Halo support

Author: Shashikant86

CHANGELOG.md

@@ -5,6 +5,19 @@ All notable changes to this project are documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.1.7] - 2026-04-30
+
+### Added
+- HALO-style `trace_analysis` RLM environment for diagnosing agent harness failures from one-span-per-line JSONL traces.
+- Trace sidecar indexing with dataset rollups for trace counts, span counts, error traces, services, models, agents, token totals, and sample trace ids.
+- Bounded trace inspection actions: `get_dataset_overview`, `query_traces`, `count_traces`, `view_trace`, `search_trace`, and `view_spans`.
+- Large-trace safeguards: per-attribute truncation, oversized trace summaries, and higher-cap selected-span reads.
+- Tests for trace indexing, querying, searching, selected-span viewing, and trace environment actions.
+- Trace analysis documentation under the Core Engine docs.
+
+### Changed
+- `/rlm` command help now advertises `env=trace_analysis` for run, chat, and doctor workflows.
+
 ## [0.1.6] - 2026-02-20
 
 ### Added
@@ -56,3 +69,4 @@ Initial public release of **RLM Code**.
 
 [0.1.5]: https://github.com/SuperagenticAI/rlm-code/releases/tag/v0.1.5
 [0.1.6]: https://github.com/SuperagenticAI/rlm-code/releases/tag/v0.1.6
+[0.1.7]: https://github.com/SuperagenticAI/rlm-code/releases/tag/v0.1.7

README.md

@@ -25,21 +25,20 @@ RLM Code implements the [Recursive Language Models](https://arxiv.org/abs/2502.0
 
 RLM Code wraps this algorithm in an interactive terminal UI with built-in benchmarks, trajectory replay, and observability.
 
-## Release v0.1.6
+## Release v0.1.7
 
-This release adds the new CodeMode path as an opt-in harness strategy.
+This release adds HALO-style trace analysis as a new RLM environment.
 
-- New harness strategy: `strategy=codemode` (default remains `strategy=tool_call`)
-- MCP bridge flow for CodeMode: `search_tools` -> typed tool surface -> `call_tool_chain`
-- Guardrails before execution: blocked API classes plus timeout/size/tool-call caps
-- Benchmark telemetry for side-by-side comparison: `tool_call` vs `codemode`
-- Dedicated docs section for CodeMode: quickstart, architecture, guardrails, evaluation
-- Multi-backend setup docs for UTCP (local) and Cloudflare (remote MCP)
+- New `trace_analysis` environment for diagnosing agent harness failures from OTel-shaped JSONL traces
+- Sidecar trace indexing with dataset overview, query, count, search, full-trace view, and selected-span view actions
+- Bounded payload handling for large traces, including oversized summaries and higher-cap surgical span reads
+- `/rlm` help/docs updated for `env=trace_analysis`
+- Dedicated trace analysis docs under the Core Engine section
 
 Example:
 
 ```text
-/harness run "implement feature and add tests" steps=3 mcp=on strategy=codemode mcp_server=utcp-codemode
+/rlm run "Find systemic harness failures trace=./traces.jsonl" env=trace_analysis steps=6
 ```
 
 ## Documentation

docs/core/trace-analysis.md

@@ -0,0 +1,63 @@
+# Trace Analysis
+
+`rlm-code` includes a HALO-style trace analysis environment for diagnosing
+agent harness failures from one-span-per-line JSONL traces.
+
+The environment is named `trace_analysis`. It indexes a trace file into a
+sidecar cache, exposes bounded trace-inspection actions to the RLM planner, and
+keeps large payloads under control by returning summaries or selected spans
+instead of blindly loading full traces into context.
+
+## Usage
+
+```text
+/rlm run "Find systemic harness failures trace=./traces.jsonl" env=trace_analysis steps=6
+```
+
+The task can include either `trace=<path>` or `trace_path=<path>`. The planner
+can also explicitly load a file with the `set_trace_path` action.
+
+## Actions
+
+The environment supports these planner actions:
+
+| Action | Purpose |
+|---|---|
+| `set_trace_path` | Load and index a trace JSONL file |
+| `get_dataset_overview` | Return dataset-level trace, span, service, model, agent, token, and error counts |
+| `query_traces` | List matching trace summaries with pagination |
+| `count_traces` | Count matching traces without materializing summaries |
+| `view_trace` | Read all spans for a small trace, or return an oversized summary |
+| `search_trace` | Search one trace for a literal substring |
+| `view_spans` | Read selected spans at a higher per-attribute cap |
+| `final` | Return the final evidence report |
+
+Supported filters are `has_errors`, `model_names`, `service_names`,
+`agent_names`, and `project_id`.
+
+## Trace Shape
+
+The first implementation expects one JSON object per line. Each line should
+represent one span with fields such as:
+
+```json
+{
+  "trace_id": "trace-1",
+  "span_id": "span-1",
+  "parent_span_id": null,
+  "name": "agent.Root",
+  "kind": "SPAN_KIND_INTERNAL",
+  "start_time": "2026-01-01T00:00:00Z",
+  "end_time": "2026-01-01T00:00:01Z",
+  "status": {"code": "STATUS_CODE_ERROR"},
+  "resource": {"attributes": {"service.name": "my-agent"}},
+  "attributes": {
+    "inference.project_id": "my-project",
+    "inference.agent_name": "Root",
+    "inference.llm.model_name": "gpt-test"
+  }
+}
+```
+
+This is intentionally compatible with the HALO/OpenTelemetry-style file export
+pattern where trace data is stored as JSONL and queried through a sidecar index.

