Skip to content

feat(graph): follow GitHub sub-issue links - #54

Merged
Railly merged 6 commits into
mainfrom
railly/recreate-38-s1
Nov 14, 2026
Merged

Railly merged 6 commits into
mainfrom
railly/recreate-38-s1

Conversation

@Railly

@Railly Railly commented Nov 14, 2026 •

Copy link
Copy Markdown
Member

Problem

The crawl reads mentions, cross-references, connected events, and closing references, but not GitHub sub-issue links, so an epic seed does not reach sub-issues it never mentions and a sub-issue seed does not reach its parent or siblings.

Changes

  • The node query reads each issue's parent and its first 50 subIssues; they become parent-of and child-of edges beside any mention or cross-reference between the same issues.
  • The crawl follows sub-issue links from either end under the existing depth, budget, and hub rules.
  • The hub threshold counts distinct linked items, so an epic that also lists its sub-issues in a task list counts each one once.
  • Sub-issue links are structure: they add no Heat, are never closing links, and an unfetched or capped sub-issue does not block its parent in plan.
  • Reports label the edges parent of and child of; the dashboard draws one parent-to-child arrow and counts sub-issues as connections in Impact, links, and blast.
  • Parents past the 50 sub-issue limit carry subIssuesPastLimit, and sub-issues GitHub returns without data are counted in subIssuesUnreadable.
  • Docs and the bundled skill describe the new edges, limits, and hub counting.

Replaces #38.

Visual proof

A synthetic acme/roadmap epic with two sub-issues that never mention it. Before, the crawl stops at the epic:

Before

After, it follows the sub-issue links both ways:

After

The 32 new tests fail 24 of 32 on base f2fa275 and pass on ff4e8ea.

Railly and others added 5 commits November 13, 2026 21:28
Parsing of the parent and sub-issue fields, crawling from either end
of a link, the hub guard, rank, mentionedBy, the per-node limit and
unreadable sub-issues, and how the report, terminal view, cluster
prompt and dashboard show the link. A mention between two linked
issues keeps its own edge, inbound reference and author.

Co-Authored-By: Lars Buur <3585707+LarsBuur@users.noreply.github.com>
An issue attached to an epic only through a GitHub sub-issue link was
missing from the graph, because the node query never read the
sub-issue fields.

The query now reads each issue's parent and its first 50 sub-issues.
The parent gets a `parent-of` edge to each readable sub-issue and the
sub-issue a `child-of` edge to its parent, so the crawl follows the
link from either end under the usual depth, budget and hub rules.
These edges sit beside any mention, cross-reference or connection
between the same two issues instead of replacing it, so linking two
issues keeps the reference in rank and in mentionedBy. The links add
no inbound references and never count as closing links.

The report labels the edges `parent of` and `child of`, and the
dashboard draws one arrow from parent to sub-issue. A parent past the
50 limit records `subIssuesPastLimit`; sub-issues GitHub returns
without data are counted in `subIssuesUnreadable`.

Co-Authored-By: Lars Buur <3585707+LarsBuur@users.noreply.github.com>
The dashboard model puts sub-issue links in each node's out and in
lists, so neighbors() counts them: Impact lists an epic's open
sub-issues as follow-ups, and the Swarm links and blast metrics count
them, as they already do for Linear parent and sub-issue relations.
The docs only said the links add no inbound references.

Document the behavior in the graph and dashboard pages and the bundled
skill, and pin neighbors, follow-ups and Impact scores in a test.

Co-Authored-By: Lars Buur <3585707+LarsBuur@users.noreply.github.com>
A sub-issue link and a mention of the same issue are two edges, so an
epic that also lists its sub-issues in a task list counted each one
twice and became a hub at half the threshold, leaving its sub-issues
unfetched. The hub guard, the hub line in the report, and the
component hub in Markdown and the dashboard now count each linked item
once.

Co-Authored-By: Lars Buur <3585707+LarsBuur@users.noreply.github.com>
The plan treated every edge to a failed or capped-out node as an
unresolved reference, so a parent whose sub-issue fell past the node
cap, or a sub-issue whose parent failed to fetch, moved to the blocked
lane with readiness 0. A sub-issue link is structure, not a reference;
only reference edges now block on an unresolved target.

Co-Authored-By: Lars Buur <3585707+LarsBuur@users.noreply.github.com>
@vercel

vercel Bot commented Nov 14, 2026 •

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
issue-graph Ready Ready Preview, v0 Nov 14, 2026 4:20am UTC

Comment thread src/github.ts Resolved

@lahfir lahfir left a comment •

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM 🔥

A new sub-issue link to a node that was already referenced diffed as
"no changes" because the previous-edge set held targets only. Key it by
`via` and target so a `parent-of` or `child-of` link beside a mention is
reported as a new link.
@Railly
Railly merged commit 2a3ba85 into main Nov 14, 2026
13 checks passed
@Railly
Railly deleted the railly/recreate-38-s1 branch November 14, 2026 05:04

This branch was successfully deployed

1 active deployment
Preview — 94bd73c3 Deployed Nov 14, 2026 by vercel[bot]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants