Skip to content

Changelog

All notable changes to this project are documented in this file.

[1.3.0] - 2026-08-13

Added

  • Unified findings model — consolidates code-side architectural risks (god modules, layer violations, dependency explosions, custom rule violations) and network-side findings (open ports, drift violations) into one stable Finding shape with consistent id, domain, category, severity, confidence, and evidence fields. Code findings get deterministic IDs (ARCH-001, ARCH-002, ...) and all inherit confidence: 1.0 (static analysis is deterministic). Network findings get NET-* IDs and retain their variable confidence scores from endpoint scanning. Existing risks, architecture.ruleViolations, and expose.findings consumers remain unchanged; the new unified AnalysisResult.findings list is additive and ready for follow-on roadmap items (memory network section, network diff, web UI rule panels) without breaking any current integration.
  • archmap.core.analyzer.findings module — new build_code_findings() transforms architecture/risk/cycle data into the unified Finding list, and exports SEVERITY_LEVELS and max_severity() helper (deduplicating prior ad hoc severity ranking from correlate.py). Public API: stable Finding TypedDict in archmap.types.

Fixed

  • archmap init --from-analysis layer hints — was reading nonexistent fromLayer/toLayer fields off layer-violation entries, always silently returning empty strings. Now correctly reads sourceLayer/targetLayer (the actual keys produced by risk_analyzer) so derived archmap.toml hints reflect observed layer violations.

[1.2.0] - 2026-08-12

Added

  • archmap memory — renders the project's architecture (health score, top risk files, cycles, layer/rule violations) into a compact markdown digest (default .archmap/memory.md) meant for AI coding agents to read at session start instead of re-exploring the codebase from scratch. Reuses archmap analyze's own fingerprint cache, so repeat calls after a no-op edit are nearly instant, and writes are idempotent (a call that finds no architectural change skips rewriting the file rather than touching its mtime over a timestamp-only diff). --print prints the digest instead of writing a file (for a Claude Code SessionStart hook to inject straight into context); the default write behavior pairs with a PostToolUse hook to keep the file fresh after every edit — see docs/cli/memory.md for ready-to-paste hook config. Also exposed as a new get_project_memory MCP tool (8 tools total) for agents that use MCP instead of hooks.
  • archmap expose confidence scoring — every finding now carries a confidence (0.0–1.0) and confidenceReasons. A new endpoint scanner (core/exposure/endpoint_scanner.py) looks across the project (source files, .env files, and common config formats) for literal connection endpoints — URIs with a known scheme (redis://host:6379) and paired *_HOST/*_PORT config keys — and an exact host:port match against the scanned target is much stronger evidence than the existing service-name- to-package-import guess, which now only backs the lowest confidence tier.
  • Network rules & drift detection — .archmap.toml gains a [network.rules] section (forbid/allow, same "source-tag -> target-tag" syntax as [architecture.rules]) to declare which parts of the codebase are expected to talk to which network services. When a high-confidence connection (one ArchMAP can attribute to a specific file) breaks a declared rule, expose flags it as a driftViolation on that finding, bumps its severity to at least high, and lists it in a new top-level driftViolations array — "declared vs. observed architecture" extended past the codebase to the network boundary. get_network_exposure (MCP) picks up both features automatically.
  • archmap.core.analyzer.rule_engine — the tag-based rule-matching engine backing [architecture.rules] was extracted into its own reusable module (no behavior change for existing architecture rules) so the new network rules could reuse it instead of duplicating the matching logic.

[1.1.1] - 2026-08-12

Added

  • archmap expose — correlates archmap netscan results with the codebase's dependency graph: for each open port whose service (Redis, PostgreSQL, MongoDB, Elasticsearch, etc.) has a known client library, finds matching pkg:<name> nodes in the analyzed project and surfaces their already-computed blast radius (impactCount/impactedFiles/risk) alongside the port's network-side risk rating, combined into one severity. Accepts every netscan option plus a project path. No new graph analysis — every package node already carries a precomputed impact field, this just cross-references two things ArchMAP already computes separately.
  • 3 new MCP tools — archmap mcp now exposes 7 tools instead of 4:
  • trace_reachability — BFS from an entrypoint through outgoing dependency edges, wrapping the same analyzer archmap trace uses, for dead-code detection and understanding what an entrypoint actually pulls in.
  • diff_architecture — architecture diff between two git refs (default HEAD~1 vs HEAD), wrapping the same analyzer archmap diff uses, so an AI agent can check whether a change degrades architecture before proposing a commit.
  • get_network_exposure — runs a live archmap expose-style scan/ correlation against a target and the analyzed project, so an agent can ask whether an open port maps to code it depends on.

