chore(deps): bump criterion from 0.5.1 to 0.8.2

.github/ISSUE_TEMPLATE/bug_report.md

@@ -1,33 +1,38 @@
 ---
 name: Bug Report
 about: Report a bug in m1nd
-title: "[bug] "
 labels: bug
-assignees: ''
 ---
 
-## Description
-<!-- What happened? -->
+**Describe the bug**
 
-## Steps to Reproduce
-1.
-2.
-3.
+A clear and concise description of what the bug is.
 
-## Expected Behavior
-<!-- What should have happened? -->
+**To Reproduce**
 
-## Environment
-- OS:
-- Rust version:
-- m1nd version:
-- MCP client: <!-- Claude Code, OpenCode, etc. -->
+Steps to reproduce the behavior:
+1. Run `m1nd.ingest path=...`
+2. Call `m1nd.activate query=...`
+3. Observe error
 
-## Graph Size
-- Nodes:
-- Edges:
+**Expected behavior**
+
+A clear description of what you expected to happen.
+
+**Actual behavior**
+
+What actually happened, including any error messages or unexpected output.
+
+**Environment**
+
+- OS: (e.g., macOS 14.2, Ubuntu 22.04)
+- Rust version: (output of `rustc --version`)
+- m1nd version: (output of `./m1nd-mcp --version` or git commit SHA)
+- MCP client: (e.g., Claude Code, Cursor, Windsurf)
+- Codebase size: (approximate nodes/edges from ingest output)
+
+**Logs / output**
 
-## Logs
-<!-- Paste relevant stderr output -->
 ```
+Paste relevant output here
 ```

.github/ISSUE_TEMPLATE/feature_request.md

@@ -1,19 +1,21 @@
 ---
 name: Feature Request
-about: Suggest a new tool or capability
-title: "[feature] "
+about: Suggest a new feature or tool
 labels: enhancement
-assignees: ''
 ---
 
-## Problem
-<!-- What problem does this solve? -->
+**Is your feature request related to a problem?**
 
-## Proposed Solution
-<!-- How should it work? -->
+A clear description of the problem. E.g., "When I call `activate()` on large monorepos, I can't distinguish results by semantic meaning vs. keyword proximity..."
 
-## Use Case
-<!-- Concrete example of when you'd use this -->
+**Describe the solution**
 
-## Alternatives Considered
-<!-- Other approaches you thought about -->
+What you'd like to happen. Be specific — which crate would this live in? What would the MCP tool signature look like?
+
+**Alternatives considered**
+
+Describe any workarounds or alternative solutions you've considered.
+
+**Additional context**
+
+Any other context, benchmarks, or references that would help implement this feature.

