memtrace 0.8.63 → 1.0.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (8) hide show
  1. package/hooks/userprompt-claude.sh +14 -5
  2. package/installer/package.json +1 -1
  3. package/installer/skills/commands/memtrace-quality.md +17 -5
  4. package/installer/skills/workflows/memtrace-refactoring-guide.md +2 -2
  5. package/lib/cuda-ep.js +3 -3
  6. package/package.json +7 -7
  7. package/skills/commands/memtrace-quality.md +17 -5
  8. package/skills/workflows/memtrace-refactoring-guide.md +2 -2
package/hooks/userprompt-claude.sh CHANGED
@@ -162,11 +162,20 @@ LOCK_FILE="$LOCK_DIR/$SESSION_ID.lock"
162
162
  # entirely so every fire proceeds (useful for debugging & tests).
163
163
  if (( DEBOUNCE_SECS > 0 )) && [[ -f "$LOCK_FILE" ]]; then
164
164
  NOW=$(date +%s)
165
- # `stat -f %m` is BSD/macOS, `stat -c %Y` is GNU/Linux. Try
166
- # both; if neither works (extremely unusual) we treat the lock
167
- # as fresh enough to suppress, on the principle that "we just
168
- # touched it" is the safer default than "spam the daemon".
169
- LAST="$(stat -f %m "$LOCK_FILE" 2>/dev/null || stat -c %Y "$LOCK_FILE" 2>/dev/null || printf '%s' "$NOW")"
165
+ # `stat -c %Y` is GNU/Linux, while `stat -f %m` is BSD/macOS.
166
+ # GNU stat accepts `-f` for a different operation and exits zero,
167
+ # so an `cmd_a || cmd_b` fallback can capture filesystem details
168
+ # instead of an mtime. Validate each result before accepting it.
169
+ LAST=""
170
+ if LAST="$(stat -c %Y "$LOCK_FILE" 2>/dev/null)" && [[ "$LAST" =~ ^[0-9]+$ ]]; then
171
+ :
172
+ elif LAST="$(stat -f %m "$LOCK_FILE" 2>/dev/null)" && [[ "$LAST" =~ ^[0-9]+$ ]]; then
173
+ :
174
+ else
175
+ # If neither form works (extremely unusual), treat the lock as
176
+ # freshly touched rather than spamming the daemon.
177
+ LAST="$NOW"
178
+ fi
170
179
  if [[ "$LAST" =~ ^[0-9]+$ ]]; then
171
180
  AGE=$((NOW - LAST))
172
181
  if (( AGE < DEBOUNCE_SECS )); then
package/installer/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "memtrace-skills",
3
- "version": "0.8.63",
3
+ "version": "1.0.1",
4
4
  "description": "Memtrace skills for AI coding agents — codebase exploration, temporal evolution, impact analysis, and more.",
5
5
  "type": "module",
6
6
  "bin": {
package/installer/skills/commands/memtrace-quality.md CHANGED
@@ -9,7 +9,7 @@ allowed-tools:
9
9
  - mcp__memtrace__get_evolution
10
10
  metadata:
11
11
  author: "Syncable <support@syncable.dev>"
12
- version: "1.0.0"
12
+ version: "1.1.0"
13
13
  category: development
14
14
  user-invocable: true
15
15
  ---
@@ -22,7 +22,7 @@ Code quality via graph analysis — dead code, complexity hotspots, repository s
22
22
 
23
23
  | Tool | Purpose | Required | Key optional |
24
24
  |------|---------|----------|--------------|
25
- | `find_dead_code` | Zero-caller symbols | `repo_id` | `include_tests`, `limit`, `kinds[]` |
25
+ | `find_dead_code` | Symbols unreachable from roots | `repo_id` | `include_tests`, `limit`, `min_confidence`, `include_historical` |
26
26
  | `find_most_complex_functions` | Ranked complexity hotspots | `repo_id` | **`top_n`** (def 20) — not `limit` |
27
27
  | `calculate_cyclomatic_complexity` | Score one symbol | `repo_id`, **`target`** | — |
28
28
  | `get_repository_stats` | Repository overview | `repo_id` | `branch` |
@@ -42,10 +42,20 @@ Full parameter spec for every Memtrace tool: `references/mcp-parameters.md` (bun
42
42
  ### 2. Dead code
43
43
 