Fixed

  • MCP server stale-cache bug — the in-memory analysis cache only invalidated on .archmap.toml mtime changes, so a long-lived MCP session could keep serving stale impact_analysis/get_file_context results after the AI assistant edited source files. It now keys on the same content fingerprint (config + every source file's mtime/size) the on-disk cache already uses, so any source change is detected. The dead _invalidate() function (never called) was removed.
  • MCP impact_analysis recomputed data that was already free — it now reads the node's precomputed impact field directly instead of rebuilding the dependents map from scratch on every call. This also makes it explicit that the tool works on any node id, including pkg:<name> package nodes.

[1.1.0] - 2026-08-12

Network mapping release. ArchMAP grows past static code analysis with a built-in network scanner, plus a round of mobile/Termux fixes found while dogfooding the web UI and CLI on Android.

Added

  • archmap netscan — nmap-style network discovery and port scanning, built stdlib-only so it runs without root on Termux (Android) as well as on regular Linux/macOS/Windows. Accepts single hosts, CIDR blocks, or dash ranges; supports host discovery (ping with TCP-probe fallback), threaded TCP port scanning, service banner fingerprinting, and JSON output. --use-nmap delegates to a system nmap install for deeper scans (-sV, -O, etc. via --nmap-args). First step toward ArchMAP mapping more than just source code.
  • archmap netscan report redesign — human-readable output is now a clean aligned table (PORT/STATE/SERVICE/INFO per host, plus an up/duration summary line and a totals footer) instead of a flat list, so it reads the same way regardless of engine. New --os-detection flag surfaces nmap's -O OS guess when the nmap engine runs; nmap's own scan stats, version, and richer service info (product/version/extrainfo) are now parsed and shown too.
  • nmap is now the default netscan engine when installed — it's simply a more capable scanner than a from-scratch one. --use-nmap now only forces it (erroring if missing) instead of opting in; pass --no-nmap to fall back to the built-in stdlib-only scanner even when nmap is present.
  • Per-port analysis — every open port is now checked against a static table of known-risky ports/services (Telnet, SMB, RDP, VNC, exposed Redis/MongoDB/Elasticsearch/Memcached, unauthenticated Docker API, etc.) and flagged with a [LEVEL RISK] tag and reason in the report, on either engine. The Python engine's fingerprinting now does a real HTTP(S) GET (including over TLS on 443/8443/...) to extract the page title and Server header instead of just the first response line. New --scripts flag runs nmap's default NSE scripts + version detection (-sC -sV) for deeper per-port and per-host analysis (page titles, TLS cert info, known misconfigurations); script and risk info are parsed into the JSON report and shown indented under each port in the human-readable output, with a risky-port count added to the summary footer.

Fixed

  • Web UI (archmap serve) mobile/Termux layout — the desktop toolbar (fixed-width search box, spacer-pushed buttons) no longer gets squeezed into an unreadable single row once the workspace collapses to its single-column mobile layout (≤960px, e.g. tablets and phones in landscape). The toolbar now wraps into title/tabs/search/action rows, export/refresh buttons go icon-only with accessible labels, the nav rail scrolls horizontally with larger touch targets instead of overflowing off-screen, and the mini-map shrinks (then hides below 460px) instead of overlapping graph content. Verified at common phone/tablet/landscape viewport sizes.
  • Web UI "Open project" button doing nothing on Termux/headless servers — the native folder picker (/api/open) needs tkinter and a display, which Termux doesn't have; the server already detected this and returned a manual_path fallback response, but the frontend silently ignored anything that wasn't a 200 and dropped it on the floor. It now prompts for the path directly and posts it to /api/project, with a visible error message on failure instead of nothing happening. /api/project also now expands ~ in the entered path (previously only worked with absolute paths).
  • Silent "0 files analyzed" on a nonexistent project path — analyze, serve, and every other command now fail fast with project path does not exist: ... when given a path that isn't actually there, instead of quietly reporting a "successful" scan with 0 files and 100/100 health. This was the confusing symptom behind a real Termux issue: Android's scoped storage can make a project folder look empty (or the path resolve to nothing at all) from Termux's side even though the files are genuinely on the device — see the new Termux setup docs.