.github/PLANNED_ISSUES.md

@@ -0,0 +1,159 @@
+# Planned Issues — m1nd
+
+Issues staged for filing on GitHub. These represent the v0.3.0 and v0.4.0 roadmap.
+
+---
+
+## 1. feat: Embedding-based semantic similarity (fastembed)
+
+**Label**: enhancement, v0.3.0
+
+Replace keyword matching in `activate`/`seek` with vector embeddings for true semantic search.
+`fastembed` 5.12.1 is verified on crates.io and supports ONNX inference locally with no API calls.
+Currently deferred from v0.2.0 due to ONNX build complexity on cross-compiled targets.
+
+The semantic dimension in the 4D activation score currently uses trigram TF-IDF. This works for
+identifier overlap but fails on "find code that *means* authentication but never uses the word."
+Embedding-based similarity would close this gap while preserving zero-API-call semantics.
+
+---
+
+## 2. feat: ast-grep pattern integration for scan tool
+
+**Label**: enhancement, v0.3.0
+
+`ast-grep-core` 0.41.1 is available on crates.io and provides structural pattern matching via YAML
+rule files. The current `scan` tool in `m1nd-mcp` uses heuristic regex patterns against node
+identifiers, which produces false positives on minified code and misses structural patterns.
+
+Integrating `ast-grep-core` would allow rules like "find all functions that call X without
+checking the return value" — structural queries the regex approach cannot express.
+
+---
+
+## 3. feat: Desktop tray application
+
+**Label**: enhancement, v0.4.0
+
+A system tray app (Tauri or similar) that shows:
+- Graph health (node/edge counts, last ingest time)
+- Active agents + lock holders (from `lock.*` tools)
+- Quick-fire common queries without opening a terminal
+- Plasticity drift indicator (when the graph has diverged from disk state)
+
+Useful for teams where multiple agents share the same m1nd instance. The tray app would connect
+to the existing JSON-RPC stdio server via a local socket bridge.
+
+---
+
+## 4. perf: flow_simulate optimization for dense graphs
+
+**Label**: performance
+
+BFS particle propagation in `flow_simulate` (used internally by `activate` for the causal
+dimension) can be slow on graphs with >50K edges. In the current implementation, each particle
+is tracked individually through the BFS queue, leading to O(particles × edges) work.
+
+Observed: on a 50K-edge test graph, `activate` latency jumps from ~31ms to ~340ms when
+`infection_rate` is set above 0.4. The epidemic model saturates too quickly on dense subgraphs
+even at default settings (see related bug #9).
+
+Candidates: parallel particle simulation via Rayon, bitset-based batch propagation similar to
+the `RemovalMask` approach used in counterfactuals.
+
+---
+
+## 5. docs: Add video tutorial for first-time setup
+
+**Label**: documentation
+
+The README has solid text instructions and the wiki has detailed tool reference, but a video
+walkthrough would lower the friction for first-time users who want to see ingest → activate →
+learn in a real codebase before committing to building from source.
+
+Suggested content:
+1. Clone + `cargo build --release` (real terminal, real timing)
+2. Add to Claude Code config
+3. First `ingest` call — show what the output means
+4. `activate` → `why` → `missing` sequence
+5. `learn` feedback — explain Hebbian plasticity visually
+
+---
+
+## 6. feat: WebSocket transport for real-time graph updates
+
+**Label**: enhancement
+
+Currently the `m1nd-mcp` server is JSON-RPC over stdio only. A WebSocket transport would enable:
+- Bidirectional communication for a future GUI (m1nd-ui)
+- Real-time push of plasticity weight changes as agents learn
+- Browser-based graph visualization without a Tauri wrapper
+
+The stdio transport would remain the primary MCP path; WebSocket would be additive for
+GUI/dashboard use cases.
+
+---
+
+## 7. test: Increase test coverage for perspective navigation
+
+**Label**: testing
+
+`perspective.suggest` and `perspective.affinity` have minimal test coverage in the current
+test suite. `test_perspective_usecases.py` covers the happy path for `start`/`follow`/`back`
+but misses:
+- `suggest` with no open perspective session (should return graceful error)
+- `affinity` on disconnected graph components
+- `branch` + `compare` round-trip with conflicting node weights
+- Concurrent perspective sessions from two agents on the same graph
+
+These cases have produced silent wrong results in past manual testing.
+
+---
+
+## 8. feat: Git blame integration for temporal dimension
+
+**Label**: enhancement
+
+The temporal dimension currently uses co-change frequency and Hebbian feedback to weight recency.
+Git blame data would add actual authorship timestamps per line, allowing the temporal score to
+reflect real modification recency rather than just session-level feedback.
+
+Implementation path: `m1nd-ingest` already runs `git log --follow` for co-change history.
+Extending it to run `git blame --porcelain` per file and map blame timestamps to node weights
+would give the temporal dimension a ground-truth signal independent of agent feedback.
+
+---
+
+## 9. bug: epidemic model saturates too quickly on dense subgraphs
+
+**Label**: bug
+
+The BFS epidemic model used in `activate`'s causal dimension can reach 97% infection coverage
+in 3 iterations on dense subgraphs, even with `infection_rate=0.1` (the default).
+
+Steps to reproduce:
+1. Ingest a codebase with a highly-connected hub module (e.g., a shared `utils.py` with 200+
+   import edges)
+2. Call `activate(query="utils", agent_id=...)` with default parameters
+3. Observe: causal scores are uniformly high across nearly all nodes — signal saturates
+
+The v0.2.0 release partially mitigated this with early-stopping when coverage exceeds 80%, but
+edge cases remain when the hub is the query target itself (self-loop in the epidemic start node).
+
+Expected: causal scores should discriminate. Actual: they flatten to ~0.95 across the graph.
+
+---
+
+## 10. feat: Export graph to Neo4j/Cypher format
+
+**Label**: enhancement
+
+Currently `m1nd` supports Mermaid export (used for the architecture diagram in README). Adding
+Cypher export for Neo4j would let teams:
+- Visualize the full graph in Neo4j Browser or Bloom
+- Run Cypher queries alongside m1nd's activation queries
+- Integrate with existing graph databases in data engineering stacks
+
+The export would serialize the CSR graph to `.cypher` CREATE statements. Node properties would
+include the four dimension base scores; relationship properties would include the current
+plasticity weight.

.github/pull_request_template.md

@@ -1,22 +1,18 @@
 ## What
 
-<!-- One sentence: what does this PR do? -->
+Brief description of the change.
 
 ## Why
 
-<!-- Motivation or issue link -->
-
-## How
-
-<!-- Key changes, design decisions -->
+Motivation and context.
 
