Handles methods, classes, modules, attr_* properties, require imports,
and call graph including bare identifier calls (no-arg method invocations).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Implements the LanguageExtractor interface for Go, handling functions,
methods with receivers, structs, interfaces, imports, exports (via
capitalization convention), and call graph extraction. Includes 25 tests.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Implements the LanguageExtractor interface for Python, extracting functions
(with type annotations, defaults, *args/**kwargs), classes (methods +
annotated properties), imports (plain, from, aliased, wildcard), exports
(top-level defs), and caller-callee call graphs. Includes 31 tests using
the real tree-sitter parser.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Extract all TypeScript/JavaScript-specific AST extraction functions
(extractParams, extractReturnType, extractImportSpecifiers, processTopLevelNode,
extractFunction, extractClass, extractVariableDeclarations, extractImport,
processExportStatement, and call graph walking) from TreeSitterPlugin into the
new TypeScriptExtractor class. TreeSitterPlugin now dispatches to registered
LanguageExtractor instances, defaulting to TypeScriptExtractor for backward
compatibility. All 426 existing tests pass unchanged.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
GraphBuilder maintained its own ~60-line extension-to-language mapping that
duplicated and could diverge from the canonical LanguageRegistry. Now delegates
language detection to LanguageRegistry.getForFile(), eliminating the duplication
and ensuring new language configs are automatically picked up everywhere.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
CLI --open overrides server.open config, stripping the token. Remove
--open from CLI and set server.open to /?token= so the browser opens
with the access token already included.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
All five child-node loops in addNonCodeFileWithAnalysis shared the same four-line pattern: dedup check, nodeIds.add, nodes.push, and a contains edge push. A private addChildNode helper centralises this so each loop only constructs the node object specific to its type.
Default sort() uses Unicode code point ordering which can produce unexpected results for non-ASCII strings. localeCompare guarantees correct alphabetical ordering regardless of character set.
build() was returning direct references to the builder's internal arrays, allowing callers to mutate graph.nodes or graph.edges and corrupt the builder's state. Spreading into new arrays at build time prevents this at negligible cost.
addImportEdge and addCallEdge previously pushed edges unconditionally, allowing duplicate relationships if multiple agents reported the same import or call. A shared edgeKeys set keyed on type|source|target silently skips any edge that has already been recorded.
The endpoint name template was evaluated twice — once for name and once for summary. A single const removes the duplication and resolves the nested template literal lint warning.
addNonCodeFileWithAnalysis was reconstructing the fileId string independently of addNonCodeFile, creating a silent correctness risk if the ID construction logic ever changed. addNonCodeFile now returns the ID it used so the caller cannot go out of sync.
The same filePath.split("/").pop() ?? filePath expression appeared three times across addFile, addFileWithAnalysis, and addNonCodeFile. A private static helper centralises the logic and makes call sites easier to read.
Previously addNonCodeFileWithAnalysis rebuilt a full Set from this.nodes on every call, making duplicate checks O(n) per file and O(n²) overall. Moving nodeIds to a class field and updating it incrementally at each insertion reduces duplicate detection to O(1) per check.
Previously the mapping object was recreated on every mapKindToNodeType call. Moving it to module level means it is allocated once at load time instead of once per definition node processed.
marketplace.json declares plugins[0].source as ./understand-anything-plugin,
so Claude Code resolves the plugin manifest at
./understand-anything-plugin/.claude-plugin/plugin.json. That file did not
exist — only .claude-plugin/plugin.json at the repo root — causing
"Marketplace sync failed. Some plugins in this marketplace have validation
errors." when adding the marketplace via /plugin marketplace add.
Adds the manifest at the path the marketplace source expects, so marketplace
sync validates and the plugin can be installed. Version matches the current
root plugin.json (2.3.1).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The skill's inline Node.js generator was duplicated from core and missed
the .gitignore seeding and bin/ header fix. Updated the one-liner to read
.gitignore patterns, deduplicate against defaults with trailing-slash
normalization, and remove bin/ from the built-in defaults header comment.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Read the project's .gitignore at starter-file generation time and include
non-default patterns as commented suggestions in .understandignore. Patterns
already covered by hardcoded defaults are deduplicated (with trailing-slash
normalization). This is a one-time inclusion — users can later remove patterns
for files they want analyzed without the filter re-reading .gitignore.
Also fixes the starter header listing bin/ as a built-in default when it is
intentionally excluded from DEFAULT_IGNORE_PATTERNS (bin/ is used by Node/Ruby
CLI launchers).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Connected edges now highlight (thicker, full opacity) when a node is
selected/focused. Unconnected edges dim to near-invisible. Edge labels
only show on connected edges to reduce visual noise.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
P1: Separate force layout computation from visual state updates so
clicking/searching/touring doesn't re-randomize node positions. Layout
only recomputes when the graph data or filters change.