Docs

  • Added a Termux (Android) shared-storage troubleshooting section (termux-setup-storage, the separate Android "All files access" permission, and copying into Termux's own home directory as the most reliable fix) to the installation guide, linked from the README.

[1.0.3] - 2026-06-11

Tree-sitter engine overhaul. Tree-sitter becomes a resilient primary parser with automatic per-file fallback to the regex path. Fully backward-compatible; the optional [tree-sitter] extra is still optional.

Added

  • Per-language grammar availability — each tree-sitter grammar now loads independently. A single missing or broken grammar (e.g. tree-sitter-php) no longer disables tree-sitter for every other language; the affected language simply uses its regex fallback. New ts_engine.language_available(name).
  • Automatic per-file fallback (ts_engine.try_extract) — the parsers call a single orchestrator that returns the structured tree-sitter result, or None to request the regex fallback when tree-sitter cannot be trusted for that file: the grammar is unavailable, parsing/querying raised an exception, or the parse contained syntax errors and recovered no imports. Previously a file that made tree-sitter raise was dropped from the graph entirely.
  • Structured C# extraction — using directives are read from typed AST nodes, so using Alias = Real.Namespace; resolves to the real namespace (not the alias), and global using / using static are handled. This matches the regex fallback's behaviour so both paths agree.
  • CI now tests both paths — a dedicated tree-sitter job runs the suite with the AST grammars installed, while the existing quality job continues to cover the regex fallback. The tree-sitter code path was previously untested in CI.

Changed

  • Rust regex fallback now strips comments before matching, consistent with the other regex fallbacks introduced in 1.0.2.
  • ts_engine extractors were refactored to parse-then-extract from a shared tree, enabling the reliability check without changing their public signatures.

[1.0.2] - 2026-06-11

Parser-accuracy release. Every change reduces false positives or raises the resolution rate of the regex fallbacks (used when the optional [tree-sitter] extra is not installed). Fully backward-compatible.

Added

  • Comment-aware regex fallbacks — a shared, string-aware strip_comments pass removes //, /* */ (and # for PHP) comments before import matching, so import-like text inside comments no longer produces phantom dependencies. Applied to JavaScript/TypeScript, Java, C#, PHP and C/C++. String literals (e.g. require("x"), #include "x") are preserved, and line structure is kept intact for re.MULTILINE matching.
  • Go replace directives — local replacements in go.mod (replace x => ./local) now resolve imports to the replaced directory instead of reporting them as external packages.
  • PHP composer PSR-4 autoload — use statements resolve through the autoload/autoload-dev psr-4 map in composer.json (longest prefix wins), replacing the previous fragile suffix-matching heuristic for configured roots.
  • C# alias and global using — using Alias = Real.Namespace; now extracts the right-hand namespace (previously captured the alias name), and global using directives are recognized.
  • Java inner-class resolution — import com.example.Outer.Inner; resolves to com/example/Outer.java when no Inner compilation unit exists.
  • C/C++ include-dir resolution — a header included as "foo/bar.h" resolves to a unique include/foo/bar.h or src/foo/bar.h via path-suffix matching, covering the common -I include-directory layout.

Changed

  • C# using static System.Math; now parses to the System.Math namespace (the static directive keyword is no longer kept in the parsed value).

[1.0.1] - 2026-06-11

Hardening and correctness release. No public API breakage — every change is backward-compatible with 1.0.0.