 ## Testing
 
-- [ ] `cargo test` passes
-- [ ] `cargo clippy` clean
-- [ ] `cargo fmt` applied
-- [ ] Tested with MCP client (tool name: ___)
+- [ ] `cargo test --workspace` passes
+- [ ] `cargo clippy --workspace` clean
+- [ ] Documentation updated if needed
+- [ ] New tools registered in `server.rs` dispatch
 
 ## Breaking Changes
 
-<!-- None / describe -->
+List any breaking changes to MCP tool signatures, or write "None".

.github/wiki/Home.md

@@ -96,6 +96,7 @@ Antibody system (bug immune memory), Epidemic engine (SIR propagation prediction
 | [Superpowers Extended](API-Reference#superpowers-extended-9-tools) | antibody_scan, antibody_list, antibody_create, flow_simulate, epidemic, tremor, trust, layers, layer_inspect | 9 |
 | [Surgical](API-Reference#surgical-4-tools) | surgical_context, apply, surgical_context_v2, apply_batch | 4 |
 | [v0.4.0 — Search & Efficiency](API-Reference#v040-search--efficiency-5-tools) | search, help, panoramic, savings, report | 5 |
+| [v0.5.0 — Verified Writes & File Tools](API-Reference#v050-verified-writes--file-tools) | apply_batch verify, view, glob | 3 |
 
 ---
 
@@ -160,7 +161,7 @@ Claude Code · Cursor · Windsurf · GitHub Copilot · Zed · Cline · Roo Code
 
 ## On crates.io
 
-m1nd v0.4.0 is published on [crates.io](https://crates.io/crates/m1nd-core). Add as a library dependency or build the binary from source — both paths are fully supported.
+m1nd v0.5.0 is published on [crates.io](https://crates.io/crates/m1nd-core). Add as a library dependency or build the binary from source — both paths are fully supported.
 
 ---
 

.github/wiki/Roadmap.md

@@ -82,6 +82,18 @@ The 5-layer verification pipeline runs automatically on every `apply` call:
 
 ---
 
+## Shipped — v0.5.0
+
+**Added in v0.5 — Verified writes, file tools, enhanced search (DONE):**
+- `apply_batch` **5-layer post-write verification** (`verify=true`) — trivial-return detection, compilation check, structural diff, semantic coherence, and regression guard
+- `m1nd.view` — lightweight file reader with auto-ingest. Zero-token alternative to Read for graph-tracked files
+- `m1nd.glob` — file pattern matching with auto-ingest. Graph-aware alternative to Glob
+- `m1nd.search` **enhanced** — invert matching, count mode, multiline, auto-ingest on miss, glob filtering
+
+**Total in v0.5: 63 MCP tools**
+
+---
+
 ## Shipped — v0.4.0
 
 **Added in v0.4 — Search, efficiency, and panoramic analysis (DONE):**

.github/workflows/ci.yml

@@ -1,4 +1,5 @@
 name: CI
+
 on:
   push:
     branches: [main, master]
@@ -7,17 +8,29 @@ on:
 
 env:
   CARGO_TERM_COLOR: always
-  RUSTFLAGS: -Dwarnings
 
 jobs:
-  test:
-    name: Test
+  check:
+    name: Check
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v4
       - uses: dtolnay/rust-toolchain@stable
       - uses: Swatinem/rust-cache@v2
-      - run: cargo test --all
+      - run: cargo check --workspace
+
+  test:
+    name: Test (${{ matrix.os }})
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest, macos-latest]
+    steps:
+      - uses: actions/checkout@v4
+      - uses: dtolnay/rust-toolchain@stable
+      - uses: Swatinem/rust-cache@v2
+      - run: cargo test --workspace
 
   clippy:
     name: Clippy
@@ -28,7 +41,7 @@ jobs:
         with:
           components: clippy
       - uses: Swatinem/rust-cache@v2
-      - run: cargo clippy --all -- -D warnings
+      - run: cargo clippy --workspace -- -D warnings
 
   fmt:
     name: Format
@@ -38,16 +51,17 @@ jobs:
       - uses: dtolnay/rust-toolchain@stable
         with:
           components: rustfmt
-      - run: cargo fmt --all -- --check
+      - run: cargo fmt --check
 
   build:
     name: Build (${{ matrix.os }})
     runs-on: ${{ matrix.os }}
     strategy:
+      fail-fast: false
       matrix:
         os: [ubuntu-latest, macos-latest]
     steps:
       - uses: actions/checkout@v4
       - uses: dtolnay/rust-toolchain@stable
       - uses: Swatinem/rust-cache@v2
-      - run: cargo build --release
+      - run: cargo build --release --workspace