@nodemint/projectmind - npm Package Compare versions

Comparing version 0.4.3 to 0.4.4

CHANGELOG.md

@@ -6,2 +6,18 @@ # Changelog
 
+## [0.4.4] - 2026-07-04
+
+### Changed
+- Strengthened `mind_digest`/`mind_context` tool descriptions and the
+  `setup`-generated rules block (`CLAUDE.md` etc.) to name the exact trigger
+  explicitly — "before running ls/find/glob/grep or reading files" — instead
+  of the vaguer "before reading source files," which real usage showed models
+  don't always map to directory listing.
+
+**Honesty note:** this is a best-effort prompt-engineering improvement, not a
+guarantee. No MCP server (this one included) can force a model to call a
+specific tool — the agent still weighs the nudge against its own instincts.
+If you see an agent skip `mind_digest` on an orientation question, that's a
+known, inherent limit of instruction-following, not a wiring bug — verify the
+tool itself works by asking the agent to call it directly.
+
 ## [0.4.3] - 2026-07-03
@@ -8,0 +24,0 @@

package.json

 {
   "name": "@nodemint/projectmind",
-  "version": "0.4.3",
+  "version": "0.4.4",
   "mcpName": "io.github.Nodemint-dev/projectmind",
@@ -5,0 +5,0 @@ "publishConfig": {

src/mcp/server.js

@@ -16,3 +16,3 @@ #!/usr/bin/env node
     description:
-      "Read this FIRST at the start of any task, before reading source files or asking the user about project structure. Returns a compact map of modules, dependencies, decisions, and conventions. Reading this instead of scanning the codebase is the primary way this project saves you tokens.",
+      "Call this BEFORE running ls, find, glob, grep, tree, or reading any file, whenever the task is to explain, describe, summarize, or give an overview of this project or its architecture, or when you otherwise need to orient yourself in an unfamiliar codebase. This includes questions like \"what is this project\", \"explain this codebase\", \"what's the tech stack\", or \"how is this structured\" — call mind_digest first, THEN use its output to decide whether further file exploration is even needed. Returns a compact map of modules, dependencies, decisions, and conventions in a few hundred tokens, replacing what would otherwise be many file reads.",
     inputSchema: { type: "object", properties: {}, additionalProperties: false },
@@ -23,3 +23,3 @@ },
     description:
-      "Task-scoped context. Given the files you're about to edit (and/or a node id or keyword), returns ONLY the relevant part of the map: those modules in full plus their direct dependencies. Use this over mind_digest when you already know which files a task touches — it's the most surgical, token-cheap option.",
+      "Call this BEFORE reading/greping files, when you're about to work on specific files or a module and want its purpose, notes, and direct dependencies without opening them. Given the files you're about to edit (and/or a node id or keyword), returns ONLY the relevant part of the map. Use this over mind_digest when you already know which files a task touches — it's the most surgical, token-cheap option.",
     inputSchema: {
@@ -184,3 +184,3 @@ type: "object",
   const server = new Server(
-    { name: "projectmind", version: "0.4.3" },
+    { name: "projectmind", version: "0.4.4" },
     { capabilities: { tools: {} } }
@@ -187,0 +187,0 @@ );

src/setup/index.js

@@ -26,3 +26,3 @@ // Multi-agent wiring. Writes the MCP server config and a workflow rules block
   "## projectmind",
-  "At the start of a task, call `mind_digest` before reading source files or asking about project structure.",
+  "Before running ls/find/glob/grep or reading files to explain, describe, or orient in this project (e.g. \"what is this project\", \"explain this codebase\", \"how is this structured\"), call `mind_digest` first — it answers most of that in a few hundred tokens.",
   "Use `mind_context({ files })` for a task-scoped subgraph, or `mind_query(<id>)` for one module's files/notes.",
@@ -29,0 +29,0 @@ "After a structural change, architectural decision, or newly learned convention, call `mind_update` (only the fields that changed).",

New alerts

AI-detected potential code anomaly

Supply chain risk

AI has identified unusual behaviors that may pose a security risk.

Found 2 instances

Fixed alerts

AI-detected potential code anomaly

Supply chain risk

AI has identified unusual behaviors that may pose a security risk.

Improved metrics

Total package byte prevSize

105390

1.4%

Lines of code

1702

0.12%

Worsened metrics

Number of low supply chain risk alerts

10

11.11%

No dependency changes