Security

  • serve binds to 127.0.0.1 by default (was 0.0.0.0). Exposing the server on the network now requires an explicit --host and prints a warning.
  • All state-changing / local endpoints gated to loopback — /api/project (switch analyzed directory), /api/reanalyze, and /api/advise now require a loopback client, matching /api/open and /api/open-file. Previously a host on the same network could repoint the analyzer at any directory and read its structure via /api/graph, or trigger outbound LLM requests (SSRF).
  • /api/advise validates base_url — only well-formed http/https URLs are forwarded server-side; file:// and other schemes are rejected.

Fixed

  • Whole-graph impact analysis is now O(V+E) instead of O(V·E). The backward adjacency map is built once and shared across nodes (impact_analyzer.build_dependents_map), and the BFS uses a deque instead of list.pop(0). Large repositories no longer spend most of the analysis time in impact calculation.
  • LLM advisor layer-violation rendering — the prompt read non-existent fromLayer/toLayer keys and emitted None -> None. It now reads the sourceLayer/targetLayer/source fields actually produced by the risk analyzer.
  • Mermaid label escaping — labels containing \n, [ ], { }, < >, " or \ no longer break the generated diagram; labels are quoted and the breaking characters are escaped or substituted.

Added

  • tsconfig / jsconfig path-alias resolution — JS/TS imports that rely on compilerOptions.baseUrl and compilerOptions.paths (e.g. @app/* → src/*) now resolve to internal files instead of being reported as external packages, raising resolutionRate on real-world TypeScript projects. JSONC comments and trailing commas in the config are tolerated.

Changed

  • Coverage gate enforced in CI — --cov-fail-under=85 added to the pytest configuration; the suite now fails if coverage regresses below 85%.
  • Dockerfile pinned to python:3.13-slim (was the unreleased 3.14-slim).

[1.0.0] - 2026-05-25

Added

  • archmap temporal — temporal coupling analysis via git log. Detects files that frequently change together in commits — hidden coupling not visible in static imports. Outputs ranked pairs with co-change count and coupling strength (0–1). Flags: --min-commits N, --top N, --json. Zero new dependencies (stdlib subprocess only).
  • Web UI — SVG export — "SVG" button in the canvas toolbar next to PNG. Uses Cytoscape.js native cy.svg({ full: true }) API and downloads as archmap-graph.svg. Scalable vector output, zero new dependencies.
  • MCP server (archmap mcp) — exposes the ArchMAP analysis engine as a JSON-RPC 2.0 server over stdio. Four tools: get_architecture_summary, get_file_context, impact_analysis, run_checks. Compatible with Claude Code, Cursor, Windsurf, and any MCP-capable AI assistant.

Changed

  • Stable public Python API — from archmap import analyze_project, AnalysisResult is now fully typed and considered stable. All types exported from archmap.types. Zero-breaking-change commitment for all 1.x releases.
  • mypy strict enforcement — 0 type errors enforced as a blocking CI gate. All new code must pass mypy at the same strictness level.
  • Test coverage — raised to 88% (from 85% in v0.9.0). New test files: test_reachability_analyzer, test_generic_parser, test_rust_go_parsers, test_temporal_analyzer.

Migration from 0.x

No breaking changes. The analyze_project() function and all return types are backward-compatible with 0.9.0. The only additions are: - New archmap temporal subcommand. - New SVG export button in the Web UI. - New archmap mcp subcommand (available since 0.9.x, now stable). - from archmap import AnalysisResult and all sibling types are now part of the public API contract.

[Unreleased]

