diff --git a/run.json b/run.json
index 6838c06ee..bc196dce7 100644
--- a/run.json
+++ b/run.json
@@ -515,14 +515,106 @@
     }
   },
   "web_url": "http://127.0.0.1:32276/runs/01KSAPQQSVK4FY3CWHYVZYD7T6",
-  "start": null,
-  "status": {
-    "kind": "starting"
+  "start": {
+    "start_time": "2026-05-23T15:20:09.658986Z",
+    "run_branch": "fabro/run/01KSAPQQSVK4FY3CWHYVZYD7T6",
+    "base_sha": "a64a58d567e3de775aff86503703934c64fe2d38"
   },
-  "status_updated_at": "2026-05-23T15:19:53.456033Z",
-  "last_event_at": "2026-05-23T15:20:08.819373Z",
+  "status": {
+    "kind": "running"
+  },
+  "status_updated_at": "2026-05-23T15:20:09.659052Z",
+  "last_event_at": "2026-05-23T15:20:11.504629Z",
   "pending_control": null,
-  "checkpoints": [],
+  "checkpoints": [
+    {
+      "seq": 19,
+      "checkpoint": {
+        "timestamp": "2026-05-23T15:20:11.504317Z",
+        "current_node": "start",
+        "completed_nodes": [
+          "start"
+        ],
+        "node_retries": {},
+        "context_values": {
+          "failure_signature": "",
+          "graph.rankdir": "LR",
+          "current_node": "start",
+          "graph.model_stylesheet": "\n            * { model: claude-opus-4-7; }\n        ",
+          "internal.node_visit_count": 1,
+          "graph.goal": "# Agent Compaction API Usage Baseline Plan\n\nDate: 2026-05-23\n\n## Summary\n\nChange Fabro's agent compaction trigger from a whole-history `chars / 4`\nestimate to a Claude Code-style hot-path estimate: use the latest real\nassistant response's stored `usage.total_tokens()` as the baseline, then add\nlocal estimates for turns appended after that response. This avoids token-count\nprovider API calls while making compaction sensitive to actual\nprovider-reported context usage, including cache and reasoning tokens.\n\nNo provider token-count API calls should be added in this change.\n\n## Key Changes\n\n- Replace the current compaction estimate in `fabro-agent` with a new\n  active-context estimator.\n- Find the newest assistant turn whose `usage.total_tokens() > 0`.\n- Use that `usage.total_tokens()` as the baseline.\n- Add local estimates only for turns after that assistant turn.\n- If no usable assistant usage exists, fall back to the existing local\n  whole-history estimate.\n- Reuse a shared per-turn local estimate helper so fallback and post-baseline\n  delta counting stay consistent.\n- Keep the estimator local and in-process. Do not call\n  `llm_client.count_input_tokens()` from `compact_if_needed()`.\n\n## Implementation Details\n\n- Update `lib/crates/fabro-agent/src/compaction.rs`:\n  - Add an estimator that returns both token count and method, for example\n    `ApiUsagePlusLocalDelta` or `LocalEstimate`.\n  - Make `check_context_usage()` use the new estimator and include the method\n    in warning `details`.\n  - Make `compact_context()` report the same improved estimate in\n    `CompactionStarted`.\n  - Move `CompactionStarted` emission after the\n    `original_turn_count <= preserve_count` no-op check, so a no-op compact\n    cannot emit started without completed.\n- Update `lib/crates/fabro-agent/src/history.rs`:\n  - In `History::compact()`, invalidate preserved assistant usage by replacing\n    preserved assistant `usage` with `TokenCounts::default()`.\n  - Keep provider parts, response IDs, text, and tool calls unchanged.\n  - Rationale: preserved assistant usage reflects the pre-compaction context and\n    must not become the next baseline. Billing remains available from emitted\n    run events, so mutable runtime history should prefer compaction correctness.\n- Leave public run event names and schemas unchanged:\n  - `agent.compaction.started`\n  - `agent.compaction.completed`\n  - Existing warning event remains a warning with richer `details`.\n\n## Test Plan\n\n- Add unit coverage in `lib/crates/fabro-agent/src/compaction.rs`:\n  - No assistant usage: estimator matches current local whole-history behavior.\n  - Latest assistant usage present: estimator uses `usage.total_tokens()` plus\n    only later tool/user/steering turns.\n  - Usage fields include cache and reasoning through `TokenCounts::total_tokens()`.\n  - Earlier assistant usage is ignored when a later assistant usage exists.\n- Add unit coverage in `lib/crates/fabro-agent/src/history.rs`:\n  - `History::compact()` preserves assistant content, tool calls, and provider\n    parts, but resets preserved assistant usage to default.\n  - Existing OpenAI opaque stripping and Anthropic thinking preservation tests\n    still pass.\n- Add session coverage in `lib/crates/fabro-agent/src/session.rs`:\n  - A short assistant response with high `usage.total_tokens()` triggers\n    compaction even when text length is small.