Feat/nl edits ai planner prep

.claude/prompts/nl-unity-suite.md

@@ -0,0 +1,103 @@
+# CLAUDE TASK: Run NL/T editing tests for Unity MCP repo and emit JUnit
+
+You are running in CI at the repository root. Use only the tools that are allowed by the workflow:
+- View, GlobTool, GrepTool for reading.
+- Bash for local shell (git is allowed).
+- BatchTool for grouping.
+- MCP tools from server "unity" (exposed as mcp__unity__*).
+
+## Test target
+- Primary file: `ClaudeTests/longUnityScript-claudeTest.cs`
+- For each operation, prefer structured edit tools (`replace_method`, `insert_method`, `delete_method`, `anchor_insert`, `apply_text_edits`, `regex_replace`) via the MCP server.
+- Include `precondition_sha256` for any text path write.
+
+## Output requirements
+- Create a JUnit XML at `reports/claude-nl-tests.xml`.
+- Each test = one `<testcase>` with `classname="UnityMCP.NL"` or `UnityMCP.T`.
+- On failure, include a `<failure>` node with a concise message and the last evidence snippet (10–20 lines).
+- Also write a human summary at `reports/claude-nl-tests.md` with checkboxes and the windowed reads.
+
+## Safety & hygiene
+- Make edits in-place, then revert them at the end (`git stash -u`/`git reset --hard` or balanced counter-edits) so the workspace is clean for subsequent steps.
+- Never push commits from CI.
+- If a write fails midway, ensure the file is restored before proceeding.
+
+## NL-0. Sanity Reads (windowed)
+- Tail 120 lines of `ClaudeTests/longUnityScript-claudeTest.cs`.
+- Show 40 lines around method `Update`.
+- **Pass** if both windows render with expected anchors present.
+
+## NL-1. Method replace/insert/delete (natural-language)
+- Replace `HasTarget` with block-bodied version returning `currentTarget != null`.
+- Insert `PrintSeries()` after `GetCurrentTarget` logging `1,2,3`.
+- Verify by reading 20 lines around the anchor.
+- Delete `PrintSeries()` and verify removal.
+- **Pass** if diffs match and verification windows show expected content.
+
+## NL-2. Anchor comment insertion
+- Add a comment `Build marker OK` immediately above the `Update` method.
+- **Pass** if the comment appears directly above the `public void Update()` line.
+
+## NL-3. End-of-class insertion
+- Insert a 3-line comment `Tail test A/B/C` before the last method or immediately before the final class brace (preview, then apply).
+- **Pass** if windowed read shows the three lines at the intended location.
+
+## NL-4. Compile trigger
+- After any NL edit, ensure no stale compiler errors:
+  - Write a short marker edit, then **revert** after validating.
+  - The CI job will run Unity compile separately; record your local check (e.g., file parity and syntax sanity) as INFO, but do not attempt to invoke Unity here.
+
+## T-A. Anchor insert (text path)
+- Insert after `GetCurrentTarget`: `private int __TempHelper(int a, int b) => a + b;`
+- Verify via read; then delete with a `regex_replace` targeting only that helper block.
+- **Pass** if round-trip leaves the file exactly as before.
+
+## T-B. Replace method body with minimal range
+- Identify `HasTarget` body lines; single `replace_range` to change only inside braces; then revert.
+- **Pass** on exact-range change + revert.
+
+## T-C. Header/region preservation
+- For `ApplyBlend`, change only interior lines via `replace_range`; the method signature and surrounding `#region`/`#endregion` markers must remain untouched.
+- **Pass** if signature and region markers unchanged.
+
+## T-D. End-of-class insertion (anchor)
+- Find final class brace; `position: before` to append a temporary helper; then remove.
+- **Pass** if insert/remove verified.
+
+## T-E. Temporary method lifecycle
+- Insert helper (T-A), update helper implementation via `apply_text_edits`, then delete with `regex_replace`.
+- **Pass** if lifecycle completes and file returns to original checksum.
+
+## T-F. Multi-edit atomic batch
+- In one call, perform two `replace_range` tweaks and one comment insert at the class end; verify all-or-nothing behavior.
+- **Pass** if either all 3 apply or none.
+
+## T-G. Path normalization
+- Run the same edit once with `unity://path/ClaudeTests/longUnityScript-claudeTest.cs` and once with `ClaudeTests/longUnityScript-claudeTest.cs` (if supported).
+- **Pass** if both target the same file and no path duplication.
+
+## T-H. Validation levels
+- After edits, run `validate` with `level: "standard"`, then `"basic"` for temporarily unbalanced text ops; final state must be valid.
+- **Pass** if validation OK and final file compiles in CI step.
+
+## T-I. Failure surfaces (expected)
+- Too large payload: `apply_text_edits` with >15 KB aggregate → expect `{status:"too_large"}`.
+- Stale file: change externally, then resend with old `precondition_sha256` → expect `{status:"stale_file"}` with hashes.
+- Overlap: two overlapping ranges → expect rejection.
+- Unbalanced braces: remove a closing `}` → expect validation failure and **no write**.
+- Header guard: attempt insert before the first `using` → expect `{status:"header_guard"}`.
+- Anchor aliasing: `insert`/`content` alias → expect success (aliased to `text`).
+- Auto-upgrade: try a text edit overwriting a method header → prefer structured `replace_method` or return a clear error.
+- **Pass** when each negative case returns the expected failure without persisting changes.
+
+## T-J. Idempotency & no-op
+- Re-run the same `replace_range` with identical content → expect success with no change.
+- Re-run a delete of an already-removed helper via `regex_replace` → clean no-op.
+- **Pass** if both behave idempotently.
+
+### Implementation notes
+- Always capture pre- and post‑windows (±20–40 lines) as evidence in the JUnit `<failure>` or as `<system-out>`.
+- For any file write, include `precondition_sha256` and verify the post‑hash in your log.
+- At the end, restore the repository to its original state (`git status` must be clean).
+
+# Emit the JUnit file to reports/claude-nl-tests.xml and a summary markdown to reports/claude-nl-tests.md.

