Reply Current

Overview

Use this skill whenever you need to return a normal successful result to QQ. Do not ask the bridge to mirror your last assistant message automatically. Send the result yourself.

When to Use

• You finished a task and need to answer the current QQ conversation.
• You generated an image, markdown file, report, or other artifact under .run/artifacts/.
• You need to send a short text result directly.

Do not use this skill for failures. Bridge-generated errors are handled by the runtime.

Rules

• Reply only to the current active conversation. Do not invent QQ IDs or group IDs.
• Use exactly one local reply_current.py command per message you want to send.
• Always pass --context-file <path>. The path is given to you in the "# Reply context" section of your developer_instructions and is specific to THIS thread's conversation. The legacy singleton at .run/default/run/reply_context.json is shared across all active tasks and may point at the wrong group under concurrency — skill calls that omit --context-file can silently deliver the reply into another chat.
• Attachments must already exist under .run/artifacts/.
• Prefer text for short answers, image for visual output, and file for markdown/report artifacts.
• If you want line breaks in QQ, put real newline characters in --text. The plain "..." form in bash does NOT turn \n into a newline; use $'line1\nline2' (ANSI-C quoting), or split a "..." string across two source lines, or call this skill once per line. The bridge defensively decodes a stray \n/\r\n/\t if one slips through, but writing real newlines is cleaner.
• Group messages reach you with mention markers preserved: @<bot> is the placeholder for an @ aimed at the bot itself, and @nickname<QQ:1234567> (or @<QQ:1234567> when the original at segment carried no name) is the placeholder for an @ aimed at any other user. Read those markers when relevant; do not echo them back into your reply — a bare QQ id is meaningless to a human reader. The bridge will defensively strip any marker that slips through (and downgrade @nickname<QQ:...> to @nickname), but write clean text to begin with. Only @ someone when the sender asked for it; do not invent extra pings.
• When the inbound message quoted another message, the bridge prepends a context block in the form [quote<msg:12345> @nickname<QQ:1111>: 原文] (or [quote<msg:12345>] when the fetch failed). Use it as read-only context. If the quoted content matters and the block is missing the real body — especially for quoted images — recover the original payload first with qq-quoted-image-recovery instead of inventing a summary.

Commands

Send plain text (default: @-mentions the original sender and quotes the triggering message):

python3 skills/reply-current/reply_current.py \
  --context-file /abs/path/from/developer_instructions/contexts/group_111.json \
  --text "处理完啦～"

--context-file is mandatory (the absolute path is given in your developer_instructions "# Reply context" section); omit it and you risk cross-talk between concurrent groups.

Send plain text and @-mention specific users instead of the sender:

python3 skills/reply-current/reply_current.py \
  --context-file /abs/path/from/developer_instructions/contexts/group_111.json \
  --text "这是你要的结果" \
  --at 1234567

python3 skills/reply-current/reply_current.py \
  --context-file /abs/path/from/developer_instructions/contexts/group_111.json \
  --text "你们看看这个" \
  --at 1234567 7654321

Quote a specific earlier message instead of the triggering one (the reply pill in QQ will land on that message):

python3 skills/reply-current/reply_current.py \
  --context-file /abs/path/from/developer_instructions/contexts/group_111.json \
  --text "找到了，就是这条" \
  --reply-to 12345

Send an image artifact:

python3 skills/reply-current/reply_current.py \
  --context-file /abs/path/from/developer_instructions/contexts/group_111.json \
  --image .run/artifacts/result.png

Send a file artifact:

python3 skills/reply-current/reply_current.py \
  --context-file /abs/path/from/developer_instructions/contexts/group_111.json \
  --file .run/artifacts/report.md

Choosing who to @

By default the bridge @-mentions the person who sent the original message. Use --at to override this when the context makes it clear the sender wants someone else to see the reply:

• Sender @-mentioned another user alongside the bot — e.g. @bot 帮 @小明<QQ:1234567> 看看这个. The sender wants 小明 to see the answer. Pass --at 1234567 (or --at 1234567 <sender_qq> if you also want to @ the sender).
• Sender explicitly asked you to reply to someone — e.g. "把结果发给 @小明QQ:1234567". Pass --at 1234567.
• No special mention context — omit --at entirely; the bridge will @ the sender as usual.

Read the QQ id from the @nickname<QQ:...> placeholder in the incoming message text. Do not guess QQ ids.

Choosing which message to quote

The reply pill in QQ defaults to the message that addressed you. Override with --reply-to <msg_id> when jumping to a different line reads more naturally:

• Sender asked you to find an earlier chat record — e.g. "帮我翻一下昨天小明说过的那句关于部署的话". After you locate the target message id (from [quote<msg:...>] in the inbound block, or from history the sender pasted), pass --reply-to <that_msg_id> so the pill lands on the actual target.
• Sender replied-to an earlier message while asking you a follow-up — if your answer is about that quoted message itself (not the sender's latest sentence), pass --reply-to <quoted_msg_id> so the pill rebounds to that line.
• No special context — omit --reply-to; the default reads naturally.

Never fabricate a --reply-to id. Read ids from placeholders in the inbound text; do not invent numbers.

Common Mistakes

• Do not use send-private or send-group for normal task results.
• Do not point to files outside .run/artifacts/.
• Do not assume bridge will send your final assistant text for you.