jsdoc-scribe blog
Guides, product notes, and practical writing about documentation automation.
LinearB's 2026 benchmarks put a number on the AI-code review slowdown — agentic PRs wait 5.3x longer for reviewer pickup. Faros's data shows why. A worked example of exactly which part of that time generated documentation removes, and which part it explicitly doesn't.
On March 31, 2026, axios — one of the most-downloaded packages on npm — was compromised via maintainer credential theft. That incident, and the 1.2M+ malicious packages now blocked cumulatively across the registry, make a devDependency's own dependency tree a real audit question, not a nitpick. Here's how to actually check one, using jsdoc-scribe's package.json as the worked example.
Two independently-run 2026 due-diligence checklists just added the same new category — not who wrote a system, but whether anyone can explain it. Here's what actually answers that question, and why a clean demo stopped being enough.
A 2026 benchmark of 50+ engineering leaders found PRs per author up 20% year-over-year and incidents per PR up 23.5% — and named documentation as one of three foundations that separate teams who absorb that safely from teams who don't. What that actually means for an EM's review queue.
A self-audit three weeks ago found and fixed a jsdoc-scribe release that shipped with no changelog entry. Today the same gap reappeared on two newer releases — proof that a manual fix doesn't hold, and the honest case for Conventional Commits plus automated release tooling.
2026 research on AI coding assistants found repositories using them leak secrets at a materially higher rate than repositories that don't. A documentation generator built on the same mechanism — read your code, send it to a model — carries the same exposure. Here is a five-question checklist to run before you adopt one, with the data behind each question.
Backstage TechDocs is migrating off an end-of-life MkDocs toward Zensical, which means every team running it now has a reason to open mkdocs.yml and catalog-info.yaml before November 2026. Here is exactly how to use that same visit to wire in generated, drift-checked documentation.
Technical due diligence still treats "bus factor of one" as a valuation red flag, using git-blame authorship as a proxy for who understands a system. Agentic coding has quietly broken that proxy — here's what still works as a comprehension signal once authorship stops meaning anything.
code-multivitals just went live on GitHub Marketplace as a composite Action. Not a repackaging exercise — a genuinely different distribution model from the peerDependency jsdoc-scribe has used internally since Track C, and we dogfooded it in our own CI the same day it shipped.
A copy-pasteable Node script that walks every package in an npm workspaces monorepo and builds one merged documentation site using jsdoc-scribe's programmatic API — the working example the TypeDoc monorepo-mode comparison post promised but didn't include.
SyntaxScribe is the clearest name-adjacent competitor we've found in sixteen days of tracking — a real product with its own domain, showcase sites, and a Medium tutorial. Here is a sourced, feature-by-feature look at where it actually differs from jsdoc-scribe, and where it doesn't.
AI agents now account for roughly half to two-thirds of documentation traffic on major hosted-docs platforms, up from under 20% at the start of the year. Here's what that shift actually changes about how documentation should be generated and packaged, for engineering leaders deciding where to spend on tooling next.
CSA's AI-CAIQ v1.1 adds 320 AI-specific self-assessment questions on top of the base CAIQ. Here is the applicability test that determines how many of them actually apply to a given tool, and why a "no AI" architecture answers a whole cluster with one fact.
Backstage TechDocs and its peers fixed where documentation lives — co-located with code, reviewed in the same PR. That never guaranteed anyone updates it. Here is the staleness gap platform teams are standardizing into their catalogs, and the mechanical fix for it.
TypeScript 7.0 shipped stable with a 10x-faster Go-native compiler and no programmatic API until 7.1. Here's why that matters for any AST-based tool, and why jsdoc-scribe already pins below it.
A reference for security and procurement teams evaluating jsdoc-scribe against CAIQ and SIG — mapped domain by domain to specific, verifiable answers, ready to paste into an actual vendor review.
A full self-audit of jsdoc-scribe's own repo turned up a hardcoded 220-line cap quietly truncating the public changelog, a release that shipped with no changelog entry, a plugin whose install instructions 404, one real npm vulnerability, and zero ESLint or Prettier setup on the codebase itself. Here is every fix, and exactly how to use the result.
TypeDoc's dedicated monorepo mode still has open issues around cross-package references and per-package config duplication in 2026. Here is what that mode actually requires, and the simpler path jsdoc-scribe already ships for workspace-based repos.
When security teams go looking for unapproved AI tool use inside engineering, source code is the most commonly exposed data type. Here is what a shadow-AI review actually turns up in 2026, and why some tooling can't appear in the finding at all.
A copy-pasteable GitHub Actions workflow for publishing a jsdoc-scribe documentation site to GitHub Pages on every push to main, taken directly from this project's own CI.
The EU AI Act's high-risk obligations became enforceable on August 2, 2026. A day later, here is what genuinely went live, what was widely misreported in the run-up, and the one-week test most engineering orgs would fail if a regulator asked today.
SonarQube's free tier caps out at 50k lines of code and five users. Here is what jsdoc-scribe's optional --quality dashboard covers instead, what it deliberately doesn't try to replace, and where the honest line between the two sits.
A concrete, screen-by-screen walkthrough of what jsdoc-scribe's automatic Architecture Insight page shows a new engineer, and why it replaces the "grab a senior engineer and ask them to explain the codebase" ritual most teams still run.
A new arXiv paper names the failure mode where stale documentation silently degrades AI coding agents. Here is what the research actually measured, and why deterministic, drift-checked docs are a mechanical fix rather than a cultural one.
Bus factor has real measurement methodology now — silo ratios, truck-factor algorithms, dedicated tooling — and a hard dollar cost for getting it wrong. Here is how to measure it for your team and the one lever that raises it without adding headcount.
A security questionnaire is the single biggest cause of stalled B2B SaaS deals in 2026, and the newest line item on it is "does any AI component touch our source." Here is what that question costs a vendor that cannot answer it in one sentence, and what changes when it can.
jsdoc-scribe is built to never call an LLM. That is an easy line to sell and a harder one to defend honestly — here is the real trade-off, costs included, for founders, CTOs, and engineers who are skeptical of both sides of the AI argument.
ESLint core dropped require-jsdoc and valid-jsdoc in v9. Here's what fills the gap, why catching documentation drift at PR time beats catching it later, and how jsdoc-scribe's ESLint plugin autofixes 10 of its 12 rules.
Nearly half of startup deals collapse during technical due diligence. Here's why documentation quality — not code quality — is often the tell reviewers act on, and what changes when it's generated continuously instead of assembled for the data room.
The story of jsdoc-scribe — a free, 100% local, AST-based JSDoc generator and doc-site builder — and the real numbers behind adopting it.
A practical case study on how a client cut engineering transition overhead by 50% and eliminated technical debt using automated documentation.
A practical look at why teams should prefer repeatable AST-based documentation over hand-written boilerplate or generated guesses.