Sloth-ninja makes Clio matter search stop guessing

The fork tightens how its legal-workflow assistant handles missing data, page-by-page results, and internal failures.

integrationworkflow

A missing response from Clio, the practice-management platform, now becomes an error instead of a confident claim that no matters were found. Clients still receive a plain, safe message, while the team gets enough internal detail to investigate what actually broke.

Search-result page links stay tied to the filters that created them, preventing a later page from widening into an unfiltered matter search.
Long searches are rejected before they can produce unusable follow-up links.
Expected chat-stream failures avoid noisy duplicate logs, while unexpected failures retain full diagnostics.

The changes are backed by repeatable tests around those edge cases.

So what Firms relying on matter search in client-facing workflows should care: this is the unglamorous work that keeps an assistant from turning an integration hiccup into a misleading answer.

View this fork on GitHub →

Spotted something wrong? Or know the PR text has fresher detail than the writeup above?

Commits in this thread

2 commits from Sloth-ninja/JessicaOSS, oldest first. Source extracted verbatim from the harvested git log.

SHA	Subject	Author	Date
`4913ee97`	Composed-range review fix wave for the #71/#72 train	sloth-ninja	2026-08-05	↗ GitHub
commit body Six review findings: Clio tool executors log original failures ([clio/tools], validation errors excluded); null Clio body throws instead of shaping into an authoritative empty matter list; the at-least-one-filter guard re-applied to decoded page tokens; query capped at 256 chars at token mint; stream routes log only name/message for the AssistantStreamError rethrow wrapper; CLAUDE.md safeError.ts module-map entry + status sentence. Addendum: all seven real-scrypt suites get a 120s file-level timeout (contention flakes; DURABLE_LESSONS 2026-08-05). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
`ad46a640`	Review micro-wave: honest oversize-token handling, sibling null-body guards, CLAUDE.md accuracy	sloth-ninja	2026-08-05	↗ GitHub
commit body The 256-char query cap only bounds the common case (reviewer measured minted tokens over the 1024 decode cap with CJK queries / long cursors): the mint site now drops any token its own decoder would refuse, keeping has_more true. findContact and listMatterDocuments get the same null-body guard as findMatter (deleteTimeEntry untouched - 204 is success there). CLAUDE.md: stale 604-test figure -> 656 (post-#73), '#71/#72 deployed' claim corrected to merged-only (deploy pending owner action), status header date bumped. Three new tests (suite 656). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Capture this thread into my fork

Download a single Markdown prompt that tells Claude how to port every commit above into your working tree — adapting paths and structure to match your repo. Run it via claude -p < capture-thread-1050.md from inside the repo you want the changes in.

⬇ Download capture-thread-1050.md