Tooling and documentation
The product is Clarua, built with heavy use of coding agents (share not yet measured). The rule that started ClaimCheck applies here too: every agent report says what it did not establish.
Merged pull requests, all time
Merged in September 2026
Merged in October 2026 (to Oct 23)
Source: GitHub search on the product repo (private), Oct 23 2026 ~1:50pm CT. PR numbers below are from the same repo.
Agent tooling: recent merged work
| PR | Merged | Change |
|---|---|---|
| #8847 | 2026-10-15 | /code: project memory carried between sessions, and /memory to see and forget it |
| #8844 | 2026-10-15 | /code: review_rules applies the repository's .supacode/review.md to /review |
| #8854 | 2026-10-15 | /code loop: retry a context overflow after compacting, bound a quiet stream, report a missing stop reason |
| #8849 | 2026-10-15 | Landing-task autofix: push to the PR's own branch, one fix per commit |
| #8850 | 2026-10-15 | Measure a 10x-cheaper model before building one |
| #8859 | 2026-10-16 | harnesses: newbot and supabot-voice measure the product as it is |
| #8773 | 2026-10-01 | Stop /code and Supa Bot runs whose tool calls keep failing |
| #8731 | 2026-10-01 | Supa Bot: one in-turn correction round when the record contradicts a report |
| #8734 | 2026-10-01 | Stop calling a gappy crawl complete |
Documentation: docs-only PRs
58 merged PRs have "docs" in the title. A pattern in them: docs that correct earlier claims with measured numbers.
| PR | Merged | Change |
|---|---|---|
| #8842 | 2026-10-15 | docs(web-capture): record the call-scan sweep, tsc and build |
| #8648 | 2026-09-26 | docs: quote router tiers from TIER_MODEL |
| #8463 | 2026-09-24 | docs(api): name the public docs page Clarua |
| #8112 | 2026-09-13 | docs(supa-intelligence): acceptance specification |
| #8107 | 2026-09-13 | docs(lattice): correct the answer-path authorization reading, with measured counts |
| #8085 | 2026-09-13 | docs(ci): correct what the concurrency change actually bought |
Watching
I read GitHub trending daily for anything that overlaps with ClaimCheck. Closest so far: verifier patterns that try to disprove each finding before accepting it, and labeling each fact as extracted or inferred. Neither checks whether an agent's "done" claim matches its logs.