docs(jsr): fit package taglines in the 250-byte cap + gate it in CI

[rickylabs]

Summary

Sixteen published package descriptions are truncated mid-sentence on jsr.io. This fixes them and adds the gate that stops it recurring.

The trap

The JSR package description is not read from deno.json. It is derived from each package README's bold tagline — the first prose paragraph after the H1 and badge block — by .llm/tools/release/jsr-set-package-settings.ts, and PATCHed onto JSR after publish.

JSR validates that description in BYTES (Rust String::len), not characters, and caps it at 250. An em-dash costs 3 bytes. Over the cap, it is silently truncated at a word boundary.

Which is why the live descriptions read like this today:

@netscript/telemetry — "…and a telemetry query read model — all linking sched"

Cut mid-word. On the public registry page. For 16 packages.

What changed

Commit
952afca1	New gate: .llm/tools/validation/check-jsr-tagline-length.ts + deno task docs:tagline:check. It extracts the tagline exactly as the release tool does and measures it in bytes.
822453f6	16 over-cap taglines fitted.
458879fb	Gate wired blocking into the quality CI job.

Result: checked=35 over=0 (was over=16).

The technique is not truncation. The extractor stops at the first blank line, so each tagline was tightened into a complete, well-formed sentence inside the cap, with the surplus demoted to a second paragraph — which stays in the README for human readers and never reaches JSR.

plugins/ai (760 bytes, the worst offender) turned out to have no bold tagline at all — that is why extraction swallowed its entire 9-line lead paragraph. It got a real tagline; no claim was dropped.

The gate is wired blocking last, only after the branch makes it green. Wiring a gate that would immediately fail is theatre.

The JSR registry is NOT touched

jsr-set-package-settings.ts and jsr-provision-packages.ts were not run. No call that writes to jsr.io was made.

The descriptions on jsr.io will not change when this merges. They update at the next publish, when the settings step runs — an owner-supervised action. This PR only fixes the source the release tool reads.

Why it matters now

@netscript/mcp is a first publish in #715. This is the one chance to land its description whole rather than cut mid-sentence like its siblings. (packages/cli and packages/mcp were already fixed on that branch — they are the worked example.)

Validation

• deno task docs:tagline:check → checked=35 over=0
• deno task docs:links → 96 docs, 0 broken links
• deno fmt --check clean on every touched README

Harness

Tier-D Codex slice, reviewed by the Tier-A supervisor (opposite family). Run dir: .llm/runs/beta10-non-dashboard--claude/, slice jsr-tagline-byte-cap. Not self-certified.

Refs #715

🤖 Generated with Claude Code

[rickylabs]

⚠️ The green ticks on this page are a thin signal — do not read them as merge-safe

.github/workflows/ci.yml triggers only on PRs into main:

on:
  pull_request:
    branches: [main, "feat/package-quality"]

This PR targets feat/beta10-integration, so the full ci workflow — check-test and quality (check + lint + fmt:check) — does not run on it at all. Only surface-diff and code-quality execute. The "0 fail" is real, and nearly empty.

The merge evidence is the gates run locally against this branch, recorded in the PR body — not the ticks above.

Worth knowing across the beta.10 set: the gates these PRs newly make blocking (repo-drift in #772, tagline length in #771, the JSR specifier guard in #769's branch) all live in the quality job — which never executes on their own PRs. None of them has run in CI even once. They fire for the first time on the feat/beta10-integration → main PR. Expect that one to be the loud one, and treat it as the first honest CI verdict for this whole wave.

[rickylabs]

[PHASE: IMPL]

Reconciled PR #771 with the advanced feat/beta10-integration base.

Reconciliation

• Fetched and merged base at 3265b516 via merge commit a2cd2002.
• Resolved the sole conflict in .github/workflows/ci.yml by preserving both sides: docs(jsr): fit package taglines in the 250-byte cap + gate it in CI #771's blocking JSR tagline length step and the base's Generated asset freshness step, along with all other CI/ruleset updates already landed on base.
• Preserved all docs(jsr): fit package taglines in the 250-byte cap + gate it in CI #771 intent: the exact release-tool tagline extraction semantics, the 250-byte check, all 16 README tagline fixes (including the real plugins/ai tagline), task/lock wiring, and blocking quality-lane integration.
• Dropped as obsolete/superseded: nothing.
• Applied Deno's current formatter to 13 already-touched READMEs in 14a07686; these are mechanical wrapping changes discovered by rerunning the PR's own format gate after the base merge.

Validation

• deno task docs:tagline:check — PASS: checked=35 over=0.
• deno task docs:links — PASS: docs=96 broken-links=0 broken-anchors=0 orphans=0.
• deno fmt --check <16 touched README files> — PASS: Checked 16 files.
• deno run --allow-read --allow-run .llm/tools/run-deno-check.ts --root .llm/tools/validation --ext ts,tsx — PASS: 8 files selected, 1 batch, 0 failed batches, 0 diagnostics.
• git diff --check — PASS.
• Added-line scan for deno-lint-ignore, quality-allow, @ts-ignore, @ts-expect-error, as unknown as, and as any — PASS: zero new suppression-like constructs.

Pushed explicitly as HEAD:refs/heads/docs/jsr-tagline-byte-cap at 14a07686. No evaluator was dispatched and the PR was not merged.

[rickylabs]

IMPL-EVAL — PASS (supervisor-dispatched; Claude · Opus 4.8 · high, review_codex_light per the #794 ladder)

Verdict: .llm/runs/beta10-cli--orchestrator/slices/771-rebase/evaluate.md. Reconcile (14a0768) verified: nothing dropped silently, taglines still under the 250-byte cap post-formatter alignment, gate proven functional. status:ready-merge; merging once check-test is green on the branch (currently blocked repo-wide by the markdown-hydration CI failure, fix in flight).

🤖 Generated with Claude Code