Editing and Fixing the Graph
Graphs get messy. Titles have typos, links point the wrong way, nodes need status updates. Here is how to fix them.
Decision graphs are living documents. As work progresses, you will inevitably need to rename nodes, remove bad connections, update statuses, or delete nodes that should never have existed. Deciduous provides editing commands for all of these.
Claude is excellent at reconstructing connections, fixing broken links, and reorganizing nodes. If something gets messy — wrong links, duplicate nodes, missing connections — just tell Claude to clean it up. It will query the graph, identify the structural issues, and fix them. This is one of the easiest things to get right because Claude can see the entire graph structure and reason about what should connect to what. Think of it like asking Claude to refactor code — it just works.
Fixing a Title
The most common edit: a typo in a node title, or a title that no longer accurately describes the decision.
The node ID stays the same. All edges pointing to or from node 61 remain intact. Only the title changes.
Removing a Wrong Link
Sometimes Claude creates an edge that doesn't make sense—linking the wrong nodes together, or duplicating a connection. The unlink command removes it.
The canonical flow is goal → option → decision. A direct link from goal #57 to decision #60 bypasses option #58, which misrepresents how the decision was reached. Removing the shortcut edge restores the correct chain.
Deleting a Node
For nodes that should not exist at all—duplicates, test nodes, or nodes created by mistake—use delete:
Deleting a node also removes all edges connected to it—both incoming and outgoing. If the node is in the middle of a chain, this will break the chain. Make sure to re-link the remaining nodes after deleting a mid-chain node, or use edit instead if you just need to fix the title.
Updating Stored Prompts
If the stored prompt on a goal node is wrong or incomplete, update it with deciduous prompt:
Prompts are stored alongside the node and are visible in the web viewer's detail panel. They are used by the /recover skill to rebuild context after session loss.
Updating Node Status
As work progresses, nodes move through statuses. The status command updates them:
The full set of statuses and when to use each:
| Status | Meaning | Issue #93 Example |
|---|---|---|
pending |
Not yet started or resolved | A goal or action waiting for work |
active |
Currently in progress or current truth | Decision #60 while implementation is underway |
completed |
Done and working | Action #61 after the workflow passes tests |
superseded |
Replaced by a newer approach | An old CI approach replaced by a better one |
abandoned |
Started but given up on | An approach that was partially implemented then dropped |
rejected |
Considered but not chosen | Option #59: single workflow approach |
A Typical Editing Session
Here is what cleaning up a graph looks like in practice. Claude usually does this unprompted before syncing or exporting, but you can ask for it explicitly:
Dangling outcomes: 0
Dangling outcomes: 0
Run deciduous pulse before exporting or syncing. It catches orphan nodes (no connections), dangling outcomes (outcomes with no parent action), and other structural issues. Cleaning these up before sharing ensures teammates get a coherent graph, not a pile of disconnected nodes.
With the graph clean and statuses up to date, you're ready to share your work. The next section covers exporting the graph for GitHub Pages, generating PR writeups, and syncing with teammates.