Saved workflows: arguments, launcher, Settings page and /workflow (stacked on #719) - #724
Open
katulevskiy wants to merge 50 commits into
Open
katulevskiy wants to merge 50 commits into
katulevskiy wants to merge 50 commits into
Conversation
Claude TodoWrite, OpenCode, ACP plans and Codex update_plan expose an in-progress step that TodoItem flattened into done/not-done. Add a serde- defaulted status (written only for in-progress, so existing docs are byte- identical and old builds/edge ignore it; unknown values derive from done), map it in every normalizer, and wire Codex turn/plan/updated into the live plan chip.
The agent's checklist is now a tray stacked above the queue/composer: a collapsed header (Todo 2/5 plus the current item), an expanded list with check / active / pending glyphs, and a 3-item focus window with N earlier / N later fold rows for lists over 6. Open state is remembered per chat for the app run, a finished list tidies itself to a compact done state once the turn is idle, and every control is keyboard reachable with tooltips. The transcript chip's detail marks the in-progress item [~].
ZERON_MOCK_TODO walks an 8-item list through every panel state for live checks; docs/todo-panel.md records the data model, behaviour and limits.
Child ask (engine::ask): hidden child chat under the requester, run-scoped submit_result validated against a caller-supplied JSON Schema (boon), repair rounds, nudge, typed failures, cancellation and archive-on-exit, plus a FakeAsk backend. Goal mode: additive session-doc goal state, goal commands on the command plane, machine-origin queue/message origin, pure state machine and prompts, doc-host controller with restart recovery, and tests.
The expanded goal list was cut off by the todo tray tucked over it. Every expanded list now keeps a bottom clearance equal to the tuck overlap (the todo tray too, a one-line change in todo_panel.rs), the goal list's height share shrinks as trays stack below it, and it follows the newest round. Retook the goal-mode screenshots and added a stacked-trays one.
The in-process verifier entry was dropped when the verifier returned, but the verdict was applied only after taking the controller lock. A tick (turn end, queue watcher, status change) in that window saw 'verifying' with no verifier and judged the same round again. Reserve the entry under the lock together with the Verifying ledger write, release it under the lock after the verdict is applied, and make a second start for the same (goal, round) a no-op.
…erInputQuestion.meta
…ysis and host seam
…PC and command plane
… reason in the end marker
…g; steadier tests; demo host ignores workflow commands
…r run lines, result row
…d), docs/workflows-ui.md, fmt
…der, no env reads per sync, no run clone per state change
…ts, doc touch-ups
…ples The frontmatter parser/writer lives in zeron-workflow (strict, line:col errors, all collected); types and the argument rules (validate_args, parse_text, slug names) live in zeron-proto so every client shares them. Three built-in workflows (pr-review, fix-until-green, repo-audit) run against the fake host in tests.
… with args, run history)
Saved files are scanned on demand from the project, the global folder and the
built-ins, with shadowing and per-file errors. start_workflow accepts saved
{name, scope, args}: arguments are validated before approval, the approval shows
the saved name and values, and runs record saved_name/saved_scope in their
header. Saving asks through the existing input-question mechanism and writes
atomically inside the workflows folder only. WorkflowSaved* RPCs are forwardable;
capability workflows-saved-v1.
…ved} MCP tools, guide section The tools steer the model (prefer a saved workflow, save only with the user's agreement); save always asks the user; an agent can never pass byUser. The guide gains a Saved workflows section whose frontmatter example runs against the fake host in tests.
…d args form, /workflow, Save as workflow and Run again Settings → Workflows lists Global, per-project and built-in workflows (rescanned on demand) with a detail view: arguments table, highlighted script, recent runs, Run / Open file / Delete. The launcher turns a workflow's declared arguments into typed fields (defaults prefilled, required markers, inline errors) and starts the run in the open chat or a new chat in a project. /workflow lists saved workflows in the slash popup and starts one with key=value arguments, opening the launcher when a required argument is missing. Run cards show the saved name; ended runs offer Run again (same arguments) or Save as workflow…. The approval block names the saved workflow and its argument values.
… apart from the provider catalog docs/workflows.md (saved workflows), workflows-ui.md (Settings, launcher, /workflow, card actions), mcp.md; nine screenshots from the live app. The composer fetches the saved workflows independently of the provider's command catalog, run lists trust a settled run's store status over a lagging doc projection, and run history says how long ago.
…s in the new files
katulevskiy
force-pushed
the
workflows-saved
branch
from
October 2, 2026 00:04
797b0ee to
a3a34a8
Compare
Contributor
Author
|
Force-pushed: rebased the stack onto current |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stacked on #719 (workflows card), which is stacked on #712, #708 and #707. Until those merge, this PR's diff also contains their commits; review
git log workflows-card..workflows-saved(6 commits). It is independent of #723 (mobile), which is also stacked on #719.Summary
Saved (reusable) workflows with typed arguments. A saved workflow is a Starlark file with a frontmatter comment block, kept in
<project>/.zeron/workflows,~/.zeron/workflowsor built in. Agents list, start and save them through MCP; people use Settings → Workflows, a launcher with a generated args form,/workflow, and "Run again" / "Save as workflow…" on the cards.Screenshots
Captured from the real desktop app on an isolated fixture against the mock harness (scripted agents, not real models). The Settings paths are the fixture's temp directories.
Settings → Workflows: Global, per-project (with an override and an unusable file) and Built-in
Detail view: description, arguments, highlighted script, recent runs
Recent runs of one workflow
Launcher: generated args form with a validation error
/workflowin the composerApproval for a saved workflow, showing the name and argument values
"Save as workflow…"
Card header with the saved chip and Run again, and an ad-hoc run with Save as workflow…
Design
name,description,when_to_use,argswithstring | int | number | bool | json,required,default,description). Every error carriesline:coland all errors are reported at once; hostile input (huge files, duplicate keys, path traversal in names) is covered by tests. The parser lives inzeron-workflow; the shared types and argument rules (validate_args, slug names) live inzeron-protoso engine, desktop and mobile share one contract without linking the interpreter.start_workflow {saved: {name, scope?, args}}validates arguments before approval, and the approval shows the saved name, scope and values.save_workflowalways asks (stock labelsSave workflow/Deny), writes atomically inside the workflows folder only, and refuses to overwrite unless the question said it would.list_saved_workflowslists them. New RPCsWorkflowSavedList/Get/Save/Delete/Runsare forwardable, so a client addressing a chat's host lists the host's workflows (capabilityworkflows-saved-v1).savedName/savedScopein the header,WorkflowGetreturns the run'sargsandsaved, and resume still replays the stored copy, so editing the file afterwards doesn't matter./workflowor Run again is the approval (there is no live agent turn to carry the question), so the launcher shows exactly what the approval would. The engine honours thatbyUsershortcut only together withsaved, and the MCP tools never send it.pr-review,fix-until-green,repo-audit, each run against the fake host in tests; the guide's frontmatter example is also executed in a test.docs/workflows.md,docs/workflows-ui.md,docs/workflow-guide.md,docs/mcp.md.Testing
zeron-workflow97,zeron-proto88,zeron-mcp40,zeron-rpc16,zeron-enginelib 450 (2 ignored),zeron-uilib 1523.workflows21,workflow_e2e2,workflows_saved21 (new),ask_child16.cargo check --workspace --all-targetsclean; the new files are clippy-clean (older warnings elsewhere, e.g. incomposer.rs, are untouched).Not verified
list()prefers the store's status for a settled run when the doc projection lags behind (no dedicated test), and saved-workflow rows in the composer load apart from the provider catalog so a slow catalog can't leave/workflowon a skeleton.