docs/index.md

@@ -6,7 +6,7 @@
 
 <p class="rlm-tagline">Research Playground & Evaluation OS for Recursive Language Model Agentic Systems</p>
 
-<span class="rlm-badge rlm-badge--purple">v0.1.6</span>
+<span class="rlm-badge rlm-badge--purple">v0.1.7</span>
 <span class="rlm-badge rlm-badge--green">Python 3.11+</span>
 <span class="rlm-badge rlm-badge--blue">Apache 2.0</span>
 
@@ -46,6 +46,13 @@ Run **Pure RLM** (paper-compliant with context-as-variable), **CodeAct** (contex
 
 <div class="rlm-feature-card" markdown>
 
+### 🔎 Trace Analysis
+Run HALO-style trace diagnosis with `env=trace_analysis` over OTel-shaped JSONL traces to find repeated harness failure modes.
+
+</div>
+
+<div class="rlm-feature-card" markdown>
+
 ### 🧪 Harness CodeMode
 Opt into `strategy=codemode` for MCP tool discovery, guarded single-program generation, and chain execution via `call_tool_chain`.
 

docs/reference/index.md

@@ -15,6 +15,7 @@ RLM Code is organized into the following top-level packages. Each module is docu
 | `rlm_code.rlm.events` | Event bus with 27+ event types, collector, and subscriber system | [Event System](../core/events.md) |
 | `rlm_code.rlm.termination` | FINAL/FINAL_VAR detection, code block extraction, answer formatting | [Termination Patterns](../core/termination.md) |
 | `rlm_code.rlm.memory_compaction` | LLM and deterministic memory compaction strategies | [Memory Compaction](../core/memory-compaction.md) |
+| `rlm_code.traces` | HALO-style trace indexing and bounded trace query helpers | [Trace Analysis](../core/trace-analysis.md) |
 | `rlm_code.rlm.repl_types` | REPLVariable, REPLEntry, REPLHistory, REPLResult data types | [REPL Types](../core/repl-types.md) |
 | `rlm_code.rlm.trajectory` | Trajectory event logging, viewing, and comparison | [Trajectory Logging](../core/trajectory.md) |
 | `rlm_code.rlm.comparison` | Paradigm comparison engine (Pure RLM vs CodeAct vs Traditional) | [Paradigm Comparison](../core/comparison.md) |

mkdocs.yml

@@ -124,6 +124,7 @@ nav:
     - "\U0001F4E1 Event System": core/events.md
     - "\U0001F6D1 Termination": core/termination.md
     - "\U0001F9F9 Memory Compaction": core/memory-compaction.md
+    - "Trace Analysis": core/trace-analysis.md
     - "\U0001F4DF REPL Types": core/repl-types.md
     - "\U0001F4C8 Trajectory": core/trajectory.md
     - "\U0001F504 Paradigm Comparison": core/comparison.md

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "rlm-code"
-version = "0.1.6"
+version = "0.1.7"
 description = "RLM Code: Research Playground & Evaluation OS for Recursive Language Model Agentic Systems"
 readme = "README.md"
 license = "Apache-2.0"

rlm_code/__init__.py

@@ -5,5 +5,5 @@
 through natural language interactions.
 """
 
-__version__ = "0.1.6"
+__version__ = "0.1.7"
 __author__ = "Super Agentic AI"

rlm_code/commands/slash_commands.py

@@ -1684,7 +1684,7 @@ def cmd_rlm(self, args: list):
         Manage RLM runs.
 