Added

  • Tree-sitter multi-language parser — optional [tree-sitter] extra (pip install archmap[tree-sitter]) replaces every regex-based parser with a proper AST engine. When the extra is installed, all nine language parsers use tree-sitter grammars:
  • JavaScript / TypeScript — tree-sitter-javascript / tree-sitter-typescript (+ TSX variant). Multiline imports, import type, export type all handled correctly. Import-like text in string literals and comments no longer produces false positives.
  • Java — tree-sitter-java. Static imports, wildcard imports (com.example.*), and standard imports all resolved correctly.
  • Go — tree-sitter-go. Both single-import and block-import forms captured without regex edge cases.
  • Rust — tree-sitter-rust. use, mod, and extern crate declarations extracted and normalised using the existing _normalize_rust_use logic.
  • PHP — tree-sitter-php. Handles both single-quoted and double-quoted strings, and the parenthesized call form (require('x')).
  • C# — tree-sitter-c-sharp. using, using static, and alias directives.
  • C / C++ — tree-sitter-c / tree-sitter-cpp. #include <system> and #include "local" correctly distinguished by node type.
  • The regex fallback is preserved in all parsers; no behaviour change when the extra is not installed.
  • Per-import unresolved tracking — parser now resolves each import statement individually. importsTotal, importsResolved, and unresolvedImports fields on every ParsedFile. metrics.unresolvedImports and metrics.unresolvedImportsTotal in the graph report.
  • --show-unresolved[=N] — prints a detailed list of import statements that could not be resolved (file + specifier). N controls how many are shown (default 20 when flag is present).
  • --fail-on-budget-violations — CI gate that exits with code 2 when any file exceeds coupling budgets. Works with CLI overrides or .archmap.toml.
  • --max-outgoing-per-file N — CLI override for maximum allowed outgoing dependencies per file.
  • --max-incoming-per-file N — CLI override for maximum allowed incoming dependencies per file.
  • [analysis.budgets] in .archmap.toml — max_outgoing_per_file and max_incoming_per_file config keys for project-wide coupling limits.
  • --ignore-external — strips external package nodes (e.g. pkg:requests) and their edges from the output graph. Metrics are recomputed after filtering. Useful for focusing on internal structure only.
  • --summary-format {text,table,markdown} — controls the terminal summary output style. text (default) is the existing [ok]/[warn] format; table renders an aligned ASCII table with separator lines; markdown renders a pipe-delimited markdown table for pasting into docs or PRs.

Fixed

  • SVG injection in /api/badge — health grade string is now XML-escaped via xml.sax.saxutils.escape before being embedded in the SVG response. Prevents XSS if a grade value ever contained <, >, or &.
  • Web UI Config navigation — Config panel is now accessible via a dedicated gear-icon button in the nav rail (navBtnConfig). The ArchMAP logo click is restored to its original behavior: navigating to the Graph view. Previously the logo opened Config, leaving no dedicated button and no way to return to the graph.
  • _VALID_LLM_PROVIDERS scoping in server.py — moved to module-level frozenset constant; previously re-declared as a local variable inside _handle_advise on every request.

[0.9.0] - 2026-05-08

Added

  • Parser resolution rate — resolutionRate metric (0.0–1.0) now included in every analysis report. Tracks what fraction of import statements were successfully resolved to a file or package. Visible in the CLI summary (warns when below 70%), in the Web UI Insights panel, and in JSON output (metrics.resolutionRate). New --min-resolution-rate PCT flag gates CI on this metric (e.g. --min-resolution-rate 70).
  • Shell completion — archmap --print-completion bash|zsh|fish prints the shell-specific setup command. Install the argcomplete optional dep (pip install archmap[completion]) and add the printed line to your shell profile for full tab-completion of all subcommands and flags.
  • Web UI — PNG export — "PNG" download button in the canvas toolbar. Exports the current dependency graph as a 2× scale PNG using Cytoscape.js's native cy.png() API. Zero new dependencies.
  • Docker image — ghcr.io/kaua-kgzin/archmap:latest published automatically on each GitHub release. Includes :X.Y.Z and :X.Y version tags. Use docker run -p 3000:3000 -v $(pwd):/project ghcr.io/kaua-kgzin/archmap to start the Web UI.
  • Web UI — i18n — Full interface internationalization: English, Português (Brasil), Español, Français, Deutsch, 中文 (简体). Language selector in the Config panel; setting persists across sessions and applies instantly without reload.
  • Web UI — Config panel — Clicking the ArchMAP logo opens a persistent settings panel: Interface (language), General (auto-save config, animations toggle), Graph (default layout: cose/breadthfirst/concentric/circle/grid), LLM Advisor, Scan Languages. All settings saved to localStorage.
  • Web UI — Advisor persistence — LLM provider, model, base URL, and API key are saved to localStorage and restored on every session. Auto-save mode writes on every keystroke.