44
44
  ```json
45
- { "repo_id": "memdb", "include_tests": false, "limit": 50 }
45
+ { "repo_id": "memdb", "include_tests": false, "limit": 50, "min_confidence": "high" }
46
46
  ```
47
47
 
48
- Exported symbols and entry points excluded by default. Results are candidates, not proof — callers via dynamic dispatch or reflection are invisible to the graph; verify before deleting.
48
+ `find_dead_code` seeds **roots** (`main`, exported symbols, HANDLES targets,
49
+ `trait_impl`, Magalz/framework allowlists), BFS-walks CALLS / REFERENCES /
50
+ MAKES_API_CALL forward, and reports Function/Method nodes **unreachable from
51
+ roots**. Each row includes `confidence` (`high` / `medium` / `low`) and
52
+ `confidence_score` (Skylos-style penalties for dynamic languages, ambiguous
53
+ hook names, methods). Use `min_confidence: "high"` for CI-safe cleanup;
54
+ omit or `"low"` to see all bands.
55
+
56
+ Results are still candidates, not proof — unmodeled dynamic dispatch can look
57
+ live or dead incorrectly. Verify before deleting. Prefer false-live over
58
+ false-dead.
49
59
 
50
60
  ### 3. Complexity hotspots
51
61
 
@@ -72,7 +82,7 @@ Ordered by call-graph out-degree.
72
82
 
73
83
  | Result | Carries |
74
84
  |--------|---------|
75
- | Dead-code entry (`find_dead_code`) | Symbol name + kind with zero graph callers — e.g. `format_legacy` (Function) |
85
+ | Dead-code entry (`find_dead_code`) | Name, kind, path, `reason` (`unreachable from roots`), `confidence`, `confidence_score` |
76
86
  | Complexity row (`find_most_complex_functions`) | Symbol, complexity score, risk level — e.g. `processOrder` — 27, Critical |
77
87
  | Single score (`calculate_cyclomatic_complexity`) | Cyclomatic complexity for `target` |
78
88
  | Stats (`get_repository_stats`) | Node counts by kind, edge counts, community and process counts for `repo_id`/`branch` |
@@ -81,6 +91,8 @@ Ordered by call-graph out-degree.
81
91
 
82
92
  | Mistake | Reality |
83
93
  |---------|---------|
94
+ | Treating every dead-code row as deletable | Filter with `min_confidence: "high"`; still verify framework/dynamic use |
95
+ | Assuming “no inbound edge” is the model | Dead means unreachable from roots, not merely unreferenced |
84
96
  | `find_most_complex_functions(limit: 10)` | Param is **`top_n`** |
85
97
  | `calculate_cyclomatic_complexity(symbol_id=...)` | Required param is **`target`** |
86
98
  | Only looking at the highest complexity | Medium-complexity functions that are growing (check `get_evolution`) are often more urgent |
package/installer/skills/workflows/memtrace-refactoring-guide.md CHANGED
@@ -35,7 +35,7 @@ Run these three tools in parallel to build a candidate list:
35
35
  Call `find_most_complex_functions` with `top_n: 20`
36
36
 
37
37
  **b) Dead code:**
38
- Call `find_dead_code` to find unused symbols
38
+ Call `find_dead_code` (prefer `min_confidence: "high"`) for symbols unreachable from roots
39
39
 
