kendreaditya/claude/skills/blind

Blind

Read-only CLI for teamblind.com. Wraps scripts/blind.py, a single-file Python tool that hits Blind's public Next.js RSC payloads — no login, no cookies, no encryption.

When to invoke

Reach for this skill whenever the user wants real anonymous tech-worker discussion on a topic. Blind is the source of truth for layoff rumors, severance negotiation tactics, comp data points, interview leaks, and internal-feel gossip about specific companies. If the question is "what's the chatter on X", "any Blind threads about Y", or contains a teamblind.com URL, use this.

Setup

Run the script directly with python3. No dependencies — stdlib only. The script is at scripts/blind.py relative to this skill's directory.

python3 scripts/blind.py --help

Commands

search — find posts by query

python3 scripts/blind.py search "<query>" [--limit N] [--json]

Returns first page of search results (typically ~20 posts) with title, author company tag, like/comment/view counts, ~300-char snippet, and canonical article URL.

python3 scripts/blind.py search "severance negotiation" --limit 5

read — fetch an article + comments

python3 scripts/blind.py read <url-or-alias> [--json]

Accepts a full Blind URL (https://www.teamblind.com/article/4dzzibmq, /post/4dzzibmq, or /company/<co>/posts/<slug>) OR a bare alias (4dzzibmq). Renders title, channel, author, body excerpt (~150 chars — see "Body limitation" below), and the full threaded comment tree with company tags, OP markers, like counts, and reply nesting.

python3 scripts/blind.py read m58k2r2o
python3 scripts/blind.py read https://www.teamblind.com/article/4dzzibmq

feed — browse a channel feed (or home)

python3 scripts/blind.py feed [channel] [--since DURATION] [--limit N] [--json]

Channel names go after feed (e.g. Layoffs, Tech-Industry, India). Omitting the channel returns the home feed. --since filters client-side to posts written within a duration like 10m, 6h, 2d, 1w. First page only — pagination is not supported.

python3 scripts/blind.py feed Layoffs --since 24h
python3 scripts/blind.py feed                    # home

If the user gives a channel name that doesn't exist, the script errors with "no such channel" rather than silently returning the home feed.

company — browse a company page

python3 scripts/blind.py company <Name> [--since DURATION] [--limit N] [--json]

Lists posts on a specific company's page. Use the company's URL alias as Blind has it (e.g. Meta, Google, Amazon, Salesforce).

python3 scripts/blind.py company Meta --limit 10
python3 scripts/blind.py company Salesforce --since 7d

export — dump article + comments to a file

python3 scripts/blind.py export <url-or-alias> [--format md|json]

Writes a self-contained Markdown (default) or JSON document to stdout. Useful for archiving threads the user wants to keep around — Blind posts get buried fast and accounts can vanish.

python3 scripts/blind.py export 4dzzibmq --format md > thread.md
python3 scripts/blind.py export m58k2r2o --format json > thread.json

Common flags (work on every command)

• --json — emit structured JSON instead of pretty text. Pipe into jq or other tooling. Schema is the Post / Article dataclass.
• --limit N — cap output to N items.
• -v / --verbose — print fetched URLs to stderr for debugging.

Common flags must appear after the subcommand name, e.g. blind.py feed Layoffs --limit 5 (not blind.py --limit 5 feed Layoffs).

How to choose the right command

| User intent | Command | |---|---| | "What's Blind saying about X?" / "Find Blind threads about Y" | search "<X>" | | "What are people on Blind saying about Meta/Google/etc?" | company <Name> | | "Show me the Layoffs channel" / "Tech Industry feed" | feed <Channel> | | "What's hot on Blind right now?" | feed (no channel = home) | | "Read this Blind post: <url>" | read <url> | | "Save/export this Blind thread" | export <url> --format md | | "Show me last 24h of layoffs posts" | feed Layoffs --since 24h |

When the user gives a teamblind.com URL, use read. When they describe a topic, start with search. When they name a specific company, use company. When they reference a channel/forum, use feed.

Body limitation (important — communicate this to the user)

Blind lazy-loads the full text of long article bodies via a Server Action that runs after page render. The skill cannot replicate that without reverse-engineering action hashes (which change every Blind deploy). Implications:

• read shows ~150 chars of body (the SEO meta description) plus the full comment thread. The output includes a clear warning when this happens.
• Feed/search/company listings show up to 300 chars per post (Blind truncates content at 300 in listings, with contentLength reporting the real total). The renderer appends […+N chars] when truncated.
• Comments are NOT truncated — full threaded discussion is always available, including like counts, company tags, and reply nesting.
• For the FULL body of a long post, the user has to open the URL in a browser. The skill always prints the URL.

In practice this is fine for most Blind use cases because the high-value content is the comment discussion, and most Blind posts are short questions or rants under 300 chars anyway. But it's worth telling the user "I can show you the title, ~150 chars of the post, and all 67 comments — open the URL for the full body" rather than silently truncating.

Other limitations

• First page only on search, feed, company. No "load more".
• Default sort only — ?sort=... query params don't work on feeds (sort is a Server Action). They DO work on a topic-aggregation page but not on /article/<alias>, so the skill doesn't expose a --sort flag.
• Read-only. No login, no posting, no commenting, no voting, no DMs. Those would require defeating Blind's SJCL+RSA encrypted /api/* endpoints, which is out of scope.
• Engagement counts (likes/comments/views) are present in feed listings but rendered as compact icons (♥, 💬, 👁).

Output handling tips

• Default text output is human-readable with tree-indented comments; pipe to less -R if it's long.
• For programmatic use (counting threads, filtering by company, sorting by engagement), always use --json and process with jq or load into Python.
• The JSON Post schema has: alias, title, content, content_length, author_nickname, author_company, channel_name, created_at (relative, e.g. "4d"), written_at (ISO 8601 absolute), like_count, comment_count, view_count, url.
• The JSON Article schema adds: body, body_is_excerpt, pinned comment list, comments comment tree, total_comment_count. Each Comment has id, parent_id, author_nickname, author_company, content, like_count, is_op, replies[].

Error handling

The script exits with:

• 0 on success
• 1 on errors (network, parse failure, etc.)
• 2 on "not found" — bad article alias, nonexistent channel, or unknown company page. Blind silently redirects unknown topic names to the home feed; the skill detects this via og:url and reports "no such channel" rather than returning home posts.