it in the digest.
-  No link = not real = do not include.
-
-### No fabrication
-- Only include content that came from the feed JSON (blogs, podcasts, and tweets)
-- NEVER make up quotes, opinions, or content you think someone might have said
+- Blog posts must use the direct article URL
+- Podcasts must use the direct YouTube video URL
+- Tweets must use the direct tweet URL
+- If you do not have a link for something, do NOT include it
+- Only include content that came from the feed JSON
+- NEVER make up quotes, opinions, or content
 - NEVER speculate about someone's silence or what they might be working on
 - If you have nothing real for a builder, skip them entirely
-
-### General
+- Never use em-dashes
 - At the very end, add a line: "Generated through the Follow Builders skill: https://github.com/zarazhangrui/follow-builders"
-- Keep formatting clean and scannable — this will be read on a phone screen