Repository navigation
Releases
Releases · productweaver/productweaver-plugin
v0.17.2
What's fixed
- No more maps in half Italian, half English. In a study in another
language, the map could show text such as "Copia the intervista guide":
the assistant had swapped single words instead of writing each label in
the language, and the plugin kept it. Nowkb_tool labels --addrefuses a
label that still reads as English and says which one. Labels saved before
that read as English show in plain English and count as missing, so your
assistant writes them again in your language the next time you work on the
study.
v0.17.1
What's new
- Which version do I have? Ask "Which version of Product Weaver do I
have?" or saypw-about. The newpw-aboutskill shows the version, the
folder the plugin is loaded from, your assistant and model, your system and
Python, the study in the folder and whether usage statistics are on. Add
"for a bug report" to get a block to paste: it hides your home folder and
login name, and holds no study content. - Am I up to date? Only when you ask,
pw-aboutchecks
productweaver.com for a newer version and says how to update. Nothing else
in it goes online.
v0.17.0
What's new
- Talk to someone again. When you hear again from someone you already
interviewed, a follow-up call, notes or an answer by email, your assistant
adds it to the same person, not to a new one. The map shows every
conversation with them ("Interactions with" their name) and what changed
between them. Emails and chats are kept as text in.pw-kb/transcripts/. - The plan can suggest one person to ask again. When a person is the best
way to settle an open question, the plan for your next interviews says
"Ask" and their name, and the interview kit gets a sheet for that
conversation.
What's improved
- Who uses a tool follows the latest conversation. If someone told you
in March that they use a tool and in June that they stopped, the study now
shows that they stopped.
What's fixed
- An open map no longer stays on "Updating" after a follow-up. A batch
that held only a follow-up conversation never counted as read, so the map
kept saying the study was being updated. The map now counts it, and while
your assistant is still writing the overview it says what is in so far
instead of "Updated just now".
Good to know
- Once a study holds a follow-up conversation or a planned follow-up,
everyone who works on it needs version 0.17.0. Older versions refuse such
a study and ask you to update. - Your assistant saves emails and chats without headers, addresses, phone
numbers or signatures, and the merge refuses a copy that still holds an
email address or a header line.
v0.16.0
What's new
- Fewer approval prompts, less waiting. Your assistant needs fewer
commands for each batch of interviews. Each command checks everything
before it changes your study, so the assistant runs it once, with no test
run first. In Claude Code, that means fewer questions asking you to
approve a command. - The map is up to date after every step. When your assistant adds
interviews, writes the Overview, groups the measures of success into value
themes or acts on one of your decisions, the map and the review sheet are
updated in the same step. A map you have open updates itself after each
one.
What's improved
- The Overview is refused less often. A sentence of the Overview that is
a little over its word limit, by a quarter at most, is now kept, and your
assistant gets a note to make it shorter. The statements about your
interviews keep their limits. - Your assistant starts the Overview from a form.
kb_tool overview --skeleton Agives the file to fill in, with the word limits of your
study, so the assistant gets its shape right the first time. A form saved
without filling it in is refused. - You hear how far a batch is. In a batch of 4 or more interviews, your
assistant tells you when half of them are read. When it starts a study, it
says how long the first batch takes: about 1.5 minutes per interview, plus
2, the same estimate an open map shows. - The guide for your first interviews stays until your study's first
Overview, so the kinds of people you planned to meet shape the first plan.
What's fixed
- Claude Code refused some commands. The assistant sometimes kept the
script's path in a shell variable, or ran it in a loop. Claude Code cannot
check such a command, so it asked you to approve it, or refused it. The
skills now tell the assistant to write each command out in full.
Good to know
- A map you keep open while your assistant works reloads twice for a batch:
after the new interviews, and after the Overview.
v0.15.0
What's new
- The discovery map updates itself. Leave it open while you work. A few
seconds after each run it reloads in place: the same view, zoom, chosen
goal and open panel. It waits while a window is open, you are printing or
you have text selected, and shows "New version ready · Show it" meanwhile.
A map in a background tab is already up to date when you come back to it. - It says when it changed. The title bar shows "Updated 5 minutes ago".