\n  - A compact no-op due to `turns.len() <= preserve_count` does not emit\n    `CompactionStarted`.\n  - Compaction disabled still prevents compaction even if the API usage\n    baseline exceeds threshold.\n- Run targeted verification:\n  - `cargo nextest run -p fabro-agent compaction`\n  - `cargo nextest run -p fabro-agent history`\n  - If those pass, run `cargo nextest run -p fabro-agent`.\n\n## Assumptions\n\n- Runtime/session `Message::Assistant.usage` is safe to invalidate after\n  compaction because authoritative billing comes from emitted workflow/run\n  events, not preserved mutable agent history.\n- A zero-token `TokenCounts::default()` should be treated as no usable API\n  baseline.\n- Provider token-count APIs remain available for future near-threshold\n  confirmation, but are intentionally out of scope for this change.\n",
+          "internal.run_id": "01KSAPQQSVK4FY3CWHYVZYD7T6",
+          "internal.fidelity": "compact",
+          "internal.retry_count.start": 0,
+          "internal.thread_id": null,
+          "failure_class": "",
+          "internal.work_dir": "/home/daytona/workspace/fabro",
+          "outcome": "succeeded"
+        },
+        "node_outcomes": {
+          "start": {
+            "status": "succeeded",
+            "usage": null
+          }
+        },
+        "next_node_id": "toolchain",
+        "node_visits": {
+          "start": 1
+        }
+      },
+      "diff": {}
+    },
+    {
+      "seq": 0,
+      "checkpoint": {
+        "timestamp": "2026-05-23T15:20:14.019032Z",
+        "current_node": "toolchain",
+        "completed_nodes": [
+          "start",
+          "toolchain"
+        ],
+        "node_retries": {},
+        "context_values": {
+          "internal.work_dir": "/home/daytona/workspace/fabro",
+          "current_node": "toolchain",
+          "graph.rankdir": "LR",
+          "internal.run_id": "01KSAPQQSVK4FY3CWHYVZYD7T6",
+          "outcome": "succeeded",
+          "internal.thread_id": "start",
+          "thread.start.current_node": "toolchain",
+          "internal.retry_count.toolchain": 0,
+          "internal.node_visit_count": 1,
+          "failure_class": "",
+          "graph.model_stylesheet": "\n            * { model: claude-opus-4-7; }\n        ",
+          "internal.retry_count.start": 0,
+          "internal.fidelity": "compact",
+          "command.output": "blob://sha256/fc14b2ba2d770e5cd3169df7a29525c962adfc4cfa3097b9098c63ebd61a748c",
+          "graph.goal": "# Agent Compaction API Usage Baseline Plan\n\nDate: 2026-05-23\n\n## Summary\n\nChange Fabro's agent compaction trigger from a whole-history `chars / 4`\nestimate to a Claude Code-style hot-path estimate: use the latest real\nassistant response's stored `usage.total_tokens()` as the baseline, then add\nlocal estimates for turns appended after that response. This avoids token-count\nprovider API calls while making compaction sensitive to actual\nprovider-reported context usage, including cache and reasoning tokens.\n\nNo provider token-count API calls should be added in this change.\n\n## Key Changes\n\n- Replace the current compaction estimate in `fabro-agent` with a new\n  active-context estimator.\n- Find the newest assistant turn whose `usage.total_tokens() > 0`.\n- Use that `usage.total_tokens()` as the baseline.\n- Add local estimates only for turns after that assistant turn.\n- If no usable assistant usage exists, fall back to the existing local\n  whole-history estimate.\n- Reuse a shared per-turn local estimate helper so fallback and post-baseline\n  delta counting stay consistent.\n- Keep the estimator local and in-process. Do not call\n  `llm_client.count_input_tokens()` from `compact_if_needed()`.\n\n## Implementation Details\n\n- Update `lib/crates/fabro-agent/src/compaction.rs`:\n  - Add an estimator that returns both token count and method, for example\n    `ApiUsagePlusLocalDelta` or `LocalEstimate`.\n  - Make `check_context_usage()` use the new estimator and include the method\n    in warning `details`.\n  - Make `compact_context()` report the same improved estimate in\n    `CompactionStarted`.\n  - Move `CompactionStarted` emission after the\n    `original_turn_count <= preserve_count` no-op check, so a no-op compact\n    cannot emit started without completed.\n- Update `lib/crates/fabro-agent/src/history.rs`:\n  - In `History::compact()`, invalidate preserved assistant usage by replacing\n    preserved assistant `usage` with `TokenCounts::default()`.\n  - Keep provider parts, response IDs, text, and tool calls unchanged.