Follow-up to #2001, which covered only the primary fix.
Problem
find_citations (src/jev/citations.rs) scans the whole state text, fenced code blocks and inline code spans included. A quoted error message or example such as owner/repo#123 or o/r#1 then becomes a citation. Since #2001 this no longer aborts the run, but it still:
- sends a pointless
gh api graphql lookup, and
- adds noise to
route's reference_fetch_failures. The live omni-dev ai jev route 2000 now reports owner/repo#123, o/r#1 and o/r#2 as failures, all quoted examples.
Proposal
Have find_citations skip fenced code blocks (``` / ~~~) and inline code spans.
Care needed
verify-decision also uses find_citations, both for the comment's citations and in resolve_cites_index. A decision comment that deliberately puts a real reference in backticks (for example `#1614`) would stop being checked against its source. Decide whether both commands skip code, or only route. Pin the choice with tests on both sides.
Follow-up to #2001, which covered only the primary fix.
Problem
find_citations(src/jev/citations.rs) scans the whole state text, fenced code blocks and inline code spans included. A quoted error message or example such asowner/repo#123oro/r#1then becomes a citation. Since #2001 this no longer aborts the run, but it still:gh api graphqllookup, androute'sreference_fetch_failures. The liveomni-dev ai jev route 2000now reportsowner/repo#123,o/r#1ando/r#2as failures, all quoted examples.Proposal
Have
find_citationsskip fenced code blocks (```/~~~) and inline code spans.Care needed
verify-decisionalso usesfind_citations, both for the comment's citations and inresolve_cites_index. A decision comment that deliberately puts a real reference in backticks (for example`#1614`) would stop being checked against its source. Decide whether both commands skip code, or onlyroute. Pin the choice with tests on both sides.