summarize-thread

Summarize a Pachca thread

Summarize the Pachca threads the user gave with the command: one or more thread links, message links or thread ids. If there is none, ask for it. With several, summarize each thread separately, in the order given. The user may add a window («since Monday», «since my last message») and a question to answer («did anyone reply to me», «what needs fixing»).

Follow the pachca-search skill and its thread procedure. A link can name the thread itself (thread_id), the message the thread hangs under (thread_message_id), or a single message (sidebar_message or message); that message can be the one the thread hangs under, a comment inside the thread or a message with no thread. For a thread id or a thread_id link, pachca_get_thread gives the thread's chat_id and message_id. Otherwise find the thread chat with the "Find the thread chat" table in the skill's references/threads.md. A message with thread is the parent itself: its id is the thread's message_id. A comment inside a thread has no thread: its own chat_id is the thread chat, and its entity_id is the thread id, so pachca_get_thread by entity_id gives the thread's message_id. If the message has no thread (thread is null and its entity_type is not thread), say so and do not create one. Read the thread chat by its chat_id, not the parent chat.

1. Read every message of the thread (pachca_list_chat_messages by the thread's chat_id, all pages, oldest first: pass sort=id together with order=asc on every call, cursor pages included, because order alone is ignored; or read newest first and reverse; check the order by created_at) and the parent message (pachca_get_message by the thread's message_id). A standalone thread has no parent: its message_id is null, so skip that call.

2. Resolve each author once (pachca_get_user) when the response does not name them. A 404 means the author was deleted or is hidden from you: refer to them by user_id, do not guess a name from the text, and keep their messages in the summary.

3. Answer in the user's language:

   • Context: what the thread is about, in two or three sentences;
   • Decisions: what was agreed, who agreed, link to the message;
   • Open questions: what is still unanswered and who should answer;
   • Action items: what, who, by when, if the thread says so.

   Later messages override earlier ones. Quote nothing you did not read.

   With a window, read the whole thread for context but summarize only what is new in it. «Since my last message» starts at the user's own last message in the thread (its user_id is yours). «Since last time» in the same session starts after the newest message your previous summary of this thread covered. With a question, end with a direct answer to it.

4. Remarks. When the question is about feedback («what needs fixing», «what remarks are there», «go through the reviews»), the answer is a numbered list rather than a retelling, so the user can pick by number. For each remark: what is wrong and where (a screen, a page, a method), who said it, how it should be, the message link and its attachments. Look at screenshots and files with pachca_download_file; a video you cannot view is named as not viewed. A later message that withdraws or settles a remark closes it. When this session has the code or the task the thread is about, check each remark against the code and its history (git log) and mark it done, to fix or a separate task, saying what you checked; a remark you could not check is marked so. «Make sure everything has been fixed» is this list with every remark checked. End by asking which remarks to take.

Taking them is the user's next request, not part of this command: change the code as they ask, create tasks only in the tracker they name and through its own tools, and reply in the thread only when they ask, on the message the remark came from (parent_message_id), following the pachca-messaging skill. When the work is done, offer one reply in the thread item by item: what was fixed, where to check it (a build, a page, a link), what was put off and why; show the text and post after a yes.

Do not post the summary unless the user asks. If they ask, show the text first and post after a clear yes, following the pachca-messaging skill. To turn action items into reminders, the user asks for that, and you follow the reminders part of the pachca-search skill.