weavatrix 1.17.1 → 1.17.3

package/README.md

@@ -58,7 +58,7 @@ weavatrix mcp .
 ```toml
 [mcp_servers.weavatrix]
 command = "npx"
-args = ["-y", "weavatrix@1.17.1", "mcp", "."]
+args = ["-y", "weavatrix@1.17.3", "mcp", "."]
 ```
 
 Pass an absolute repository path when the Codex process cwd is not the project
@@ -67,7 +67,7 @@ you intend to analyze.
 ### Claude Code
 
 ```sh
-claude mcp add weavatrix -- npx -y weavatrix@1.17.1 mcp .
+claude mcp add weavatrix -- npx -y weavatrix@1.17.3 mcp .
 ```
 
 ### Cursor
@@ -80,7 +80,7 @@ the plugin for the same server name:
   "mcpServers": {
     "weavatrix": {
       "command": "npx",
-      "args": ["-y", "weavatrix@1.17.1", "mcp", "${workspaceFolder}"]
+      "args": ["-y", "weavatrix@1.17.3", "mcp", "${workspaceFolder}"]
     }
   }
 }
@@ -89,14 +89,14 @@ the plugin for the same server name:
 Profiles expose bounded views of the same engine:
 
 ```sh
-npx -y weavatrix@1.17.1 mcp . --profile=all
-npx -y weavatrix@1.17.1 mcp . --profile=code
-npx -y weavatrix@1.17.1 mcp . --profile=seo
-npx -y weavatrix@1.17.1 mcp . --profile=n8n
-npx -y weavatrix@1.17.1 mcp . --profile=dify
-npx -y weavatrix@1.17.1 mcp . --profile=agent
-npx -y weavatrix@1.17.1 mcp . --profile=diagram
-npx -y weavatrix@1.17.1 mcp . --profile=web3
+npx -y weavatrix@1.17.3 mcp . --profile=all
+npx -y weavatrix@1.17.3 mcp . --profile=code
+npx -y weavatrix@1.17.3 mcp . --profile=seo
+npx -y weavatrix@1.17.3 mcp . --profile=n8n
+npx -y weavatrix@1.17.3 mcp . --profile=dify
+npx -y weavatrix@1.17.3 mcp . --profile=agent
+npx -y weavatrix@1.17.3 mcp . --profile=diagram
+npx -y weavatrix@1.17.3 mcp . --profile=web3
 ```
 
 The package contains native binaries for Windows x64/arm64, macOS x64/arm64,
@@ -250,12 +250,18 @@ Every operation is read-only with respect to the analyzed repository.
 Pagination and explicit limits bound large neighborhoods, histories, searches,
 and contract inventories.
 
-`architecture_inventory` describes observed components without assigning an
-architecture style. `ci_restrictions` reads local GitHub Actions workflows and
-recognizes literal checks, with optional event, base branch, and changed-path
-scenario. `explain_restriction` retrieves one finding. Dynamic conditions,
-actual runs, and remote required checks remain unknown until separate evidence
-is supplied; a workflow file alone never establishes a passing gate.
+`build_graph` and `architecture_inventory` share one typed build model. They
+preserve nested Cargo, npm/TypeScript, Go, and Python components, many-to-many
+declared memberships, collision-free identities, configuration predicates,
+and immutable source evidence. Totals and SCC/cycle witnesses are computed
+before pagination, so a response cap cannot change the graph verdict.
+
+`ci_restrictions` reuses that target-aware topology while reading local GitHub
+Actions workflows and literal checks. Root files, repeated commands, UTF-8 BOM
+offsets, optional events, base branches, and changed-path scenarios keep exact
+evidence. Dynamic conditions, mutually exclusive configuration unions, actual
+runs, and remote required checks remain explicit unknowns; a workflow file
+alone never establishes a passing gate.
 
