harmony-skills-standalone 1.0.2 → 1.0.3

package/autotest-agent/SKILL.md

@@ -1,39 +1,47 @@
 ---
 name: autotest-agent
-description: Run the repository AutoTestAgent for a declared test-execution workflow node.
+description: Execute declared test cases through the checked-in AutoTestAgent batch wrapper.
 ---
 
 # AutoTest Agent Bridge
 
-Use the checked-in `SubAgents/AutoTestAgent-master` implementation to execute the
-test intent or test-case input named in the prompt.
+Use `scripts/run_autotest_batch.py` as the only execution entry. Do not inspect
+platform backend source code, manually parse `.env`, or assemble an AutoTest
+configuration file.
 
-- Verify the device and HAP prerequisites before execution.
-- Never fabricate device interaction, screenshots, pass counts, or reports.
-- Preserve the AutoTestAgent raw report under the declared node output directory.
-- Write the declared `result` output as valid JSON containing the real command,
-  device, case totals, pass/fail counts, report paths, and blocking reason.
-- Emit structured JSON progress lines during setup, execution, attribution, and
-  report generation.
-- Fail when required device or application prerequisites are unavailable.
+From the node prompt, use the declared `Platform root`, `Python executable`,
+`Task directory`, `Progress file`, and `Result` paths to run:
 
-## Execution method
+```text
+"<Python executable>" "<Platform root>/scripts/run_autotest_batch.py"
+  --task-dir "<Task directory>/nodes/case-exec/outputs/autotest"
+  --prompt-file "<Task directory>/nodes/case-exec/prompt.md"
+  --work-dir "<Platform root>/SubAgents/AutoTestAgent-master"
+  --progress-file "<Progress file>"
+  --result "<Result>"
+```
 
-When the node declares a `progress` output (type `file`), use the wrapper
-script instead of calling `batch.py` directly so that real-time stdout is
-captured for the progress file tailer:
+The wrapper resolves HDC from `Platform/backend/config/runtime.yaml`, verifies
+that the requested device is online, converts the prompt's single or batch test
+case input to `testcases.jsonl`, injects only environment-referenced API keys
+into the AutoTest child process, streams AutoTest output to the progress file,
+and writes the required result JSON.
 
-```bash
-python "${workflow.inputs.platform_root}/scripts/run_autotest_batch.py" \
-  --task-dir "<reports-dir>" \
-  --testcases "<testcases.jsonl>" \
-  --work-dir "${workflow.inputs.platform_root}/SubAgents/AutoTestAgent-master" \
-  --category "intent-test" \
-  --progress-file "${outputs.progress}" \
-  --result "${outputs.result}"
-```
+Rules:
 
-If the node does not declare a `progress` output, call `batch.py` directly as
-before. The wrapper script sets up the same environment (PYTHONPATH, HDC_PATH,
-venv Python) as the platform's TestExecutionService and tees batch.py stdout
-to the progress file for real-time tailing.
+- Invoke the wrapper exactly once for this node, then stop. Never invoke it a second time for the same node, including after a failed test, non-zero exit, timeout, blocked setup, or missing result file.
+- Run the wrapper immediately; do not search for another execution path.
+- If a structured test-case file is already available, use `--test-case-file`
+  instead of `--prompt-file`.
+- After the wrapper returns, finish the node immediately. The result JSON is
+  the final node output; report its path without modifying it.
+- `test_status: failed` is a valid completed test result, not a command failure
+  and not a request to diagnose or repair the environment.
+- Do not fabricate device interactions, screenshots, pass counts, or reports.
+- Do not place an API key in a command, prompt, log, event, or result.
+- Do not inspect reports, progress logs, source code, dependency state, `.env`,
+  or AutoTest configuration to diagnose a completed wrapper run.
+- Do not install dependencies, edit configuration, retry, or invoke another
+  execution path after the wrapper has run.
+- On any setup or execution failure, let the wrapper write a truthful blocked
+  or failed `result.json`; never create a success result yourself.

