Repository navigation
Releases
GlanceFlow 0.34.0
The Desktop app's task list keeps the job's goal, and other chats stay private
Outcome in the Desktop app. When Claude plans with the Desktop app's own task list, GlanceFlow follows that list instead of asking for a second plan. Until now that also meant the job lost its outcome: ▤ Plan showed the steps but no Outcome, Left out or Done when. Claude now gives them in its first progress report, so the Plan shows them again:
Outcome: A landing page with a sign-up form
Left out of this job
• The pricing page
Done when: The form works on a phone
Other chats as a count. The line about other chats on your computer used to name them, from any project, so other work turned up in screenshots and shared screens. Now it only says how many:
Another chat needs you · 1 more at work
To see which chats, pick With names under Other chats in ⚙ Settings, or run /glanceflow others names. /glanceflow others off hides the line.
Update
claude plugin marketplace update claude-mods
claude plugin update glanceflow@claude-mods
Then open a new chat.
GlanceFlow 0.33.5
Needs you shows Claude's question, and answering it keeps the job
The question, not an old permission. When Claude asked you something right after a permission request, the checklist kept saying the permission ("OK to list project folder contents?") and no notice came. Now the Needs you line shows Claude's question, and the chime and desktop notice come again with it:
Claude needs you
Who is the launch post for?
Answering keeps the same job. In the Desktop app, when Claude finishes with a question in its closing box and you click one of its buttons (or type the answer), the same job finishes under its own title, instead of a new job taking over All done. A reply to a plain question at the end of a finished job ("Do you want me to deploy it?") still starts a new job.
Update
claude plugin marketplace update claude-mods
claude plugin update glanceflow@claude-mods
Then open a new chat.
GlanceFlow 0.33.4
Workflow Kit's specialists sit under their Director in the Desktop app too
In the Desktop app's ▤ Plan, a researcher, coder or critic that Workflow Kit's Director hands work to now sits one step in under the Director's row, the same as in the terminal and in the band above the prompt. Before, it showed in the right place but lined up with the Director.
GlanceFlow also no longer reads Workflow Kit's run log while the Director works. The Kit's current engine only writes when a run ends, so that never showed anything; the Director's work orders show through its specialists instead, as they did already.
Update
claude plugin marketplace update claude-mods
claude plugin update glanceflow@claude-mods
Then open a new chat.
GlanceFlow 0.33.3
Workflow Kit's specialists sit under their Director
When Workflow Kit's Director hands work to a researcher, coder or critic, that specialist now shows right under the Director's row, one step in, instead of further down the list after any other helper:
↳ ✓ gf-livecheck-3 Director live check · agency-director
↳ ✓ Report README first heading · researcher
It stays with the Director's step too, even when the plan has moved on. The same holds in ▤ Plan.
Work order ids keep their case. A name that starts with an id like gf-livecheck-3 or task_12 is no longer turned into "Gf-livecheck-3". Other names still start with a capital.
Update
claude plugin marketplace update claude-mods
claude plugin update glanceflow@claude-mods
Then open a new chat.
GlanceFlow 0.33.2
Running the script Claude wrote counts as tested
When a step changes code, a finished job says whether it was checked. Until now only the project's tests, build or type check counted, so a job where Claude wrote a script and ran it to make sure it works still ended "not tested".
Now Claude running a file it changed in the job counts too, through Python, Node, Bash and the like (python3 hello.py) or as ./run.sh. If it runs cleanly the header says
✓ All done · Write the hello script · tested · took 40s
and if it fails, tests failed. Running some other file, or only reading the changed one, still checks nothing.
Update
claude plugin marketplace update claude-mods
claude plugin update glanceflow@claude-mods
Then open a new chat.
GlanceFlow 0.33.1
Workflow Kit's specialists show, and saved ideas stay whole
You see who the Director handed the work to. When Workflow Kit's Director runs, each specialist it starts (researcher, coder, critic) now shows as a helper under the current step, named by its role. Until now you only saw the Director itself, because it hands most work straight to its specialists rather than through its run engine.