Changed

  • metrics.resolutionRate field added to the JSON report schema (always present, defaults to 1.0 when no imports are found).

[0.8.0] - 2026-05-07

Added

  • archmap trace <entrypoint> — BFS reachability analysis from any source file: shows the full dependency tree by depth level, coverage percentage, and optionally lists unreachable files (--unreachable). Supports --max-depth, --json, and --out.
  • archmap advise — LLM-powered architectural advisor. Sends a compact report summary to an LLM and returns concrete refactoring advice. Supports multiple providers with zero new runtime dependencies (pure urllib):
  • --provider claude — Anthropic API (reads ANTHROPIC_API_KEY)
  • --provider openai — OpenAI API (reads OPENAI_API_KEY)
  • --provider ollama — local Ollama instance (default http://localhost:11434)
  • --provider custom --base-url <url> — any OpenAI-compatible endpoint (LM Studio, etc.)
  • --model, --api-key, --timeout flags for full control.
  • archmap init --from-analysis — analyzes the project first and derives .archmap.toml layer rules from actual dependency direction (files with higher fan-in rank as more foundational), including detected forbid rules from live violations.
  • archmap diff --snapshot-a <file> --snapshot-b <file> — compare two exported JSON snapshots directly, without requiring git refs. Useful for cross-machine or cross-deploy comparisons.
  • VS Code extension (vscode-extension/) — zero-config IDE integration:
  • Inline diagnostics in the Problems panel for cycles, layer violations, god modules, and custom rule violations.
  • Status bar item with live health score and grade.
  • ArchMAP: Analyze Project command.
  • ArchMAP: Open Web UI command (launches archmap serve and opens browser).
  • ArchMAP: Trace File Reachability command — opens a webview with the BFS depth tree.
  • archmap.analyzeOnSave setting for CI-style continuous feedback.

[0.7.2] - 2026-04-27

Added

  • Web UI — multi-view navigation: nav rail now switches between three independent left-panel views (Graph, Insights, Rules & Layers).
  • Web UI — Layer Diagram tab: new canvas tab renders the dependency graph with a hierarchical breadth-first layout and a floating legend.
  • Web UI — Dependency Matrix tab: new canvas tab renders the top-25 most-connected files as a colour-coded dependency matrix (dependency, circular, self-reference).
  • Web UI — Insights panel: architecture metrics summary cards populated live from report data (health score, files, cycles, complexity, etc.).
  • Web UI — Rules & Layers panel: displays layer rule cards from .archmap.toml; shows a "no rules configured" placeholder when absent.
  • Breadcrumb and canvas title update dynamically when switching canvas tabs.

[0.7.1] - 2026-04-27

Fixed

  • Web UI: bundle Cytoscape 3.30.2 locally — eliminates CDN dependency and fixes broken graph (CSP was blocking unpkg.com).
  • Web UI: correct Content-Security-Policy to allow Google Fonts while keeping scripts same-origin.
  • CI (windows-exe): add pillow to PyInstaller deps so icon.png → .ico conversion works on the runner.
  • PyPI publish: align package name with PyPI project (KG-ARCHMAP) so OIDC trusted publisher accepts uploads.
  • PyPI publish: add skip-existing: true to avoid 400 errors on re-runs of the same tag.

Changed

  • Web UI: complete visual refresh — Inter + JetBrains Mono fonts, indigo design system, architecture health gauge, KPI tiles, ranked critical files list, selection grid, status bar, mini-map overlay.

[0.7.0] - 2026-04-24

Added

  • TypeScript parser with triple-slash reference support (/// <reference path="..." />).
  • Reusable GitHub Action (action.yml) — run ArchMAP analysis directly in any workflow with SARIF upload to GitHub Code Scanning.
  • Pre-commit hook via the archmap-check entrypoint — integrates with .pre-commit-hooks.yaml for zero-config quality gates.
  • Incremental parse cache — unchanged files are skipped on re-analysis, reducing runtime on large repos.
  • --sarif flag on archmap analyze — exports findings as SARIF 2.1.0 for upload to Code Scanning or SAST tools.