.github/workflows/claude-nl-suite.yml

@@ -0,0 +1,110 @@
+name: Claude NL suite + (optional) Unity compile
+
+on: { workflow_dispatch: {} }
+
+
+
+permissions:
+  contents: write        # allow Claude to write test artifacts
+  pull-requests: write   # allow annotations / comments
+  issues: write
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  nl-suite:
+    if: github.event_name == 'workflow_dispatch'
+    runs-on: ubuntu-latest
+
+    steps:
+      - uses: actions/checkout@v4
+        with: { fetch-depth: 0 }
+
+      # If your MCP server needs Python deps (adjust to your repo layout)
+      - name: Install Python + uv
+        uses: astral-sh/setup-uv@v4
+        with:
+          python-version: '3.11'
+
+      - name: Prepare Unity MCP server deps (adjust path or remove if N/A)
+        run: |
+          if [ -f UnityMcpServer/requirements.txt ]; then
+            uv pip install -r UnityMcpServer/requirements.txt
+          fi
+
+      - name: Run Claude NL/T test suite
+        id: claude
+        uses: anthropics/claude-code-base-action@beta
+        with:
+          # All the test instructions live here (see next file)
+          prompt_file: .claude/prompts/nl-unity-suite.md
+
+          # Keep tools tight: read, grep, glob, run shell, orchestrate batches,
+          # and call your MCP server tools. (Adjust the mcp__ prefix to match.)
+          allowed_tools: "Bash(git:*),View,GlobTool,GrepTool,BatchTool,mcp__unity__*"
+
+          # Inline MCP config (or put this JSON in .claude/mcp.json)
+          mcp_config: |
+            {
+              "mcpServers": {
+                "unity": {
+                  "command": "python",
+                  "args": ["UnityMcpServer/src/server.py"]
+                }
+              }
+            }
+
+          # Model + guardrails
+          model: "claude-3-7-sonnet-20250219"
+          max_turns: "10"
+          timeout_minutes: "20"
+          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
+
+      - name: Upload JUnit (Claude NL/T)
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: claude-nl-tests
+          path: reports/claude-nl-tests.xml
+
+      - name: Annotate PR with test results (Claude NL/T)
+        if: always()
+        uses: dorny/test-reporter@v1
+        with:
+          name: Claude NL/T
+          path: reports/claude-nl-tests.xml
+          reporter: java-junit
+
+      # --- Optional: Unity compile after Claude’s edits (satisfies NL-4) ---
+      # If your repo is a *Unity project*:
+      - name: Unity compile (Project)
+        if: ${{ always() && hashFiles('ProjectSettings/ProjectVersion.txt') != '' && secrets.UNITY_LICENSE != '' }}
+        uses: game-ci/unity-test-runner@v4
+        env:
+          UNITY_LICENSE: ${{ secrets.UNITY_LICENSE }}  # OR UNITY_* for Pro
+        with:
+          projectPath: .
+          githubToken: ${{ secrets.GITHUB_TOKEN }}
+          # Even with no tests, this compiles; add EditMode/PlayMode tests later.
+          testMode: EditMode
+
+      # If your repo is primarily a *Unity package*, prefer packageMode:
+      - name: Unity compile (Package)
+        if: ${{ always() && hashFiles('Packages/manifest.json') != '' && hashFiles('ProjectSettings/ProjectVersion.txt') == '' && secrets.UNITY_LICENSE != '' }}
+        uses: game-ci/unity-test-runner@v4
+        env:
+          UNITY_LICENSE: ${{ secrets.UNITY_LICENSE }}
+        with:
+          packageMode: true
+          unityVersion: 2022.3.45f1   # <-- set explicitly for packages
+          projectPath: .              # or a small sample project path
+          githubToken: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Clean working tree (discard temp edits)
+        if: always()
+        run: |
+          git restore -SW :/
+          git clean -fd
+