Repository navigation
fix(builders): name trustshell 1.6.1, and read the checked version from the doc - #52
Conversation
…om the doc npm now serves @hyperdag/trustshell 1.6.1. check:builders hardcoded 1.6.0 twice, so it FAILS on main today (measured locally) and the blocking "BUILDERS.md matches published trustshell" job would turn the next PR in this repo red for a reason unrelated to it. main last ran CI on 10-12, before 1.6.1 shipped. - check-builders.mjs reads the version BUILDERS.md names (exactly one "PUBLISHED — `x.y.z`") and compares THAT with npm. A publish still turns CI red until the doc is updated, but the checker never needs editing again. - BUILDERS.md names 1.6.1. Re-measured on 1.6.1 2026-10-14: `trustshell verify` exits 0 / 1 / 2 for PASS / VETO / NOT_CHECKED, ethers is still an optional peer, the other packages still 404, the six-interface kernel still not on main, and the section 1a curl still returns pass. - section 1a documents the `votes` field the live classify reply now carries. Checker: VERIFIED as committed; FAILED when the doc names 1.6.0; FAILED when it names two versions. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Strix Security ReviewNo security issues found. Review summaryReviewed all five changed files. The PR is a documentation/version-bump plus a refactor of the CI checker Updated for Reviewed by Strix |
README told an outside builder to `npm i -g @hyperdag/trustshell@1.6.0` in two places, which installs the old release (no my_job, no kernel client) now that npm serves 1.6.1. The install commands are unpinned: a pinned install goes stale on every publish, which is how this happened. The measured version lives in BUILDERS.md, under check:builders. CLAUDE.md and CONTRIBUTING.md now say 1.6.1 (measured 2026-10-14). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Added a second commit with the same theme: |
Why now
npm serves
@hyperdag/trustshell1.6.1.scripts/check-builders.mjshardcoded1.6.0twice, so it fails on main today:That check is the blocking CI job "BUILDERS.md matches published trustshell". Main last ran CI on 10-12, before 1.6.1 shipped, so it still looks green. The next PR here would go red for a reason unrelated to it.
Fix
BUILDERS.mdnames (exactly onePUBLISHED — \x.y.z``) and compares that with npm. A publish still turns CI red until the doc is updated, which is the intended tripwire, but the checker never needs editing again. Hardcoding the version is what made it a code edit every release.votesfield the live classify reply now returns.Re-measured on 1.6.1, 2026-10-14
npm view: trustshell1.6.1;@hyperdag/protocol,identity-erc8004,reputation-zkpstill 404.trustshell verifyfrom a clean install: Paris → exit 0, an unverifiable private claim → exit 2, Eiffel/Rome → exit 1.ethers^6.14.0is still an optional peer (peerDependenciesMeta).packages/interfacesstill has no commits onmain.curl, verbatim:{"label":"pass","by":"votes","voters":["groq","cerebras"],"deciders":[…],"votes":[{"voter":"groq","family":"gpt-oss","verdict":"TRUE"},…]}.Checker evidence
VERIFIED as committed · FAILED when the doc names 1.6.0 · FAILED when it names two versions.
🤖 Generated with Claude Code