P1: Restrict infrastructure-file skipping (index.md, log.md, etc.) to
the wiki root level only. Nested files like concepts/index.md are now
correctly treated as content articles.
P2: Track ambiguous bare basenames in the wikilink resolution map.
Duplicate basenames (e.g., a/foo.md and b/foo.md) are removed from
the flat lookup so [[foo]] doesn't silently resolve to the wrong page.
Also fixed: edge IDs now use stable source-target-type keys instead of
array indices for proper React reconciliation.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Remove bin/ from DEFAULT_IGNORE_PATTERNS (Node/Ruby CLI launchers use bin/)
- .NET users can add bin/ to .understandignore manually
- Fix project-scanner Step 2.5 to re-filter from original file list when
.understandignore exists, ensuring ! negation correctly overrides defaults
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The understand-dashboard skill now instructs agents to capture and
share the full tokenized URL from the Vite server output, so users
are not blocked by the token gate.
Bump version to 2.2.1.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Prevents TypeError crash when LLM batches emit mixed weight types
(e.g. 0.8 vs "0.7") during edge deduplication.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add Python scripts to merge knowledge graphs (closes#70) and move
mechanical normalization out of LLM context into deterministic scripts
with diagnostic reporting. Convert all agent definitions from dispatch
templates to self-contained system prompts to prevent instruction loss.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add a demo mode to the dashboard that fetches graph data from Supabase
Storage instead of the local Vite dev server, bypasses the token gate,
and deploys alongside the homepage at /Understand-Anything/demo/.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Instead of truncating cross_domain edge labels, compute ranksep from
the longest label length (~6px/char) so dagre spaces nodes further
apart. Also add label background for readability, and support
spacingOverrides parameter in applyDagreLayout.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
1. Add infrastructure edge category to EDGE_CATEGORY_MAP for 8 missing
edge types (deploys, serves, provisions, triggers, migrates, documents,
routes, defines_schema) — previously unfiltered regardless of toggles
2. navigateToDomain now switches viewMode to "domain" — previously
clicking flow buttons from structural mode was a silent no-op
3. ExportMenu JSON export for non-technical persona now keeps all
file-level types — previously dropped config/document/service/etc.
4. ProjectOverview category breakdown now includes module/concept in
Code and domain/flow/step in new Domain category
5. SearchBar type badge colors now cover all 16 node types — previously
only 5 had colors, rest fell back to file color
6. PathFinder BFS now traverses edges bidirectionally — previously
only followed forward direction, missing valid reverse paths
7. GraphView node filtering consolidated into single authoritative
type set — previously fileLevelTypes and persona conditions were
separate lists that could drift
8. SelectedNodeFitView delayed 100ms to run after layer-level fitView
on search-click navigation — previously layer fit clobbered node fit
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Function and class nodes were invisible in the graph because they
are not directly assigned to layers (only file-level nodes are).
Expand layer membership by following `contains` edges from file nodes
in the active layer to include their child function/class nodes.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The previous wording ("MUST have sub-nodes") implied every code file
must produce function nodes, which is wrong for files without significant
functions. Now correctly states: if significant functions exist in the
script output, you must create nodes for them.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Delete the 5 prompt template files from skills/understand/ now that
they live in agents/. Update auto-update-prompt.md hook to reference
agent definitions instead of deleted prompt files.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Prompt templates (file-analyzer, project-scanner, architecture-analyzer,
tour-builder, graph-reviewer) were being compressed by the orchestrator
when dispatched as subagent prompts, causing function/class extraction
to be silently skipped. Moving them to agents/ ensures the framework
loads the full prompt without compression.
- Move 5 prompt templates from skills/understand/ to agents/
- Update SKILL.md to reference agent definitions instead of templates
- Set all agent models to `inherit` for cross-platform compatibility
- Update CLAUDE.md to reflect new agent model policy
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Remove direction-inverting `implemented_by` alias (same pattern as fd0df15)
- Replace ambiguous `process` alias with `business_process`
- Fix duplicate React Flow edge IDs in DomainGraphView
- Fix navigateToDomain clearing selectedNodeId and losing history
- Preserve domain viewMode when structural graph loads after domain graph
- Add domain/flow/step to fileLevelTypes in GraphView
- Add domain edge category to EDGE_CATEGORY_MAP
- Extend COMPLEXITY_STRING_MAP with trivial/basic/mid/average/advanced
- Normalize string complexity values in normalizeBatchOutput (not just numeric)
- Infer node type from ID prefix in edge fallback normalization
- Include flow discriminator in bare-path step ID normalization
- Clean up domain-context.json intermediate file in SKILL.md
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Lum1104andClaude Opus 4.6
JohnWongandClaude Opus 4.6
Nikola Chetelyazov
jiachenandClaude Opus 4.6