Loom
In plain words: Workflow Viewer
Guide to the read-only Workflow Viewer (Loom) at /loom: open one exact Workflow Version by id and revision and inspect its stations and allowed transitions.
- 1Links must name both pattern id and revision, e.g. /loom/bridge_chat/2.
- 2Shows declared paths only; compare with the Run Inspector for what actually ran.
- 3Editing, publishing and execution from the viewer are future design.
Partial Read-only viewer works; editor is planned.
Words on this page, decoded click one for its card
Workflow Viewer Loom
Shows the exact workflow version a job used, step by step.
It's like: Viewing the recipe card that was actually used for your order.
Click a dot to open that part's card.
Open LoomWorkflow Viewer when you want to understand what a workflow permits. It presents one exact PatternWorkflow revision as a declared ScrollWorkflow Version: its stations and the routes between them. Once you can read that score, compare it with a particular RunJob Record's evidence in Orb. LoomWorkflow Viewer is currently read-only.
Find the exact score
After Altar setup, open http://127.0.0.1:7134/loom. The catalogue retains bridge_chat@1 through bridge_chat@4 and
delegated_rite@1 and delegated_rite@2; the rail pairs each title with its pattern_id@revision and marks the default,
other active routes, and retained revisions.
Opening /loom selects the configured default and replaces the URL with its exact route. Without
weaver.bridge, that is /loom/bridge_chat/1; configuring the
Bridge capability selector
selects /loom/bridge_chat/2; the
public-note configuration selects
/loom/bridge_chat/3. The Scout public-page profile
selects /loom/bridge_chat/4. An exact revision link continues to open that retained score.
A direct link takes the form /loom/{pattern_id}/{revision}. Both parts are required: /loom/{pattern_id} is rejected. This lets a review name the revision it concerns instead of leaving its identity to a later selection.
Read identity, then possibility
Read the views together:
| View | What to inspect |
|---|---|
| Header | title, description, pattern_id@revision, declared entry, and Registered from source. A durable publication store is not implemented. |
| Pattern graph | Pan and zoom through the declared score. Whole PatternWorkflow folds it into one group; Stations reveals the declared stations and permissions. |
| Graph outline | Station and permission counts; each node's label, key, kind, declared entry and permitted successors. Select a station to inspect it. |
| Selected station | Incoming and outgoing permissions with endpoint labels, keys and the declared relation; self-loops are explicit. Back to selected station returns keyboard focus to its row. |
| Identity & source | Expand the disclosure for checkpoint, declared entry, reviewed implementation revision, 64-character digest, trigger, and plain-text Mermaid source. |
The workflow canvas and semantic outline derive the same stations and permissions, including declared wait loops, from the exact manifest. Zoom changes whole-Pattern and station detail; explicit buttons offer the same operation. The shared Svelte Flow renderer supplies a read-only computed layout. It subscribes to no execution and highlights no live station. A RunJob Record's retained evidence in OrbRun Inspector establishes which movements were observed.
Hover or keyboard focus on a canvas node reveals its meaning and identity without selecting it. Escape closes that preview. The same graph adapter serves OrbRun Inspector; a plain canvas background keeps the score readable, while declared kinds remain distinct from inferred agent or tool identities.
The selected station and its outline remain usable if the graph fails. No outgoing permission declared describes the manifest; it does not infer that execution has terminated. A failed score read offers Retry reading PatternWorkflow, preserving the requested exact revision instead of silently selecting another one.
Read a station's kind as a declaration in the current manifest. The accepted grammar also gives
a SpellStep its own identity and places that exact semantic-action contract in a Scroll-local
station. Current manifests do not yet carry that independent SpellStep identity. Thus a delegate
glyph identifies a station kind; it cannot establish an independent SpellStep contract or prove
provider-backed execution.
Compare it with a Run
AnA InvocationJob Request is the admitted CircleTask Bubble; casting performs its ScrollWorkflow Version, and the RunJob Record is its durable ledger identity. Compare the declaration in LoomWorkflow Viewer with Orb's retained evidence of traversal, accounting for any gaps and omissions there.
OrbRun Inspector can attach ?run={run_id} when it opens the score. This supplies the RunJob Record you are reviewing;
the path still selects the PatternWorkflow identity. Before offering Return to RunJob Record in OrbRun Inspector, LoomWorkflow Viewer
validates the RunJob Record's pinned manifest and requires it to equal the entire registered score. Matching
only digest, PatternWorkflow ID, and revision is insufficient. It also checks that the separately fetched
displayed score still has the validated digest.
A valid return preserves the selected event or delegated-job hint. Invalid, unavailable, or mismatched RunJob Record context remains explicit. Choosing another PatternWorkflow discards the RunJob Record query.
Compare the source contract
The revision endpoint is /api/v1/loom/source/patterns/{pattern_id}/{revision}. The
current-revision shortcut is /api/v1/loom/source/workflows/{workflow}. Their separate source
namespace keeps /api/v1/loom/{pattern_id}/{revision} legal, including a revision named source.
The fixed trusted source-browser journey
checks the exact bridge_chat@2 score and keyboard return to its RunJob Record. It does not qualify the
planned editor, every graph interaction or the shared renderer's cross-engine lifecycle gate.
The planned editor
Editing, proposal, publication, execution, and layout changes belong to the future design. The Loom workflow views record tracks which parts of that design have been implemented.
Future charcoal drafts and grey SpellStep placements would first produce inert candidates and redacted
resolution reports. Until every exact placement passes Scroll admission,
the whole candidate remains castable=false; no placeholder enters the executable GraphStep Map.
Follow ADR 15's editing gate for the future editor and Spellweaver for a candidate's passage to a published revision.
Reading direction
For interface review and design, use the following composition and acceptance case.
The exact revision, source registration, entry, station counts, and declared successors should be legible before the reader opens digest or source detail. The computed graph has an outline and station inspector beside or below it as space permits. That outline must answer the same question without pan or zoom gestures.
Judge every mark against the manifest. Both views share declared nodes and permission endpoints; station detail exposes incoming and outgoing permissions. Placement cannot supply a conditional edge label, station effect, independent SpellStep identity, or execution colour. Handles must not imply connecting, editing, publishing, or running this read-only score.
Acceptance case: open a retained exact revision with RunJob Record context, find all permitted successors and a return loop using the keyboard, then fail the graph. The outline must still answer the same question and preserve selection. Return to the original OrbRun Inspector event; selecting a different PatternWorkflow discards that context. Frontend owns this target and the shared renderer gates; installation does not establish the required cross-engine lifecycle receipt. State keeps that qualification separate from the available canvas.