40
40
  **c) Architectural bottlenecks:**
41
41
  Call `find_bridge_symbols` to find chokepoints with too much responsibility
@@ -122,7 +122,7 @@ For each item, include:
122
122
  |-----------|--------|
123
123
  | Complex + volatile + high blast radius | Highest priority — but plan carefully; incremental approach |
124
124
  | Complex + stable + low blast radius | Can wait; refactor when you're already touching nearby code |
125
- | Dead code with zero callers | Run Cortex provenance/recall first; zero callers is not proof that no decision/contract keeps it |
125
+ | Dead code (`unreachable from roots`, high confidence) | Run Cortex provenance/recall first; reachability is not proof that no decision/contract keeps it |
126
126
  | Bridge symbol with many dependents | Extract interface first, then refactor implementation behind it |
127
127
  | Symbol in cross-repo API | Coordinate with consumers; backward-compatible changes only |
128
128
  | Cortex returns a held ban/contract | Preserve it or ask before overriding it |
package/lib/cuda-ep.js CHANGED
@@ -8,9 +8,9 @@
8
8
  // ships as a GitHub release asset instead (`memtrace-windows-x86_64-
9
9
  // cuda-ep.zip`, sha256 recorded in the release's `checksums.txt`), and
10
10
  // this module fetches it at install time for hosts that can actually
11
- // use it. The engine probes CUDA -> DirectML -> CPU at startup, so a
12
- // missing DLL is never fatal — NVIDIA users just silently land on
13
- // DirectML, which is what this fetch exists to avoid.
11
+ // use it. The bundled Windows runtime probes CUDA -> CPU at startup, so
12
+ // a missing DLL is never fatal — NVIDIA users safely land on CPU. A
13
+ // custom ONNX Runtime can additionally expose DirectML.
14
14
  //
15
15
  // Placement contract: the engine's dylib discovery looks next to the
16
16
  // running binary (`bridge_ort_dylib_env` in the Rust CLI), so the DLL
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "memtrace",
3
- "version": "0.8.63",
3
+ "version": "1.0.1",
4
4
  "description": "Code intelligence graph — MCP server + AI agent skills + visualization UI",
5
5
  "keywords": [
6
6
  "mcp",
@@ -40,12 +40,12 @@
40
40
  "fs-extra": "^11.0.0"
41
41
  },
42
42
  "optionalDependencies": {
43
- "@memtrace/darwin-arm64": "0.8.63",
44
- "@memtrace/linux-x64": "0.8.63",
45
- "@memtrace/linux-arm64": "0.8.63",
46
- "@memtrace/win32-x64": "0.8.63",
47
- "@memtrace/linux-x64-noavx2": "0.8.63",
48
- "@memtrace/win32-x64-noavx2": "0.8.63"
43
+ "@memtrace/darwin-arm64": "1.0.1",
44
+ "@memtrace/linux-x64": "1.0.1",
45
+ "@memtrace/linux-arm64": "1.0.1",
46
+ "@memtrace/win32-x64": "1.0.1",
47
+ "@memtrace/linux-x64-noavx2": "1.0.1",
48
+ "@memtrace/win32-x64-noavx2": "1.0.1"
49
49
  },
50
50
  "engines": {
51
51
  "node": ">=18"
package/skills/commands/memtrace-quality.md CHANGED
@@ -9,7 +9,7 @@ allowed-tools:
9
9
  - mcp__memtrace__get_evolution
10
10
  metadata:
11
11
  author: "Syncable <support@syncable.dev>"
12
- version: "1.0.0"
12
+ version: "1.1.0"
13
13
  category: development
14
14
  user-invocable: true
15
15
  ---
@@ -22,7 +22,7 @@ Code quality via graph analysis — dead code, complexity hotspots, repository s
22
22
 
23
23
  | Tool | Purpose | Required | Key optional |
24
24
  |------|---------|----------|--------------|
25
- | `find_dead_code` | Zero-caller symbols | `repo_id` | `include_tests`, `limit`, `kinds[]` |
25
+ | `find_dead_code` | Symbols unreachable from roots | `repo_id` | `include_tests`, `limit`, `min_confidence`, `include_historical` |
26
26
  | `find_most_complex_functions` | Ranked complexity hotspots | `repo_id` | **`top_n`** (def 20) — not `limit` |
27
27
  | `calculate_cyclomatic_complexity` | Score one symbol | `repo_id`, **`target`** | — |
28
28
  | `get_repository_stats` | Repository overview | `repo_id` | `branch` |
@@ -42,10 +42,20 @@ Full parameter spec for every Memtrace tool: `references/mcp-parameters.md` (bun
42
42
  ### 2. Dead code
43
43
 
44
44
  ```json
45
- { "repo_id": "memdb", "include_tests": false, "limit": 50 }
45
+ { "repo_id": "memdb", "include_tests": false, "limit": 50, "min_confidence": "high" }
46
46
  ```
47
47
 
