add roadmap

Author: vabruzzo

CLAUDE.md

@@ -1,6 +1,6 @@
 # AgentLens
 
-Harness for running multi-session Claude Code experiments and capturing trajectories in ATIF format. Built for agent interpretability research.
+Harness for running multi-session Claude Code experiments and capturing trajectories in ATIF format. Built for AI alignment and interpretability research.
 
 ## Project structure
 

docs/guide/roadmap.md

@@ -0,0 +1,31 @@
+# Roadmap
+
+AgentLens currently supports Claude Code via the Claude Agent SDK. Here's where we're headed.
+
+## Planned
+
+### Multi-agent support
+
+Extend beyond Claude Code to support other agent frameworks and LLM providers — Codex, Devin, custom agents, and any tool-using LLM system. The core trajectory format (ATIF) and analysis tooling are already agent-agnostic; the main work is adding new adapters and runner integrations.
+
+### Comparative analysis
+
+Side-by-side trajectory comparison across agents, models, and prompt variants. Run the same experiment config against different agents and compare behavior, tool usage patterns, and outcomes.
+
+### Richer intervention toolkit
+
+Programmatic intervention pipelines for systematic counterfactual testing. Build on the existing edit & resample workflow with scriptable intervention sequences, batch processing, and automated hypothesis testing.
+
+### Scoring & evaluation
+
+Built-in trajectory scoring and automated evaluation metrics. Define custom scorers that run against captured trajectories to measure agent behavior quantitatively.
+
+## Contributing
+
+We welcome PRs and contributions! Whether it's bug fixes, new features, documentation improvements, or support for additional agent frameworks — all contributions are appreciated.
+
+If you're interested in adding support for a new agent framework, the key integration points are:
+
+- **Runner** (`src/harness/runner.py`) — session execution loop
+- **Adapter** (`src/harness/atif_adapter.py`) — message-to-ATIF step mapping
+- **Proxy** (`src/harness/proxy.py`) — API request capture (optional)