qq-current-history
QQ Current History
Overview
Use this skill when the user asks about earlier messages in the current QQ conversation and the quoted preamble is not enough.
This skill is lane-scoped:
- it only queries the current group or private chat,
- it uses the current lane's reply-context token,
- it never accepts arbitrary
group_idoruser_id.
When to Use
- The user asks "刚才谁说了什么", "昨天群里那句", "上周三聊部署那条".
- The user wants the context around a quoted message.
- The user wants you to quote an earlier message back with
reply-current --reply-to. - You need to verify who said something before drafting a reply.
If the quoted message was probably an image or screenshot and you need the
pixels, switch to qq-quoted-image-recovery instead of treating this as plain
text history.
Rules
- Always pass
--context-file <path>. The path is given in your developer instructions for THIS thread. - Query only the current conversation. Never invent other chat ids.
- Support time, sender, keyword, and free-form query intent.
- The bridge scan budget is bounded. If the result says
truncated: trueand you still need more precision, narrow the query instead of broadening it. - If no result is found, say so plainly. Do not fabricate QQ history.
- Returned
message_idvalues are safe to use withreply-current --reply-to.
Commands
Free-form lookup:
python3 skills/qq-current-history/query_current_history.py \
--context-file /abs/path/from/developer_instructions/contexts/group_111.json \
--query "找今天下午提部署那句"
Keyword lookup:
python3 skills/qq-current-history/query_current_history.py \
--context-file /abs/path/from/developer_instructions/contexts/group_111.json \
--keyword "部署"
Sender-scoped lookup:
python3 skills/qq-current-history/query_current_history.py \
--context-file /abs/path/from/developer_instructions/contexts/group_111.json \
--sender-name "小明" \
--keyword "部署"
Time-bounded lookup:
python3 skills/qq-current-history/query_current_history.py \
--context-file /abs/path/from/developer_instructions/contexts/group_111.json \
--start-time 1744970400 \
--end-time 1744977600
Reading Results
messagesis the filtered normalized transcript slice.message_idis the QQ message id you can reuse with--reply-to.timestampis unix seconds.sender_nameandtextare already normalized into bridge-friendly text.truncated: truemeans the bridge hit its scan budget.
Common Mistakes
- Do not use this skill to search other groups or other private chats.
- Do not retry with a broader and broader query if the first attempt is truncated.
- Do not treat
queryas magic semantic search. It is just a bounded bridge query input, so tighter filters are better.
More from acking-you/codex-bridge
reply-current
Use when a Codex turn running inside codex-bridge must send normal results back to the active QQ conversation as text, image, or file
1gpt2api-image-generator
Use when a QQ user asks Codex to draw, paint, create, generate, or render an image through GPT2API, including Chinese requests such as 画图, 画一张, 生成图片, 出图, 做张图, or 帮我画
1invoke-capability
Call another registered model (Claude via Kiro, Gemini images, ...) to produce this turn's reply, keeping Codex as the harness but letting a better-suited model own the actual voice. Use liberally for conversational / emotional / creative turns; Codex's own voice is a poor fit for most QQ chat.
1qq-quoted-image-recovery
Use when a quoted QQ message likely contained an image or screenshot but the bridge only exposed a quote marker or flattened text
1staticflow-kiro-log-diagnoser
Use when a Codex turn needs to diagnose StaticFlow Kiro upstream failures by reading the backend error log and correlating real usage events through sf-cli
1