bbc-skill — Bilibili Comment Collector

Download all comments (top-level + nested + pinned) for a Bilibili video and produce data that Claude Code can analyze downstream.

Update check

Throttle to one check per 24 hours per installation; never mutate the skill directory without explicit user consent.

1. If <this-skill-dir>/.last_update exists and is less than 24 hours old, skip this step entirely.

2. Otherwise, fetch the latest tag from upstream:

   git -C <this-skill-dir> ls-remote --tags origin 'v*' 2>/dev/null \
     | awk '{print $2}' | sed 's|refs/tags/||' \
     | sort -V | tail -1
   

3. Compare with this skill's metadata.version from the frontmatter. If the upstream tag is strictly newer (semver), tell the user one line and ask:

   "A newer version of this skill is available: vX.Y.Z → vA.B.C. Want me to git pull?"

   If they say yes, run git -C <this-skill-dir> pull --ff-only. Refresh .last_update either way so the prompt doesn't repeat for 24 hours.

4. If upstream is the same or older, refresh .last_update silently and continue.

5. On any failure (offline, not a git checkout — e.g. ClawHub-installed copy, read-only path, no permission), swallow the error silently and continue with the user's task. Do not mention the failure.

When to use

Trigger this skill when the user:

• Asks to get / fetch / download / export / collect / analyze comments of a specific Bilibili video (BV 号, URL, or video page).
• Asks to analyze audience feedback / sentiment / keywords / top comments / IP distribution of their own Bilibili videos.
• Provides a Bilibili URL like https://www.bilibili.com/video/BVxxxxxxxxxx/.
• Mentions their UP主 UID and wants batch analysis across their videos.

Do not use for: posting / deleting comments, downloading videos, barrage (弹幕), live stream data, or private messages.

Prerequisites

1. Python 3.9+ (stdlib only — zero pip install).

2. Bilibili cookie. The user must be logged in to bilibili.com. The recommended path:

   • Install the Chrome/Edge extension Get cookies.txt LOCALLY (open-source, fully local, no upload).
   • On a logged-in bilibili.com tab, click Export → save www.bilibili.com_cookies.txt.
   • Pass via --cookie-file or set $BBC_COOKIE_FILE.

   Alternatives:

   • $BBC_SESSDATA env var with just the SESSDATA value.
   • Browser auto-detection (Firefox / Chrome / Edge on macOS) via --browser auto. Works best for Firefox; Chrome/Edge needs a logged-in profile with cookies flushed to disk.

Auth delegation (Principle 7): the skill never runs OAuth flows. The human is expected to log in via browser; the agent only consumes the resulting cookie.

Quick start

Before any fetch, verify the cookie works:

python3 -m bbc cookie-check

Success envelope (stdout):

{"ok":true,"data":{"mid":441831884,"uname":"探索未至之境","vip":false}}

Fetch all comments for a single video:

python3 -m bbc fetch BV1NjA7zjEAU

Or pass a URL:

python3 -m bbc fetch "https://www.bilibili.com/video/BV1NjA7zjEAU/"

Output (default ./bilibili-comments/<BV>/):

• comments.jsonl — one comment per line, flattened
• summary.json — video metadata + statistics + top-N
• raw/ — archived API responses
• .bbc-state.json — resume state

Commands

Call bbc <cmd> --help or bbc schema <cmd> for full parameter details — do not guess flag names.

Agent contract

Stdout vs stderr

• stdout: stable JSON envelope {"ok":true,"data":...} or {"ok":false,"error":...}. JSON is the default when stdout is not a TTY. Pass --format table for human-readable tables.
• stderr: human log lines + NDJSON progress events for long tasks.

Exit codes

Error envelope

{
  "ok": false,
  "error": {
    "code": "auth_expired",
    "message": "SESSDATA 已过期，请重新登录 B 站",
    "retryable": true,
    "retry_after_auth": true
  }
}

Error codes: validation_error, auth_required, auth_expired, not_found, rate_limited, api_error, network_error. See bbc schema for the full contract.

Dry-run

Every fetch command supports --dry-run to preview the planned request without making network calls:

python3 -m bbc fetch BV1NjA7zjEAU --dry-run

Idempotency

Re-running the same fetch command on the same output directory resumes from .bbc-state.json (skips already-fetched pages). Pass --force to refetch.

Analysis workflow (for the agent)

After fetch completes:

1. Read summary.json first (< 10 KB) to establish global context: video metadata, total counts, time distribution, top-N.
2. For thematic analysis, Grep or head/tail on comments.jsonl — each line is a flat JSON object, never load the whole file unless small.
3. Typical analyses:
   • Sentiment distribution → scan message by batch
   • Top fans → group by mid, count entries, aggregate like
   • UP 主互动 → filter is_up_reply=true
   • Audience geography → ip_location histogram
   • Feedback timeline → bucket ctime_iso by day/week

The summary.json schema is documented in references/agent-contract.md. Run the skill against any video to produce a real sample locally.

Safety tier

All commands are read-only (tier: open). No mutation, no deletion, no message sending. Dry-run available for all fetch commands.

References

• references/api-endpoints.md — Bilibili API fields used
• references/cookie-extraction.md — per-browser cookie decryption
• references/agent-contract.md — full envelope + schema contract

Limitations

• all_count returned by the API includes pinned comments. Completeness check: top_level + nested + pinned == declared_all_count.
• Very old comments (>2 years) may return thin data if the user was deleted.
• Anti-bot: aggressive --max values or repeated runs may trigger HTTP 412. The client sleeps 1s between requests and backs off on 412.