Archify
Turn an idea, plan, or codebase into an interactive, shareable single-file HTML diagram from inside your coding agent.
- Source repo
- tt-a1i/archify
- Stars
- ★ 87k
- Last updated
- today
- License
- MIT
- Primary language
- JavaScript
- FA score
- 77/100 · Good
At a glance
- Works with
- Universal · cross-platformCodex · Claude CodeClaude.ai (Partial support)ChatGPT (Unsupported)
- You'll need
- Typical use
- An engineer joining an unfamiliar codebase asks the agent to read it and produce a runtime architecture map whose nodes cite Git-verified files and line ranges
- Main limitation
- Automatic Mermaid parsing, general-purpose auto-layout, hosted sharing, and WYSIWYG editing are explicitly out of scope, so teams relying on those must change workflows
- Source review
- 77/100 · Good
What does this agent do, and when should you use it?
Archify is an agent skill that teaches coding agents such as Claude Code, Codex CLI, Cursor, and opencode to author a typed JSON IR and render it into one self-contained interactive HTML file. The repository ships its own renderer and validators (archify/bin/archify.mjs) with validate, deliver, preview, guide, and compare subcommands, and covers ten diagram types from architecture and sequence to ERD and waterfall. Before delivery, schema, layout, HTML/SVG, route, and label-to-route clearance checks must all pass, and only then does the candidate atomically replace the last known good artifact; failures come back as machine-readable repair receipts. The output opens in any browser without installing Archify and supports themes, focus, upstream/downstream reach, route probing, and PNG or share-card export.
The agent converts a description or a repository reading into typed JSON IR. node bin/archify.mjs validate <mode> <file> --quality showcase --json runs the bundled schema and layout rules; preview starts a loopback-only session that watches one JSON file and reloads only verified revisions while keeping the previous artifact on failure; deliver renders a same-directory candidate, checks it, atomically replaces the target, and can --open the exact file; guide recommends a diagram type from a natural-language scenario; compare architecture base.json head.json produces an Architecture Delta with a machine receipt. Rendered HTML restores state from #focus=, #reach=, #route=, and #lens= links and exports PNG to clipboard, static or motion files, and 1200×630 Route/Reach Share Cards.
- An engineer joining an unfamiliar codebase asks the agent to read it and produce a runtime architecture map whose nodes cite Git-verified files and line ranges
- A backend team debugging a cache miss or auth failure diagrams the call as a sequence and traces the exact route from web app to Postgres
- A reviewer comparing a design or PR generates a Before/Delta/After Architecture Delta with a machine receipt instead of a hand-drawn sketch
- Someone explaining where time went in one request uses Waterfall or Timeline to show spans on one proportional clock
- A developer sends a self-contained HTML file or a route share card to a stakeholder who has no Archify installation
How do you install or deploy this agent?
Install globally through the skills CLI (requires Node.js):
npx skills add tt-a1i/archify -gFor an explicit non-interactive Cursor install:
npx -y skills add tt-a1i/archify --skill archify --agent cursor --global --copy --yesTo try without installing:
npx skills use tt-a1i/archify@archify --agent codexOn Claude.ai, upload archify.zip under Settings → Capabilities → Skills; whether it works depends on Node.js access in the sandbox.
How do you use this agent?
After installing, invoke the skill in chat:
Use Archify to diagram a web request: Browser calls the API,
the API checks Redis, and a cache miss queries PostgreSQL and fills the cache.Then refine with add authentication or highlight the cache-miss path. To drive the CLI directly inside the repository:
cd archify
node bin/archify.mjs doctor
node bin/archify.mjs guide "Show an API request with Redis cache miss"
node bin/archify.mjs validate workflow examples/agent-tool-call.workflow.json --quality showcase --json
node bin/archify.mjs deliver workflow examples/agent-tool-call.workflow.json /tmp/workflow.html --quality showcase --open --jsonSet locale and animation in the artifact meta:
{
"meta": {
"locale": "en",
"animation": "trace",
"visual_preset": "signal-flow"
}
}What are this agent's strengths and limitations?