package/codesearch/SKILL.md

@@ -0,0 +1,197 @@
+---
+name: codesearch
+description: >-
+  Build a code knowledge graph index for Git repositories and expose it as an MCP
+  Server for AI-driven code understanding. Use when the task involves indexing a
+  codebase, searching code by keyword or semantics, querying call/inheritance
+  relations, analyzing impact of a symbol change, executing Cypher graph queries,
+  or running a local code-search MCP service. Supports JavaScript, TypeScript,
+  Python, Java, C, C++, C#, Go, Rust, PHP, and Kotlin. Triggers on phrases like
+  code search, code index, code knowledge graph, call relation, impact analysis,
+  semantic search, BM25, Cypher query, or MCP server for code.
+source: local
+upstream_version: 0.1.0
+---
+
+# CodeSearch
+
+`codesearch` builds a code knowledge graph from Git repositories and exposes it
+as an MCP Server. It provides keyword search (BM25), semantic vector retrieval,
+Cypher graph queries, call/inheritance relation traversal, and impact analysis.
+
+## Prerequisites
+
+- **Node.js >= 22.0.0**
+- **Python >= 3.13** — only required for semantic retrieval features
+  (`embedding-service`, `repo-embedding`, and the `semantic` MCP tool). Basic
+  indexing and MCP query work without Python.
+
+## Installation
+
+Install the bundled npm package into the current project:
+
+```bash
+npm install <skill-dir>/codesearch-0.1.0.tgz
+```
+
+For npm 11+ (which blocks install scripts by default), allow native module
+compilation:
+
+```bash
+echo "allow-scripts=true" >> .npmrc
+npm install <skill-dir>/codesearch-0.1.0.tgz
+```
+
+Verify:
+
+```bash
+npx codesearch --version
+npx codesearch --help
+```
+
+For Python-dependent features, install Python dependencies:
+
+```bash
+pip install -r references/requirements.txt
+```
+
+## Quick start
+
+Minimal flow — index a repo and start the MCP Server:
+
+```bash
+npx codesearch analyze /path/to/repo --output-path /path/to/index
+npx codesearch mcp --repo /path/to/index
+```
+
+## CLI commands
+
+| Command | Purpose |
+|---------|---------|
+| `analyze [path] --output-path <dir>` | Full analysis of a single Git repo |
+| `analyze-dist <input> <output>` | Distributed analysis for multi-module repos |
+| `incremental [path] [prevPath]` | Rebuild only changed modules (requires prior `analyze-dist`) |
+| `realtime <path> <output>` | File-level real-time analysis for local dev |
+| `mcp [--mode stdio\|http\|fusion] [--repo <dir>]` | Start MCP Server on an index |
+| `repo-embedding <index> <url>` | Generate embeddings + Milvus Lite index |
+| `embedding-service --mode embedding\|query` | Start Python embedding or query service |
+| `merge <target> <src1> <src2> [...]` | Merge multiple index directories |
+
+Key rules:
+
+- `--repo` accepts **index directories** (containing `kuzu/` and `meta.json`),
+  not source code directories.
+- For `analyze-dist` output, pass the `mergeRepo` subdirectory as `--repo`.
+- `analyze`, `analyze-dist`, `incremental`, `realtime`, and `mcp` do **not**
+  require Python. Only `embedding-service` and `repo-embedding` do.
+
+## MCP Server modes
+
+| Mode | Use case |
+|------|----------|
+| `stdio` (default) | Local dev, single client (Claude Desktop, Cursor, Trae) |
+| `http` | Long-running service, multi-client, remote deployment |
+| `fusion` | Local + remote hybrid; merges local and cloud results |
+
+### stdio example
+
+```bash
+npx codesearch mcp --repo /path/to/index
+```
+
+### http example
+
+```bash
+npx codesearch mcp --mode http --host 127.0.0.1 --port 4848 --repo /path/to/index
+```
+
+### fusion example
+
+```bash
+npx codesearch mcp --mode fusion --remote-mcp-url http://127.0.0.1:4848/mcp --repo /path/to/local-index
+```
+
+## MCP tools
+
+Standard mode (`stdio` / `http`) exposes:
+
+| Tool | Description |
+|------|-------------|
+| `list_repos` | List all indexed repositories |
+| `query` | Hybrid keyword + semantic search; results organized by execution flow |
+| `cypher` | Execute read-only Cypher queries on the code knowledge graph |
+| `context` | Get 360-degree context of a symbol (incoming/outgoing relations) |
+| `impact` | Analyze upstream/downstream impact of modifying a symbol |
+| `search_class_by_graph` | Search class definitions and inheritance |
+| `search_func_by_graph` | Search function definitions |
+| `search_call_relation` | Traverse caller/callee chains for a function |
+| `search_extend_relation` | Traverse parent/child inheritance for a class |
+| `bm25` | BM25 keyword retrieval |
+| `semantic` | Semantic vector retrieval (requires embedding service) |
+| `getContent` | Fetch source code content by node ID |
+
+`fusion` mode exposes only `query` and `cypher`, merging local and remote results.
+
+## Recommended workflow
+
+1. `list_repos` — confirm available indexes
+2. `query` — find relevant symbols and execution flows
+3. `context` — deep-dive into a specific symbol
+4. `impact` — assess risk before modifying a symbol
+
+## MCP client configuration
+
+### stdio
+
+```json
+{
+  "mcpServers": {
+    "codesearch": {
+      "command": "npx",
+      "args": ["codesearch", "mcp", "--repo", "/path/to/index"]
+    }
+  }
+}
+```
+
+### http
+
+```json
+{
+  "mcpServers": {
+    "codesearch-http": {
+      "url": "http://127.0.0.1:4848/mcp"
+    }
+  }
+}
+```
+
+## Enabling semantic retrieval
+
+1. Build base index: `npx codesearch analyze /path/to/repo --output-path /path/to/index`
+2. Start embedding service: `npx codesearch embedding-service --mode embedding --model-path /path/to/Qwen3-Embedding-0.6B --port 5001`
+3. Generate embeddings: `npx codesearch repo-embedding /path/to/index http://127.0.0.1:5001/embed`
+4. Start query service: `npx codesearch embedding-service --mode query --db-path /path/to/index/embedding/milvus_lite.db --port 5000`
+5. Update `/path/to/index/meta.json`:
+
+```json
+{
+  "embedding_url": "http://127.0.0.1:5001/embed",
+  "vectordb_url": "http://127.0.0.1:5000/search"
+}
+```
+
+6. Start MCP: `npx codesearch mcp --repo /path/to/index`
+
+If `embedding_url` or `vectordb_url` is not configured, the `semantic` tool
+returns empty results without affecting other tools.
+
+## Supported languages
+
+JavaScript, TypeScript, Python, Java, C, C++, C#, Go, Rust, PHP, Kotlin.
+
+## References
+
+- [CLI guide](references/cli-guide.md) — full CLI command reference with examples
+- [MCP tools guide](references/mcp-tools-guide.md) — detailed MCP tool descriptions and return examples
+- [requirements.txt](references/requirements.txt) — Python dependencies for semantic features