Guide · Agent skill
How to use the ljg-blind agent skill: blind spot scan
Updated · written with AI from the project's own documentation and checked against our catalogue data
ljg-blind is an agent skill that reads the AI conversations you had yesterday (or on a date you pass in), looks for one recurring thinking habit that keeps some kind of truth out of view, and recommends one chapter from a WeChat Reading book to help close that gap. It saves the result as a Markdown analysis note. This guide explains what the skill does, how to install it, how to trigger it, and what it needs before you run it.
Blind Spot Scannerlijigang/ljg-skills · MIT Download the skill Quick start
- 1Install the skill folder. Download the ZIP for skills/ljg-blind and unzip it into ~/.claude/skills/ or a project's .claude/skills/, or upload it under Settings → Capabilities → Skills in the Claude apps.
- 2Set your WeRead key. Export WEREAD_API_KEY with your WeRead agent key in the format wrk-xxxx before you start the agent.
- 3Confirm the tools are installed. Make sure jq, curl and bun are available on your macOS machine, since the skill calls them directly.
- 4Ask for a scan. Say 扫盲区 or ljg-blind, or pass a date such as 2025-01-15 to scan a day other than yesterday.
- 5Open the note. Find the Markdown file the agent saves in ~/Context and read the blind spot, its evidence, and the linked chapter.
What the ljg-blind skill makes the agent do
An agent skill is a folder of instructions and optional scripts that an AI agent loads when your request matches the skill's description. ljg-blind's job is narrow: find a blind spot in how you think, not a gap in what you know. The source defines a blind spot as a structural thinking habit that makes a certain kind of truth systematically invisible to you.
The agent works through five steps. It gathers yesterday's user messages from your local Codex session files, checks them against five signals, picks one blind spot, finds a matching chapter on WeChat Reading, and writes a note to a folder called ~/Context.
- 绕开点: a hard question is raised and then quickly dropped or changed
- 空转框架: one question is viewed through several frameworks, none of which lands
- 单一取景框: only one or two frameworks are used all day, the ones you lean on most
- 未检前提: an assumption runs through the whole conversation and is never questioned
- 相邻空缺: a question you should have asked but didn't, pointed to by your own clues
Install the ljg-blind skill
Download the ZIP for skills/ljg-blind from the source page, then unzip it into a skills folder. You can use your personal folder (~/.claude/skills/) or a project folder (.claude/skills/ inside the project). If you use the Claude apps instead, upload the skill under Settings → Capabilities → Skills.
The skill's own tool is referenced by the path ${HOME}/.agents/skills/ljg-blind/Tools/WeReadWebUrl.ts. If you install into ~/.claude/skills/, make sure that path also resolves to the folder, or the chapter-link step will fail. Check the folder contains SKILL.md and Tools/ with WeReadWebUrl.ts and WeReadWebUrl.help.md.
- Expected files: SKILL.md, Tools/WeReadWebUrl.help.md, Tools/WeReadWebUrl.ts
- Set the WEREAD_API_KEY environment variable before you run it (format wrk-xxxx)
export WEREAD_API_KEY=<key>Check the requirements before you run it
The skill is built for macOS. Its date commands use BSD date flags (date -v and date -j), so they will not work as written on Linux. It also needs jq to parse session files, curl for the WeChat Reading API calls, and bun to run the link tool.
The skill reads Codex session files under ~/.codex/sessions and assumes that layout. You also need a WeRead agent key. If you have a LIFEOS_DIR set, the skill reads USER/TELOS/PRINCIPAL_TELOS.md for mission context; this is optional.
- macOS with BSD date
- jq
- curl
- bun
- WEREAD_API_KEY with a WeRead agent key (wrk-xxxx)
- Local Codex session files at ~/.codex/sessions
- Optional: LIFEOS_DIR pointing to a folder with USER/TELOS/PRINCIPAL_TELOS.md
Trigger the skill with natural requests
You do not need special syntax. Ask in plain language, and the agent matches your request to the skill. The source lists these trigger phrases: 扫盲区, 盲区, 照盲区, 看看我的思维盲区, 我昨天想漏了什么, blind spot, ljg-blind, and /ljg-blind.
To scan a specific day instead of yesterday, pass a date in YYYY-MM-DD format.
- 扫盲区
- 看看我的思维盲区
- 我昨天想漏了什么
- blind spot
- /ljg-blind 2025-01-15 (pass a date in YYYY-MM-DD format; replace with your own date)
Read the output note
The skill writes one Markdown note to ~/Context, named with a timestamp and the blind-spot topic. The note has three parts: what you were thinking about yesterday, the one blind spot the agent picked and the evidence for it, and the chapter to read. The chapter entry includes the book, the chapter title, an estimated reading time, and a link that opens the chapter in the WeChat Reading web reader.
The skill is designed to say so when the data is thin. If yesterday's real user input is under 200 characters or there are no conversations, the note states that and does not invent a blind spot.
- Blind spot with the specific messages that show it
- One chapter, not a whole book
- Reading time estimate based on word count
- Web reader link generated by the bundled tool
What people use it for
- Reviewing last week's Codex conversations on a regular schedule to spot a habit you keep repeating.
- Checking a single day you remember as unproductive, by passing its date.
- Finding one WeChat Reading chapter that targets a question you kept avoiding in your AI chats.
- Keeping a personal archive of analysis notes in ~/Context for later reflection.
Frequently asked questions
Is ljg-blind free to use?
The source lists the skill under the MIT licence. It does not state any prices for the skill itself. You still need a WeRead agent key and the required tools on your own machine.
What do I need to install ljg-blind?
You need macOS (the date commands use BSD flags), jq, curl, and bun. You also need a WEREAD_API_KEY environment variable and Codex session files at ~/.codex/sessions. Install the skill by unzipping it into ~/.claude/skills/ or a project's .claude/skills/, or upload it under Settings → Capabilities → Skills.
Does ljg-blind work on Windows or Linux?
The source says it is built for macOS and uses BSD date flags such as date -v and date -j. It is not documented for Windows or Linux, so expect the date steps to fail there without changes.
Does ljg-blind send my conversations to the internet?
It reads your local Codex session files, which may contain sensitive personal or work content. The chapter steps make network calls to the WeChat Reading API, so the skill does use the network. Review your session files before you run it if that matters to you.
Why does the chapter link not open the right chapter?
The WeChat Reading web reader needs encoded IDs. A bare bookId and chapterUid in the query string returns a 404 according to the source. Use Tools/WeReadWebUrl.ts to generate the link, which combines the two encoded IDs with a k between them.
How does ljg-blind compare with ljg-think or ljg-read?
The source says ljg-blind is not for drilling into one idea (use ljg-think), reducing a field (use ljg-rank), reading a text with the agent (use ljg-read), or finding constraints in a field (use ljg-constraint). ljg-blind is for finding one blind spot in your recent AI chats and pairing it with a chapter.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use ljg-blind if you talk to an AI assistant daily, want a reflective review of your own reasoning habits, and already use WeChat Reading. Skip it if you want quick answers, do not want your session history read, or cannot run macOS with the required tools and WeRead key. For a single idea to explore in depth, use ljg-think instead.
Download the skill