Release list

AgentLayer v0.2.0-alpha.3 Latest

Latest

Qqqq5910 released this 13 Jun 17:01

v0.2.0-alpha.3

9bbdb7b

AgentLayer v0.2.0-alpha.3

AgentLayer v0.2.0-alpha.3 is a small alpha patch from real-site testing.

Fixed

Forms with fragment-only actions such as action="#" no longer break report or artifact
generation.
Generated action manifests now fall back to the source page URL when a form action is not a stable
HTTP(S) URL.
Added regression coverage so client-side demo/contact forms can still produce report.html,
actions.json, and .well-known/agents.json.

Real-World Samples

Added 10 new anonymized public-site scan summaries under examples/real-world/.
The committed feedback sample set now contains 20 anonymized scans.
New samples include high-scoring, mid-scoring, crawl-limited, and canonical-domain redirect cases.

Try It

pnpm dlx @junyi5910/agentlayer-cli generate https://your-site.com --out ./agentlayer-output --max-pages 20
open ./agentlayer-output/report.html

Still Alpha

Public pages only.
No form submission.
No authenticated/private crawling.
Generated artifacts remain review drafts.

AgentLayer v0.2.0-alpha.2

AgentLayer v0.2.0-alpha.2 is a small alpha patch for report readability during real-site testing.

Fixed

Generated report.html no longer lets long task evidence, URLs, snippets, or artifact paths push
the whole page wider than the viewport.
Task, crawl issue, form, action, and fact tables now render inside responsive table wrappers.
Table cells and compact evidence lists now wrap long unbroken text.
The report header and generated artifact list have wrapping safeguards for long URLs and paths.

Try It

pnpm dlx @junyi5910/agentlayer-cli generate https://your-site.com --out ./agentlayer-output --max-pages 20
open ./agentlayer-output/report.html

AgentLayer v0.2.0-alpha.1

AgentLayer v0.2.0-alpha.1 introduces the first public alpha shape for AgentLayer: a scoped npm alpha
CLI quickstart, clearer reports, real-world scan feedback loops, and the first local AgentLayer CI
alpha.

Highlights

agentlayer baseline <url> saves a compact agentlayer-baseline/v1 JSON report.
agentlayer compare <url> writes an agentlayer-comparison/v1 JSON report.
--fail-on task-regression blocks when a task that passed in the baseline becomes partial,
failed, or missing.
--fail-on missing-artifact blocks when a generated artifact from the baseline is absent in the
current scan.
--fail-on score-drop --min-score-delta 5 blocks only when the overall score drop exceeds the
configured threshold.
Compare reports include machine-readable comparison JSON, a human-readable CLI summary,
recommendations, and an exitCode.
Generated reports explain score weights, task reasons, evidence, missing evidence, next fixes, and
Critical/Warning/Suggestion recommendation labels.
docs/scoring.md documents the deterministic scoring model and limitations.
examples/real-world includes 10 anonymized bounded public-site scan summaries and heuristic
notes.
New issue templates collect real scan feedback, false positives, and confusing recommendations.
Two real-world heuristic issues have focused fixes: blog/resource page-type false positives and
navigation-only demo/contact wording.
examples/ci includes copyable public-site and local-fixture GitHub Actions examples.
A manual published CLI smoke workflow can verify pnpm dlx @junyi5910/agentlayer-cli after npm
publish.
docs/release-checklist.md documents publish order, smoke tests, and when to sync GitHub.
docs/launch/launch-posts.md and docs/launch/share-your-scan.md provide post-publish outreach
copy for collecting real scans.
docs/ci.md documents local baseline/compare usage and a GitHub Actions alpha workflow.

Example

pnpm dlx @junyi5910/agentlayer-cli baseline https://example.com \
  --out ./agentlayer-baseline.json \
  --max-pages 20

pnpm dlx @junyi5910/agentlayer-cli compare https://example.com \
  --baseline ./agentlayer-baseline.json \
  --out ./agentlayer-compare.json \
  --fail-on task-regression \
  --fail-on missing-artifact \
  --fail-on score-drop \
  --min-score-delta 5 \
  --max-pages 20

Limitations

Local-first only; there is no hosted AgentLayer CI service.
No PR comment bot yet.
No official compliance guarantee for MCP, WebMCP, llms.txt, Agent Skills, or future standards.
Use the scoped npm alpha CLI for public-site trials. Repo-local commands are only for repository
checkouts and fixture development.
This alpha publishes @junyi5910/agentlayer-core and @junyi5910/agentlayer-cli.
Do not use the unscoped agentlayer package name.
The scanner does not crawl authenticated/private areas, submit forms, or perform destructive
actions.

AgentLayer v0.1.0

AgentLayer v0.1.0 is the first public release of a deterministic toolkit for checking whether public
websites are readable, trusted, and operable by AI agents.

SEO made websites discoverable. AgentLayer makes websites operable by AI agents.

Install And Run

AgentLayer v0.1.0 is documented for repo-local use:

git clone https://github.com/Qqqq5910/agentlayer.git
cd agentlayer
pnpm install
pnpm build

Start the local example SaaS fixture:

pnpm dev:example

In another terminal, generate artifacts and run the doctor command:

pnpm agentlayer generate http://localhost:3001 --out ./agentlayer-output --max-pages 20 --allow-local
pnpm agentlayer doctor http://localhost:3001 --max-pages 20 --allow-local

Optionally start the web app:

pnpm dev

Open the demo report at http://localhost:3000/demo. The AcmeFlow fixture runs at
http://localhost:3001.

For a public site:

pnpm agentlayer generate https://example.com --out ./agentlayer-output --max-pages 20
pnpm agentlayer test https://example.com --tasks ./examples/tasks/b2b-saas.default.json --out ./agentlayer-report.json
pnpm agentlayer doctor https://example.com --max-pages 20

Highlights

Bounded public-site scanning with same-host, maxPages, timeout, and robots.txt guidance.
Sourced fact extraction with source URLs and confidence metadata.
Action path and form operability detection for agent-facing review.
Deterministic B2B SaaS task checks for pricing, docs, security, integrations, support, and
demo/contact paths.
Generated llms.txt, llms-full.txt, Markdown snapshots, .well-known drafts, WebMCP
suggestions, JSON reports, recommendations, and report.html.
Local Next.js demo report at http://localhost:3000/demo.
AcmeFlow fixture site for repeatable scanner demos at http://localhost:3001.

v0.1 Limitations

The recommended execution path is repo-local pnpm agentlayer; npm package execution is planned
later.
Extraction is heuristic and conservative.
Generated MCP, WebMCP, API Catalog, and Agent Skills files are draft suggestions, not compliance
claims.
AgentLayer does not guarantee compliance with MCP, WebMCP, llms.txt, or any future standard.
The scanner does not authenticate, submit forms, crawl private areas, bypass blocked sites, or
perform destructive actions.
Task checks are currently tuned for B2B SaaS-style public websites.
This is not a crawler API, AI SEO rank tracker, compliance guarantee, or browser automation
framework.

Next Roadmap

npm package publishing and package-manager execution.
WordPress plugin.
Webflow plugin.
Shopify adapter.
Next.js middleware.
Cloudflare Worker.
Real WebMCP integration.
MCP server implementation.
LLM judge plugin.
Browser-agent task replay.
Hosted SaaS version.

Releases: Qqqq5910/agentlayer

Release list