Repository Issue Activity (beta)

#244 The markup refusal's only recovery instruction is the shape that caused the fold, so callers abandon the write tools instead of retrying them

#252 Stop reports unchanged vault hygiene after every response

#201 Intake and other semantic-search commands hang when QMD vectors are unavailable

#250 Choose the embedding model instead of inheriting qmd's default, which costs 17 rank-1 points

#249 The hyde sub-query is a duplicate vector search, so question-shaped queries pay 35-40ms for a result that never changes

#248 Q: Are there `npx` AIO installs for this?

#247 The template has no prose-width rule, so every vault it creates teaches sessions to hard wrap

#246 Nothing asserts that a vault's promoted markers resolve, and splitting a note orphans them

#237 slugify keeps # in a memory filename, so the note it writes cannot be addressed by any wikilink

#243 A stream error escaping mcp-qmd-client.test.ts fails the whole file, on a change that touches none of it

#240 A client that answers no roots/list can never identify itself, so it is served only general-scope memories forever

#235 Two Windows-only CI flakes: a cleanup failure fails a passing test, and a fixed sleep stands in for waiting on a condition

#238 The first search against a fresh vault times out while health reports no warnings, so a caller concludes the record is missing

#229 The release job cannot be re-run: a failure in its last step aborts at the commit step, which exits 1 with nothing to commit

#230 The markup refusal never says which fields the server received, so a caller whose field swallowed the next one cannot see the loss

#231 recall --explain names the supersession category but not which memory superseded the withheld one

#202 A question-shaped issue has no route: it idles as a pseudo-task or gets answered implicitly by an implementation

#225 A refused tool call leaves no audit trace, and a capture written outside the tool is served by nothing

#224 A memory whose title outruns the filename slug cannot be linked by its own title, because no alias is written

#211 The correction sweep is instructed but unsupported, so a corrected fact survives in every note that paraphrased it

#215 recall serves a retired claim when its correction is itself withheld, so a second narrowing re-opens the gap #199 closed

#216 Wikilink gate: a double-quoted alias with an escaped quote never matches, because the backslash escape is not undone

#213 Memory recall: a comma in a superseding title severs the correction link, because the inline-array parser splits on every comma

#199 recall serves a superseded memory where its correction cannot reach, so narrowing a scope never takes effect

#198 MCP write tools: the markup refusal names the field but not the failure, so the caller applies the wrong repair