饭否 CLI (fanfou)

An LLM-friendly command line for the Fanfou (饭否) API, plus a bundled agent skill that works with Claude Code, Codex, Cursor, and 50+ other AI coding agents. JSON-by-default output, schema introspection, dry-run previews, and a three-layer command surface (shortcuts / resource commands / raw API).

Requirements

• Node.js ≥ 22.6 (the CLI runs TypeScript directly via type-stripping — no build step needed).

Install

# 1) The CLI (provides the `fanfou` command):
npm install -g fanfou          # or one-off, no install: npx fanfou <command>

# 2) The skill, into your AI agent(s) — via npx skills (GitHub as registry):
npx skills add Leaking/fanfou-cli -y -g
#    …or target specific agents (Claude Code, Codex, Cursor, 50+ more):
npx skills add Leaking/fanfou-cli --agent claude-code codex cursor

Run it:

fanfou --help

Developing from a clone (Node ≥ 22.6 runs the TypeScript directly — no build):

node src/index.ts --help

Quick start

# 1) Log in (XAuth). Use env vars to keep secrets out of argv.
FANFOU_USERNAME='[email protected]' FANFOU_PASSWORD='••••••' fanfou auth login

# 2) Read your timeline
fanfou +timeline --count 10
fanfou +timeline --format table

# 3) Post / reply / repost
fanfou +post "Hello 饭否"
fanfou +reply <status-id> "说得对"
fanfou +repost <status-id>

# 4) Anything else, raw:
fanfou api GET statuses/mentions.json --query count=5

Three layers

1. Shortcuts (+timeline, +post, +reply, +repost, +mentions, +me, +search, +fav, +dm) — high-frequency operations with smart defaults.
2. Resource commands — auth, timeline, status, favorite, user, friendship, dm, account, search (full Fanfou API coverage).
3. Raw API — fanfou api <GET|POST> <path> [--query ...] [--form ...].

Authentication

Two flows, stored per-profile under ~/.config/fanfou/config.json (mode 0600):

• XAuth (default): fanfou auth login -u <name> -p <pass>
• OAuth web: fanfou auth oauth-url → approve in browser → fanfou auth oauth-exchange --token … --secret … [--verifier …]

Env overrides (handy for CI / agents): FANFOU_USERNAME, FANFOU_PASSWORD, FANFOU_OAUTH_TOKEN, FANFOU_OAUTH_TOKEN_SECRET, FANFOU_CONSUMER_KEY, FANFOU_CONSUMER_SECRET, FANFOU_PROFILE, FANFOU_CONFIG_DIR.

Multiple accounts: pass --profile <name> to any command, or fanfou auth use <name> to change the default.

LLM / agent friendliness

• JSON by default on stdout; errors are JSON on stderr with exit codes (2 usage, 3 auth-required, 1 other/HTTP).
• Schema introspection: fanfou <cmd> --help --format json.
• Dry-run (--dry-run / -n) on every state-changing command prints the exact signed request without sending it.
• Output formats: --format json|ndjson|table|raw (-o).
• Robust arg parsing for Fanfou IDs that begin with -.

The bundled agent skill

A single, cross-agent skill lives at skills/fanfou/SKILL.md. It is consumable by any AI coding agent that supports the skills convention — Claude Code, Codex, Cursor, Cline, Aider, Continue, and 50+ more.

Install it globally for all of your agents at once, or scope it to a specific agent:

npx skills add Leaking/fanfou-cli -y -g                           # all agents
npx skills add Leaking/fanfou-cli --agent claude-code codex cursor # specific

From a local clone, you can also drop the skill into a Claude Code skills directory directly (other agents typically use the skills CLI above):

npm run install-skill           # -> ./.claude/skills/fanfou (project)
npm run install-skill -- --user # -> ~/.claude/skills/fanfou (global)

Develop

npm run typecheck   # tsc --noEmit
npm test            # node --test (OAuth1 signer vectors)
npm run build       # emit dist/

The OAuth1 HMAC-SHA1 signer matches the canonical RFC 5849 example signature (pinned by test/oauth1.test.ts). Fanfou signs the base string over http:// even on HTTPS requests — the signer replicates that quirk.