Saved for later keeps the whole idea. A suggested next prompt parked during a job used to lose its file names, so
Commit notes.md and hello.py
was saved as "Commit and". It's now saved word for word, with private details still hidden.
Update
claude plugin marketplace update claude-mods
claude plugin update glanceflow@claude-mods
Then open a new chat.
GlanceFlow 0.33.0
See how many jobs stayed on plan
GlanceFlow 0.32.0 keeps each job to one outcome. Now History shows whether that's working.
Each job GlanceFlow saves keeps the size of its first plan. The History's total line for the day says how many jobs ended with more steps than they started with:
1 of 2 grew past their first plan
/glanceflow week says the same for the whole week. Quick answers aren't counted, and neither are jobs saved before this version, so the numbers fill in over the next few days.
Update
claude plugin marketplace update claude-mods
claude plugin update glanceflow@claude-mods
Then open a new chat.
GlanceFlow 0.32.0
One goal per job, side ideas saved for later
A job used to keep growing: a suggested next prompt or one of Claude's follow-up ideas started new work before the first job was done. Now each job keeps to one result.
The result first. Before it starts, Claude says the one result the job delivers and what it leaves out, and asks anything only you can settle, all at once. Then it works without stopping unless it's blocked. ▤ Plan shows it:
Outcome: A contact form that emails the team
Left out of this job
Side ideas wait. Ideas outside that result, Claude's own and the suggested next prompts Claude Code or another mod would put in the prompt box, go under Saved for later in the Plan instead of into the work. When the job finishes, the checklist says so:
1 idea saved for later, under Plan
Claude stops when the job's Done when is met. After a quick answer, or once a finished job folds away, suggestions show as usual.
Workflow Kit. In a project that plans with Workflow Kit, the Kit's plan is the job's one plan, not a second one.
Update
claude plugin marketplace update claude-mods
claude plugin update glanceflow@claude-mods
Then open a new chat.
GlanceFlow 0.31.0
The checklist keeps up with Workflow Kit, agent teams and your other mods
Workflow Kit's work orders show up. When Workflow Kit's Director runs a job, each work order that has started appears under the current step as a helper, with its name (Build form), its role (coder) and whether it is running, done or failed. Claude's own plan stays as it is.
The checklist tells the truth beside other plugins.
- When another plugin's check sends Claude back to work (a test gate, say), All done goes away again until the job really ends.
- When another mod waits on you in its own panel (Blast Radius holding a command), the checklist says Needs you:
Another mod is waiting for your answer in its panel
- When something else compacts the chat (Claude Code itself,
/compact, another mod), Claude writes its checkpoint first, so the goal, decisions and next step survive. - A helper or teammate working out of sight no longer makes the job look stuck, and a team lead can message teammates before a plan.
Update
claude plugin marketplace update claude-mods
claude plugin update glanceflow@claude-mods
Then open a new chat.
GlanceFlow 0.30.5
Works more smoothly beside other plugins
GlanceFlow asks Claude to lay out a plan before it uses a tool. That check got in the way of some setups, so it's lighter now.
Looking around goes through. Reading files, searching, loading a skill, and the memory lookups mnemo and claude-mem ask for at the start no longer wait for a plan. Changing anything, such as editing a file or running a command, still waits.
Notices don't start a new job. When a background task finishes or another session sends a message, Claude carries on with the plan it already has instead of starting over, so its next step isn't refused.
Named agents aren't held up. A chat started with claude --agent <name> sets its own tools, which may not include GlanceFlow's planning tool. Such a chat now runs without waiting for a plan. Headless runs (claude -p, the Agent SDK) also stop receiving checklist instructions nobody sees.
Update
claude plugin marketplace update claude-mods
claude plugin update glanceflow@claude-mods
Then open a new chat.