Changed

  • Parser registry now self-registers at import time; bootstrap.py is a no-op retained for backwards compatibility.
  • Web UI simplified: removed i18n layer, reduced bundle size.
  • archmap-check pre-commit entrypoint defaults: --quiet --fail-on-cycles --fail-on-custom-rules.

Removed

  • architecture_analyzer, architecture_suggester, history_analyzer, human_analyzer, impact_analyzer — these experimental analyzers were removed to reduce scope and maintenance burden.
  • explain, risk, improve, history CLI commands — backends removed; commands raise NotImplementedError and will be fully cleaned from the CLI in v0.8.0.
  • i18n layer from the web UI (i18n.js).

[0.4.0-beta.0] - 2026-03-09

Added

  • Multi-Language Registry: Overhauled the parser architecture to support plugin-based language detection. Added support for Go, PHP, Java, C#, C, and C++.
  • "UI of Respect" (Web UI v2):
  • Dark Mode Support: Full dark theme with session persistence.
  • Architectural Risks Panel: Live detection of circular dependencies and "Hub modules".
  • Dynamic Project Import: New folder icon in the UI allows switching the analysis target on the fly via a native directory picker (no restart required).
  • Real-time Refresh: Manual data re-polling for instant graph updates.
  • Respectful Executable (EXE):
  • Standalone Launcher: Double-clicking the EXE now automatically launches the interactive Service Map.
  • Persistent Error Log: Added a "Press Enter to exit" pause on errors in the frozen executable for easier troubleshooting.
  • Premium Branding: Embedded high-resolution project icon and professional CLI banner.

Fixed

  • CI Stabilization: Resolved 22 linting regressions and updated the test suite to ensure 100% compatibility with the new parser architecture.
  • Repository Cleanup: Removed redundant legacy JavaScript files from the root to ensure a clean Python-centric distribution.

[0.2.1] - 2026-03-07

Added

  • OSS Professionalization: Added Issue Templates (Bug/Feature), Code of Conduct, and detailed architecture & API documentation.
  • Documentation Site: Implemented MkDocs Material site with automated GitHub Pages deployment.
  • Professional CI: Expanded CI to test against Python 3.11, 3.12, and 3.13.
  • Automated Releases: Added GitHub Actions workflow to automate package building and releases on version tags.
  • Rich Examples: Enriched examples/sample-project with intentional cycles and layer violations for demonstration.
  • Benchmark Tool: Added scripts/benchmark.py for performance testing.

Fixed

  • Static Assets: Fixed "Web UI static directory not found" error in packaged (wheel) installations by using importlib.resources.
  • Parser Precision: Fixed dependency resolution for absolute from ... import ... submodules that were previously misidentified as package-level dependencies.

[0.1.3] - 2026-03-05

Added

  • Single File Analysis: Added support to natively read and resolve dependencies starting from a single file instead of a full directory scan, reducing overhead and providing concise maps for entry files.
  • Standalone Windows Executable (.exe): Bundled the CLI and the Web UI into a completely portable, zero-dependency Windows executable via PyInstaller. The interactive graph server (archmap.exe serve .) operates directly without a Python runtime requirement.

Changed

  • Rebuilt architecture file crawler (file_utils.discover_source_files) to intercept root files before deep recursive searches.

[0.2.0-beta.0] - 2026-03-04

Added

  • Full Python implementation under src/archmap.
  • New Python CLI with commands:
  • analyze
  • serve
  • diff
  • Architecture risk engine with:
  • god module detection
  • layer violation detection
  • dependency explosion detection
  • Cytoscape exporter (cytoscape_exporter.py).
  • pyproject.toml with package metadata and scripts.
  • Pytest suite for parser, analyzer, exporters, CLI, and diff.
  • CI pipeline with Ruff lint + pytest coverage + smoke analysis.

Changed

  • Canonical runtime migrated from Node.js to Python.
  • Repository structure aligned to src/archmap layout.
  • Branch strategy documented as:
  • feature/* -> dev -> release/* -> main

Notes

  • JavaScript implementation is no longer the primary runtime.
  • Next target: v0.3.0 architecture policy and trend analysis.