48
- Exported symbols and entry points excluded by default. Results are candidates, not proof — callers via dynamic dispatch or reflection are invisible to the graph; verify before deleting.
48
+ `find_dead_code` seeds **roots** (`main`, exported symbols, HANDLES targets,
49
+ `trait_impl`, Magalz/framework allowlists), BFS-walks CALLS / REFERENCES /
50
+ MAKES_API_CALL forward, and reports Function/Method nodes **unreachable from
51
+ roots**. Each row includes `confidence` (`high` / `medium` / `low`) and
52
+ `confidence_score` (Skylos-style penalties for dynamic languages, ambiguous
53
+ hook names, methods). Use `min_confidence: "high"` for CI-safe cleanup;
54
+ omit or `"low"` to see all bands.
55
+
56
+ Results are still candidates, not proof — unmodeled dynamic dispatch can look
57
+ live or dead incorrectly. Verify before deleting. Prefer false-live over
58
+ false-dead.
49
59
 
50
60
  ### 3. Complexity hotspots
51
61
 
@@ -72,7 +82,7 @@ Ordered by call-graph out-degree.
72
82
 
73
83
  | Result | Carries |
74
84
  |--------|---------|
75
- | Dead-code entry (`find_dead_code`) | Symbol name + kind with zero graph callers — e.g. `format_legacy` (Function) |
85
+ | Dead-code entry (`find_dead_code`) | Name, kind, path, `reason` (`unreachable from roots`), `confidence`, `confidence_score` |
76
86
  | Complexity row (`find_most_complex_functions`) | Symbol, complexity score, risk level — e.g. `processOrder` — 27, Critical |
77
87
  | Single score (`calculate_cyclomatic_complexity`) | Cyclomatic complexity for `target` |
78
88
  | Stats (`get_repository_stats`) | Node counts by kind, edge counts, community and process counts for `repo_id`/`branch` |
@@ -81,6 +91,8 @@ Ordered by call-graph out-degree.
81
91
 
82
92
  | Mistake | Reality |
83
93
  |---------|---------|
94
+ | Treating every dead-code row as deletable | Filter with `min_confidence: "high"`; still verify framework/dynamic use |
95
+ | Assuming “no inbound edge” is the model | Dead means unreachable from roots, not merely unreferenced |
84
96
  | `find_most_complex_functions(limit: 10)` | Param is **`top_n`** |
85
97
  | `calculate_cyclomatic_complexity(symbol_id=...)` | Required param is **`target`** |
86
98
  | Only looking at the highest complexity | Medium-complexity functions that are growing (check `get_evolution`) are often more urgent |
package/skills/workflows/memtrace-refactoring-guide.md CHANGED
@@ -35,7 +35,7 @@ Run these three tools in parallel to build a candidate list:
35
35
  Call `find_most_complex_functions` with `top_n: 20`
36
36
 
37
37
  **b) Dead code:**
38
- Call `find_dead_code` to find unused symbols
38
+ Call `find_dead_code` (prefer `min_confidence: "high"`) for symbols unreachable from roots
39
39
 
40
40
  **c) Architectural bottlenecks:**
41
41
  Call `find_bridge_symbols` to find chokepoints with too much responsibility
@@ -122,7 +122,7 @@ For each item, include:
122
122
  |-----------|--------|
123
123
  | Complex + volatile + high blast radius | Highest priority — but plan carefully; incremental approach |
124
124
  | Complex + stable + low blast radius | Can wait; refactor when you're already touching nearby code |
125
- | Dead code with zero callers | Run Cortex provenance/recall first; zero callers is not proof that no decision/contract keeps it |
125
+ | Dead code (`unreachable from roots`, high confidence) | Run Cortex provenance/recall first; reachability is not proof that no decision/contract keeps it |
126
126
  | Bridge symbol with many dependents | Extract interface first, then refactor implementation behind it |
127
127
  | Symbol in cross-repo API | Coordinate with consumers; backward-compatible changes only |
128
128
  | Cortex returns a held ban/contract | Preserve it or ask before overriding it |