- Delivery is gated and atomic: schema, layout, route, and label-clearance checks must pass before the candidate replaces the last good file, and failures return stable rule codes with measured evidence
- The artifact is one HTML file, so node details, routes, and path exploration survive being emailed to someone without Archify
- The typed JSON IR is reproducible and diffable, and
compare architectureturns a design review into a Before/Delta/After snapshot with a machine receipt - Ten diagram types, including ERD, class, tree, timeline, and waterfall added in 3.1, instead of only boxes and arrows
- Runs across Claude Code, Codex CLI, Cursor, and opencode for free under MIT, with renderer and validators in the repo
- Automatic Mermaid parsing, general-purpose auto-layout, hosted sharing, and WYSIWYG editing are explicitly out of scope, so teams relying on those must change workflows
- Diagram hierarchy and routes are chosen by the agent, so output quality tracks the host model and still needs human visual review
- The update checker performs network requests by default and must be disabled with
ARCHIFY_UPDATE_CHECK_DISABLED=1for fully offline use - Claude.ai and Project Knowledge support depends on Node.js availability in the sandbox, making that a weak path
- The documented CLI path assumes a Node.js runtime; no alternative non-Node execution route is described
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Archify This agent | 77 · Good | — | ★ 87k | today | JavaScript | Codex · Claude Code |
| Yü Yu: Decoding the Agent Harness | 63 · Some gaps | Web appFree | ★ 4.3k | 5d ago | Python | — |
| Claude-Codex Skills Marketplace | 54 · Major gaps | Agent plugin / skillFree + model costs | ★ 582 | 3d ago | Shell | Codex · Claude Code |
| AWS Agent Skills | 27 · Major gaps | Agent plugin / skillFree + model costs | ★ 1.2k | 1mo ago | Python | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows the update check only GETs a fixed manifest, never auto-installs, can be disabled via ARCHIFY_UPDATE_CHECK_DISABLED=1, and explicitly states no version/prompt/project data is sent; deliver writes a same-directory candidate and replaces the target only after all gates pass, giving rollback-safe writes; the measurement doc states no fetch/sendBeacon/XHR and sessionStorage-only receipts. Deductions: installation relies on npx skills add and third-party skill tooling whose permission boundary is not described in-repo; numerous sponsor and third-party service links create external effects and data flows to third parties that are not individually disclosed; dependency security shows only a few devDependencies and a fast-uri override with no audit or vulnerability-handling record; publisher identity is unverified and source attribution rests on README self-description.
CI workflows are structurally thorough (multi-Node matrix, browser regression, Windows portability, zip freshness, published-manifest checks), and the cookbook documents failure receipts with stage/diagnostics/supportedFixes and states non-zero exit is never success. Deductions: these are static configs, not executed verification; dependency availability depends on Node>=18 and an optional Chrome/Chromium path with no stated degradation behavior; README and cookbook differ slightly on command paths (archify/bin vs bin), a minor self-consistency risk.
README targets Claude Code, Codex, Cursor, and OpenCode, offers ten diagram types, a scenario guide, and a no-repository start, so audience and scenarios are well covered; capability boundaries are explicit (deployment-ownership fails closed, Delta infers no impact/risk, inspect is Architecture-only, visual-check does not approve polish). Deductions: trigger precision relies on natural-language prompts with no stated trigger/non-trigger rules or counterexamples; environment fit requires Node 18+ and optional Chrome without a complete alternative path for browserless environments.
Information architecture is clear (README sections, cookbook, schema reference, multilingual READMEs), install notes are detailed including non-interactive commands and update-check details, examples and FAQ are rich (11 scenarios, Proof Lab, community cases), the MIT license file is complete, and a CHANGELOG plus 3.2.0 version badge exist. Deductions: naming stability is weakened by the bin path vs package name archify-repository mismatch; known limitations are scattered rather than consolidated; maintenance responsibility points only to an unverified individual publisher and an email with no governance or maintainer list, hence the low score.
Output is self-contained HTML that opens offline, exports PNG/WebP/JPG/PDF and share cards, giving high output usability; it has clear marginal value for agent users needing diagrams. Deductions: marginal value partly depends on the agent's own generation ability, and README marketing claims (Trending #1, 130K installs) cannot be verified from source; cost-benefit requires a Node runtime and optional Chrome, and diagram quality depends on prompting with no quantified comparison.
claim_traceability is decent with SHA-256 receipts, pinned revisions, source-range verification, and numbered citations; cross_source_corroboration is supported by CI config, cookbook, and measurement doc agreeing. Deductions: external claims such as ranking, install counts, and media coverage are only linked to third parties with no in-repo corroboration; fact_inference_separation is strong, as the measurement doc distinguishes intent signals from activation/success and Delta infers no impact, justifying the high score.
- Publisher identity is unverified by the FollowAgents registry; maintenance responsibility and update path rest on self-description and should be independently confirmed before enterprise adoption.
- Installation relies on npx skills add and third-party skill tooling whose permission scope and supply-chain risk are not described in-repo.
- README contains many sponsor and third-party service links with referral codes, implying external data flows and commercial incentives that warrant independent review.
- Claims such as ranking, install counts, and media coverage cannot be verified from repository source and should not be treated as reliability evidence.
- Dependency security shows only a few devDependencies and a fast-uri override, with no dependency audit or vulnerability-handling record.
- This is a static source review; no commands or tests were executed, so CI configuration effectiveness is unverified.
FAQ
Does Archify call a language model itself?
Can a failed run overwrite my last good diagram?
deliver renders a same-directory candidate and replaces the target only when every check passes; preview likewise refreshes only verified revisions and keeps the previous artifact visible on failure.Are timeline timestamps and waterfall durations real?
What does it send over the network?
Which agent hosts are supported?
npx skills add and documents install locations for Claude Code, Codex CLI, opencode, Claude.ai, and Project Knowledge, plus opt-in community integrations for Hermes Agent and DeepSeek Harness.