 Need measured coverage? Run Weavatrix Quality first, then `coverage_map`.
 The host will not invent percentages from an empty search path. Quality
@@ -310,11 +316,11 @@ coding agent
     |
     | MCP over stdio
     v
-weavatrix 1.17.1
+weavatrix 1.17.3
     profile catalog · refresh · watcher · MCP framing
     |
     v
-weavatrix-rust 2.17.0
+weavatrix-rust 2.17.2
     typed graph · analysis · 67 product operations including local CI, n8n, Dify, agent packages, Mermaid, and Web3
 ```
 

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "weavatrix",
-  "version": "1.17.1",
-  "weavatrixEngineVersion": "2.17.0",
+  "version": "1.17.3",
+  "weavatrixEngineVersion": "2.17.2",
   "description": "Native repository intelligence for coding agents: 67 read-only MCP operations backed by typed evidence graphs, impact, architecture, local CI restrictions, APIs, Git, search, semantics, memory, n8n workflows, Dify apps, agent packages, Mermaid diagrams, and Web3 ABIs.",
   "author": "Sergii Ziborov <sergii.ziborov@gmail.com>",
   "license": "MIT",

package/server.json

@@ -7,14 +7,14 @@
     "url": "https://github.com/Weavatrix/weavatrix",
     "source": "github"
   },
-  "version": "1.17.1",
+  "version": "1.17.3",
   "websiteUrl": "https://weavatrix.com",
   "packages": [
     {
       "registryType": "npm",
       "registryBaseUrl": "https://registry.npmjs.org",
       "identifier": "weavatrix",
-      "version": "1.17.1",
+      "version": "1.17.3",
       "runtimeHint": "npx",
       "transport": {
         "type": "stdio"

package/skill/SKILL.md

@@ -23,8 +23,16 @@ contracts. Do not call it merely because the server is available.
 1. Call `graph_stats` and confirm `repository_context.root` matches the workspace
    you intend. The MCP process is pinned to its launch root; it will not silently
    answer for another repository.
-2. Use `module_map` or `architecture_inventory` for orientation, or
-   `search_code` for a known literal.
+2. For "what is this repository's architecture?", call
+   `architecture_inventory` with no `detail` argument (bounded summary).
+   Explain its packages, component areas, dominant coupling, and limits in
+   your own concise prose; do not paste the tool's JSON as the answer. Use
+   `module_map` for a directory-focused question, or `search_code` for a
+   known literal. Request `architecture_inventory` with `detail:"full"`
+   only when a specific component, edge, or cycle needs source evidence.
+   Use `get_architecture_contract` or `verify_architecture` only when the
+   question is about declared target policy or compliance. A component
+   quotient/union cycle is a candidate, not proof of a runtime cycle.
 3. Pin decisive evidence with `inspect_symbol`, `go_to_definition`,
    `find_references`, `context_bundle`, or `read_source`.
 4. Expand only when needed: `get_dependents` or `change_impact` for risk,
@@ -52,8 +60,10 @@ Do not call `open_repo` "just in case". Cross-repository tools take explicit
 roots in their arguments. Multi-root retarget requires launching with
 `--allow-retarget`.
 
-Prefer `output_format:"text"` and a small `token_budget` for conversational
-work. Use JSON only for automation or retained evidence. Call `rebuild_graph`
+`output_format` controls the MCP envelope, not the amount of detail: `text`
+still contains JSON. Choose each tool's summary/detail or result limits for
+conversational work. Use `token_budget` only as an additional ceiling; it can
+drop evidence arrays. Call `rebuild_graph`
 only when the repository changed before automatic refresh completed or when
 deliberately changing graph mode.
 

package/skill/references/tool-routing.md

@@ -19,7 +19,7 @@ cannot silently switch to another repository. Pass an explicit
 - [`get_community`](tools/get_community.md) — Return one coupling community (containment and package edges excluded).
 - [`list_communities`](tools/list_communities.md) — List deterministic coupling communities (containment and package edges excluded).
 - [`module_map`](tools/module_map.md) — Production folder map at a requested directory depth.
-- [`build_graph`](tools/build_graph.md) — Workspace, target and runner topology from manifest evidence.
+- [`build_graph`](tools/build_graph.md) — Typed workspace, target, configuration, task, and runner topology from captured manifest evidence.
 - [`search_code`](tools/search_code.md) — Literal or Rust-regex repository search without ripgrep.
 - [`read_source`](tools/read_source.md) — Bounded source context by node or repository path.
 - [`inspect_symbol`](tools/inspect_symbol.md) — Definition, direct relationships and source evidence.
@@ -41,7 +41,7 @@ cannot silently switch to another repository. Pass an explicit
 
 ## Quality and architecture
 
-- [`architecture_inventory`](tools/architecture_inventory.md) — Observed components and typed edges without a style label.
+- [`architecture_inventory`](tools/architecture_inventory.md) — Bounded architecture overview by default, with complete edge/cycle-candidate totals; `detail:"full"` gives paged evidence. Explain the overview in prose, not as a JSON dump.
 - [`ci_restrictions`](tools/ci_restrictions.md) — Local GitHub Actions checks and conditional applicability.
 - [`explain_restriction`](tools/explain_restriction.md) — One restriction and its evidence limits.
 - [`find_duplicates`](tools/find_duplicates.md) — Deterministic Type-1/2/3 clone families.

package/skill/references/tools/architecture_inventory.md

@@ -1,22 +1,40 @@
 # `architecture_inventory`
 
-Observed package, component, and typed edge evidence. No style label or
-target architecture is inferred.
+Observed package, nested component, declared-membership, and typed edge
+evidence. The default is a bounded architecture overview, not a raw graph
+dump. No style label or target architecture is inferred.
 
 ## When to use
 
 Orient in an unfamiliar repository or compare observed structure with a
 separately declared architecture contract.
 
+Component identities distinguish paths that sanitize alike, including the
+repository root. Files can retain every matching declared membership. Edge
+totals and component SCC/cycle witnesses are computed before presentation
+limits, and each recovered fact keeps its immutable source evidence. A cycle in
+the component quotient is reported as architecture, not as symbol recursion.
+
 ## Inputs
 
-- `output_format` (`text`, `json`, or `structured`).
+- `detail`: `summary` (default) or `full`.
+- `max_results` (1–500) and `edge_cursor`: page edges in `full` mode. Supplying
+  either without `detail` implies `full`; neither shrinks the summary.
+- `token_budget`: additional output ceiling; may drop samples or evidence.
+- `output_format` (`text`, `json`, or `structured`) changes only the MCP
+  representation. `text` still contains serialized JSON, not prose.
 - `expected_repository` (optional active-root guard).
 
 ## Minimal call
 
