diff --git a/agents/hwfengcs__dm-code-agent/README.md b/agents/hwfengcs__dm-code-agent/README.md
new file mode 100644
index 0000000..d19387f
--- /dev/null
+++ b/agents/hwfengcs__dm-code-agent/README.md
@@ -0,0 +1,72 @@
+# DM-Code-Agent
+
+A local-first, auditable Python code-maintenance agent built on a **ReAct + Task Planner + Adaptive Replanning** loop. Every plan, tool call, and observation is written to a structured JSONL trace for offline replay and diff — no black-box magic.
+
+The entire core fits in ~1,500 lines of readable Python, making it a rare agent you can actually read, reproduce, and benchmark against.
+
+---
+
+## Run
+
+```bash
+npx @open-gitagent/gitagent run -r https://github.com/hwfengcs/DM-Code-Agent
+```
+
+---
+
+## What It Can Do
+
+- **Bug fixing** — Locates and fixes small-to-medium bugs, then runs the test suite to verify
+- **Regression testing** — Adds tests that cover more than just the visible failure case
+- **Code analysis** — Analyzes project structure, function signatures, dependencies, and metrics via AST tools
+- **Refactoring** — Performs small refactors or documentation-consistency fixes with file constraints
+- **Trace & benchmark reports** — Produces JSONL traces with dry-replay, tool-replay, and offline diff for agent quality auditing
+
+## Key Capabilities
+
+| Capability | Description |
+|---|---|
+| ReAct Loop | `thought / action / action_input` loop with per-step observation injection |
+| Task Planner | 3-8 step plan generated before execution; auto-replanning on failure |
+| JSONL Trace | Structured trace of every plan, tool call, LLM summary, and result — replayable offline |
+| Reflexion | Extracts failure lessons from prior trials and injects into the next (default-off) |
+| Critic | Peer-review gate before solution acceptance (default-off) |
+| Self-Consistency | N-way independent selection by majority-vote, critic score, or test-pass (default-off) |
+| Adaptive Replanning | Error-signal-to-strategy mapping with token-economics reporting (default-off) |
+| Context Memory | Mem0-style atomic memory compression — episodic, semantic, procedural |
+| MCP Integration | Attach Playwright, Context7, filesystem, SQLite, and other MCP servers |
+| Multi-LLM | DeepSeek, OpenAI (GPT-4o), Claude, Gemini, custom `base_url` |
+| Skill System | Domain-specific prompts and tools activated by task signals |
+| Maintenance Benchmarks | Hidden-test suite with changed-file constraints and trace analysis |
+
+## Example Usage
+
+```bash
+# Install
+git clone https://github.com/hwfengcs/DM-Code-Agent.git
+cd DM-Code-Agent
+python -m venv .venv && source .venv/bin/activate
+pip install -e ".[dev]"
+cp .env.example .env  # add your API key
+
+# Run a task
+dm-agent "Fix the retry boundary in retry.py and run the tests" \
+  --provider deepseek \
+  --trace traces/retry-fix.jsonl \
+  --show-steps
+
+# Inspect the trace
+dm-agent-trace analyze traces/retry-fix.jsonl
+dm-agent-trace diff traces/baseline.jsonl traces/retry-fix.jsonl
+```
+
+## Why It Stands Out
+
+- **Auditable by design** — traces record everything; debug without asking the model again
+- **Benchmarked** — ships with coding and maintenance hidden-test suites; SWE-bench Lite Tier-1 baseline published
+- **Algorithmic** — Reflexion, Critic, Self-Consistency, and Adaptive Replanning are modular, tested, default-off capabilities — not marketing
+- **Tiny core** — ~1,500 LOC vs. ~10k–50k for comparable tools; readable and extensible
+
+## License
+
+MIT — see [LICENSE](https://github.com/hwfengcs/DM-Code-Agent/blob/main/LICENSE)
diff --git a/agents/hwfengcs__dm-code-agent/metadata.json b/agents/hwfengcs__dm-code-agent/metadata.json
new file mode 100644
index 0000000..3a653c2
--- /dev/null
+++ b/agents/hwfengcs__dm-code-agent/metadata.json
@@ -0,0 +1,14 @@
+{
+  "name": "dm-code-agent",
+  "author": "hwfengcs",
+  "description": "Local-first, auditable Python code-maintenance agent: ReAct + Planner + Reflexion + Critic + Self-Consistency + MCP, with full JSONL trace replay.",
+  "repository": "https://github.com/hwfengcs/DM-Code-Agent",
+  "version": "2.0.0",
+  "category": "developer-tools",
+  "tags": ["code-agent", "react-agent", "auditable", "trace-replay", "mcp", "reflexion", "benchmarking", "local-first", "python", "swe-bench"],
+  "license": "MIT",
+  "model": "claude-sonnet-4-6",
+  "adapters": ["claude-code", "openai", "system-prompt"],
+  "icon": false,
+  "banner": false
+}