🤖 feat: improve agent profiling workflows for v0.4.1

.github/workflows/hardware.yml

@@ -17,12 +17,17 @@ jobs:
       - uses: taiki-e/install-action@just
       - run: just test-bpf-live
       - run: just test-cupti-live-cuda12
+      - run: just test-nvtx-live-cuda12
       - run: just test-cupti-live-cuda12-min
       - run: just test-injection-live-cuda12
       - run: just test-multisource-live-cuda12
       - run: just test-cupti-live
+      - run: just test-nvtx-live
       - run: just test-injection-live
       - run: just test-multisource-live
+      - run: just test-pytorch-live
+      - run: just test-pytorch-cuda-live
       - run: just benchmark-gpu
       - run: just benchmark-aggregate
       - run: just benchmark-multiprocess
+      - run: just benchmark-pytorch

.github/workflows/release.yml

@@ -86,3 +86,7 @@ jobs:
         with:
           files: dist/*.tar.gz*
           generate_release_notes: true
+      - name: Verify public release archive
+        env:
+          XPROBE_RELEASE_REPOSITORY: ${{ github.repository }}
+        run: scripts/verify-public-release.sh "${GITHUB_REF_NAME#v}"

AGENTS.md

@@ -47,13 +47,12 @@
 ## Agent workflow
 
 - Follow `skills/xprobe-measure-latency/SKILL.md` for measurement tasks.
-- Classify a workload as CPU-only or GPU/mixed before choosing collectors.
-  `discover` is only for CUDA context holders; CPU-only work proceeds with its
-  selected PID and host selectors. For GPU/mixed work, progress from readiness
-  and baseline through a broad, representative bounded inventory,
-  evidence-based selector narrowing, validation, and one detailed bounded
-  measurement per stated hypothesis. Scope breadth and capture duration are
-  independent. Do not guess selectors.
+- Choose the shortest Skill route supported by the evidence. Existing artifacts
+  do not require live readiness checks; known selectors may proceed directly to
+  validation and bounded measurement. Use broad-to-narrow inventory only when
+  selectors are unknown. Classify live work as CPU-only or GPU/mixed before
+  choosing collectors; `discover` is only for CUDA context holders. Scope
+  breadth and capture duration are independent. Do not guess selectors.
 - Read evidence pairs, artifact analysis, stream identity, collection quality,
   and profiler overhead before interpreting latency. Summed concurrent GPU
   duration is not wall time.
@@ -86,4 +85,6 @@
 - Run `just benchmark-aggregate` for aggregate inventory hot-path or capacity
   changes.
 - Run `just benchmark-multiprocess` for concurrent worker orchestration changes.
+- Run `just benchmark-pytorch` for PyTorch workflow or framework-overhead
+  changes.
 - Use emoji conventional commits, for example `🐛 fix: restore target registers`.

CMakeLists.txt

@@ -1,6 +1,6 @@
 cmake_minimum_required(VERSION 3.20)
 
-project(xprobe VERSION 0.4.0 LANGUAGES C)
+project(xprobe VERSION 0.4.1 LANGUAGES C)
 
 include(CTest)
 

Cargo.toml

@@ -10,7 +10,7 @@ members = [
 resolver = "2"
 
 [workspace.package]
-version = "0.4.0"
+version = "0.4.1"
 edition = "2024"
 license = "Apache-2.0"
 repository = "https://github.com/itdevwu/xprobe"

README.md

@@ -22,7 +22,7 @@ only installation action required from the user:
 
 ```bash
 npx skills@1 add \
-  https://github.com/itdevwu/xprobe/tree/v0.4.0/skills/xprobe-measure-latency \
+  https://github.com/itdevwu/xprobe/tree/v0.4.1/skills/xprobe-measure-latency \
   --global
 ```