         Usage:
-            /rlm run <task> [steps=N] [timeout=N] [branch=N] [depth=N] [children=N] [parallel=N] [budget=N] [framework=<see /rlm frameworks>] [env=generic|dspy|pure_rlm] [sub=provider/model]
+            /rlm run <task> [steps=N] [timeout=N] [branch=N] [depth=N] [children=N] [parallel=N] [budget=N] [framework=<see /rlm frameworks>] [env=generic|dspy|pure_rlm|trace_analysis] [sub=provider/model]
             /rlm bench [list|preset=name] [mode=native|harness|direct-llm] [strategy=tool_call|codemode] [mcp=on|off] [mcp_server=name] [pack=path[,path2]] [limit=N] [steps=N] [timeout=N] [branch=N] [framework=<see /rlm frameworks>] [env=generic|dspy|pure_rlm] [sub=provider/model]
             /rlm bench compare [candidate=<id|path|latest>] [baseline=<id|path|previous>] [min_reward_delta=N] [min_completion_delta=N] [max_steps_increase=N]
             /rlm bench validate [candidate=<id|path|latest>] [baseline=<id|path|previous>] [min_reward_delta=N] [min_completion_delta=N] [max_steps_increase=N] [--json]
@@ -1696,8 +1696,8 @@ def cmd_rlm(self, args: list):
             /rlm status [run_id]
             /rlm abort [run_id|all]
             /rlm replay [run_id|latest]
-            /rlm doctor [env=generic|dspy|pure_rlm] [--json]
-            /rlm chat <message> [session=name] [env=generic|dspy|pure_rlm] [branch=N] [depth=N] [children=N] [parallel=N] [budget=N] [framework=<see /rlm frameworks>] [sub=provider/model]
+            /rlm doctor [env=generic|dspy|pure_rlm|trace_analysis] [--json]
+            /rlm chat <message> [session=name] [env=generic|dspy|pure_rlm|trace_analysis] [branch=N] [depth=N] [children=N] [parallel=N] [budget=N] [framework=<see /rlm frameworks>] [sub=provider/model]
             /rlm chat status [session=name]
             /rlm chat reset [session=name]
             /rlm observability
@@ -1708,14 +1708,14 @@ def cmd_rlm(self, args: list):
             console.print("[bold cyan]🧠 RLM Commands[/bold cyan]")
             console.print(
                 "  [yellow]/rlm run <task> [steps=N] [timeout=N] [branch=N] [depth=N] [children=N] "
-                f"[parallel=N] [budget=N] [framework={framework_opts}] [env=generic|dspy|pure_rlm] "
+                f"[parallel=N] [budget=N] [framework={framework_opts}] [env=generic|dspy|pure_rlm|trace_analysis] "
                 "[sub=provider/model][/yellow]"
             )
             console.print(
                 "  [yellow]/rlm bench [list|preset=name] [mode=native|harness|direct-llm] "
                 "[strategy=tool_call|codemode] [mcp=on|off] [mcp_server=name] "
                 "[pack=path[,path2]] [limit=N] [steps=N] "
-                f"[timeout=N] [branch=N] [framework={framework_opts}] [env=generic|dspy|pure_rlm] [sub=provider/model][/yellow]"
+                f"[timeout=N] [branch=N] [framework={framework_opts}] [env=generic|dspy|pure_rlm|trace_analysis] [sub=provider/model][/yellow]"
             )
             console.print(
                 "  [yellow]/rlm bench compare [candidate=<id|path|latest>] [baseline=<id|path|previous>] "
@@ -1741,9 +1741,9 @@ def cmd_rlm(self, args: list):
             console.print("  [yellow]/rlm status [run_id][/yellow]")
             console.print("  [yellow]/rlm abort [run_id|all][/yellow]")
             console.print("  [yellow]/rlm replay [run_id|latest][/yellow]")
-            console.print("  [yellow]/rlm doctor [env=generic|dspy|pure_rlm] [--json][/yellow]")
+            console.print("  [yellow]/rlm doctor [env=generic|dspy|pure_rlm|trace_analysis] [--json][/yellow]")
             console.print(
-                "  [yellow]/rlm chat <message> [session=name] [env=generic|dspy|pure_rlm] [branch=N] [depth=N] "
+                "  [yellow]/rlm chat <message> [session=name] [env=generic|dspy|pure_rlm|trace_analysis] [branch=N] [depth=N] "
                 f"[children=N] [parallel=N] [budget=N] [framework={framework_opts}] "
                 "[sub=provider/model][/yellow]"
             )
@@ -2135,7 +2135,7 @@ def cmd_rlm(self, args: list):
             task = " ".join(task_tokens).strip()
             if not task:
                 show_error_message(
-                    "Usage: /rlm run <task> [steps=N] [timeout=N] [env=generic|dspy|pure_rlm] "
+                    "Usage: /rlm run <task> [steps=N] [timeout=N] [env=generic|dspy|pure_rlm|trace_analysis] "
                     "[depth=N] [children=N] [parallel=N] [budget=N] "
                     f"[framework={framework_opts}] "
                     "[branch=N] [sub=provider/model]"

rlm_code/mcp/__init__.py

@@ -17,7 +17,7 @@
 )
 from .session_wrapper import MCPSessionWrapper
 
-__version__ = "0.1.6"
+__version__ = "0.1.7"
 
 __all__ = [
     "MCPClientManager",