 ```json
-{"name":"architecture_inventory","arguments":{"output_format":"text"}}
+{"name":"architecture_inventory","arguments":{}}
 ```
 
+Turn the returned packages, component paths, dominant coupling, and complete
+edge/cycle-candidate totals into a concise answer in your own words. Never
+present a quotient/union cycle candidate as a proven executable cycle. Ask
+for `{"detail":"full"}` only to inspect a specific edge, component, or
+cycle witness. The full response can be large and paginates edges.
+
 The live MCP `tools/list` schema is authoritative.

package/skill/references/tools/build_graph.md

@@ -1,10 +1,14 @@
 # `build_graph`
 
-Workspace, target and runner topology from manifest evidence.
+Typed workspace, package, target, task, and runner topology from captured
+manifest evidence.
 
 ## When to use
 
-Inspect workspace members, targets, aggregators, and runner topology extracted from manifests.
+Inspect Cargo, npm/TypeScript, Go, and Python workspace members, targets,
+aggregators, entry points, configuration variants, and runner topology. The
+same model feeds architecture and local CI linkage; source evidence is captured
+with the analyzed snapshot rather than reread from a changing worktree.
 
 ## Inputs
 
@@ -24,3 +28,7 @@ Inspect workspace members, targets, aggregators, and runner topology extracted f
 
 Use `output_format:"text"` for compact agent interaction. The live MCP
 `tools/list` schema remains authoritative for this installed version.
+
+`max_members` limits presentation only. Returned totals still describe the
+complete recovered model, and the result reports an incomplete page instead of
+pretending omitted members do not exist.