While your assistant reads a batch, it says "Updating" and about how long
is left (about 1.5 minutes per interview, plus 2). A copy of the map that
cannot update by itself, such as one shown inside a chat, says so. - It says what changed. After each update a short card opens right
under the time in the title bar and lists it, such as "Updated just now: 3
new conversations, 12 new measures of success, a new overview". If you
were away, it waits for you, and the tab's title starts with a dot. Click
the time whenever you like to see it again, right where you clicked; it
stays until you close it. - New tags. Each new conversation, measure of success, solution and
group of solutions has a blue New tag until you have seen it. A closed
step says how many new measures it holds.
What's improved
kb_tool context --batch Nsays how long a batch of N interviews takes,
so the assistant can tell you, and an open map shows it too.kb_tool mapwrites the map in one step (a temporary file, then a swap),
so an open map never reads half a file. It also writes
pw-output/.map-live/status.js, a small file git ignores, with no names
or quotes.
Good to know
- A map made by plugin 0.14 or earlier does not update by itself. Reload it
once after this update; from then on it does. The assistant tells you
when. - A study in another language than English gets about 30 new labels to
translate at the next run.
v0.14.2
What's fixed
- Solutions and Value themes sit in the middle of the window. On a wide
window they stuck to the left edge and left an empty band on the right.
All four views (Overview, Solutions, Value themes, Next interviews) now
share one centered column with the same edges. On a wide window Value
themes shows two themes side by side, not three, so the customers' words
have more room.
What's improved
- Solutions and Value themes read like the other views. The text under
each title is larger, "Stopped using" and "In no theme yet" have the same
heading size as the headings in Next interviews, and a study with no
solutions or no themes yet says so in the same box the Overview uses.
v0.14.1
What's new
- An interview kit to print. In Next interviews, "Print the interview kit"
gives a page that plans the round, then one sheet for each call. The plan
lists who to book, what to check before you book, the message to send and
how sure the study is now. Each call sheet has the words to start and end
the call, the questions in order, what to write down and why, lines for your
notes, and what to say if an answer is short. Choose "Save as PDF" in the
print window to keep it as a file: it is named after your study and the
date. When new conversations came in after the plan was written, the kit
says so at the top. When the planned calls back up no open gap, there is
nothing to print until your assistant writes a new plan.
What's fixed
- A title that promises what the round does. "Talk to 3 more people to back
up 15 thin spots" now counts only the thin spots the planned conversations
back up (7 in that study), not every open one. - No more ".," or ".." in "Still open". The needs in the list lose their
own full stops.
v0.14.0
What's new
- A discovery map that looks like a Mac app. The map now follows the look
of macOS and Safari. A short title bar holds the study's name, the views in
the middle and two round help buttons. The people list floats on the left,
search and zoom float over the map as glass bars, and the details of a box
slide in on the right. Text and cards stay solid and easy to read. The
glass turns solid when Reduce transparency is on, and dark mode follows your
Mac. - Colors that show progress. Blue is now the color of everything you click
and read. Green, orange and red only show where something stands. The bar on
each step and measure of success is red when no one talked about it yet,
orange when 1 or 2 people did, and green from 3 people. "How sure is this?"
goes from red to orange to green. In Next interviews, "Last round closed
... gaps" turns green when your calls filled them.
What's improved
- The study's numbers moved into the Overview. Conversations, quotes,
steps, measures of success and solutions now close the Overview's first
block, in big figures. Click the solutions to open them. The title bar
gets the room for the study's title. - Two letters for each person, such as AR for Amy Richardson, so a person
never looks like the letter of a core job. - Measures of success are now cards with a blue target, and "Other people"
among the solutions is indigo, not pink, so red only means "no one yet". - In a study in another language, 5 short texts about the colors are new. The
assistant translates them at the next run, as it does for any new label.
v0.13.2
What's improved
- A clearer first paragraph in the README. It now says in plain words
that Product Weaver is a plugin for AI assistants such as Claude,
ChatGPT/Codex and Copilot, and that it builds one knowledge base from your
interviews that grows with every batch. Nothing in the plugin's behaviour
changed.
v0.13.1
What's improved
- A README that shows the result first. The plugin's front page now opens
with a picture of the discovery map and what you get, then a 10-minute way
to try it with 6 sample interviews, so you can see it work before you use
your own. Install steps come next, one section per assistant, and the
steps for the Claude desktop app now follow its menus one by one. Under
"Can you trust it?" you find what runs on your computer and what is sent.
A short list of fixes for common problems follows. Nothing in the plugin's
behaviour changed.