\n  - Rationale: preserved assistant usage reflects the pre-compaction context and\n    must not become the next baseline. Billing remains available from emitted\n    run events, so mutable runtime history should prefer compaction correctness.\n- Leave public run event names and schemas unchanged:\n  - `agent.compaction.started`\n  - `agent.compaction.completed`\n  - Existing warning event remains a warning with richer `details`.\n\n## Test Plan\n\n- Add unit coverage in `lib/crates/fabro-agent/src/compaction.rs`:\n  - No assistant usage: estimator matches current local whole-history behavior.\n  - Latest assistant usage present: estimator uses `usage.total_tokens()` plus\n    only later tool/user/steering turns.\n  - Usage fields include cache and reasoning through `TokenCounts::total_tokens()`.\n  - Earlier assistant usage is ignored when a later assistant usage exists.\n- Add unit coverage in `lib/crates/fabro-agent/src/history.rs`:\n  - `History::compact()` preserves assistant content, tool calls, and provider\n    parts, but resets preserved assistant usage to default.\n  - Existing OpenAI opaque stripping and Anthropic thinking preservation tests\n    still pass.\n- Add session coverage in `lib/crates/fabro-agent/src/session.rs`:\n  - A short assistant response with high `usage.total_tokens()` triggers\n    compaction even when text length is small.\n  - A compact no-op due to `turns.len() <= preserve_count` does not emit\n    `CompactionStarted`.\n  - Compaction disabled still prevents compaction even if the API usage\n    baseline exceeds threshold.\n- Run targeted verification:\n  - `cargo nextest run -p fabro-agent compaction`\n  - `cargo nextest run -p fabro-agent history`\n  - If those pass, run `cargo nextest run -p fabro-agent`.\n\n## Assumptions\n\n- Runtime/session `Message::Assistant.usage` is safe to invalidate after\n  compaction because authoritative billing comes from emitted workflow/run\n  events, not preserved mutable agent history.\n- A zero-token `TokenCounts::default()` should be treated as no usable API\n  baseline.\n- Provider token-count APIs remain available for future near-threshold\n  confirmation, but are intentionally out of scope for this change.\n",
+          "failure_signature": ""
+        },
+        "node_outcomes": {
+          "start": {
+            "status": "succeeded",
+            "usage": null
+          },
+          "toolchain": {
+            "status": "succeeded",
+            "context_updates": {
+              "command.output": "blob://sha256/fc14b2ba2d770e5cd3169df7a29525c962adfc4cfa3097b9098c63ebd61a748c"
+            },
+            "notes": "Script completed: command -v cargo >/dev/null || { curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y && sudo ln -sf $HOME/.cargo/bin/* /usr/local/bin/; }; cargo --version 2>&1",
+            "usage": null
+          }
+        },
+        "next_node_id": "preflight_compile",
+        "node_visits": {
+          "toolchain": 1,
+          "start": 1
+        }
+      },
+      "diff": {}
+    }
+  ],
   "conclusion": null,
   "sandbox": {
     "provider": "daytona",
@@ -543,5 +635,67 @@
   "pull_request": null,
   "superseded_by": null,
   "pending_interviews": {},
-  "stages": {}
+  "stages": {
+    "start@1": {
+      "first_event_seq": 16,
+      "prompt": null,
+      "response": null,
+      "completion": {
+        "outcome": "succeeded",
+        "notes": null,
+        "failure_reason": null,
+        "timestamp": "2026-05-23T15:20:11.504080Z"
+      },
+      "provider_used": null,
+      "diff": null,
+      "script_invocation": null,
+      "script_timing": null,
+      "parallel_results": null,
+      "output": null,
+      "started_at": "2026-05-23T15:20:11.503608Z",
+      "handler": "start",
+      "timing": {
+        "wall_time_ms": 0,
+        "inference_time_ms": 0,
+        "tool_time_ms": 0,
+        "active_time_ms": 0
+      },
+      "usage": {
+        "input_tokens": 0,
+        "output_tokens": 0,
+        "total_tokens": 0,
+        "reasoning_tokens": 0,
+        "cache_read_tokens": 0,
+        "cache_write_tokens": 0
+      },
+      "state": "succeeded"
+    },
+    "toolchain@1": {
+      "first_event_seq": 20,
+      "prompt": null,
+      "response": null,
+      "completion": null,
+      "provider_used": null,
+      "diff": null,
+      "script_invocation": {
+        "script": "command -v cargo >/dev/null || { curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y && sudo ln -sf $HOME/.cargo/bin/* /usr/local/bin/; }; cargo --version 2>&1",
+        "command": "exec 2>&1\ncommand -v cargo >/dev/null || { curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y && sudo ln -sf $HOME/.cargo/bin/* /usr/local/bin/; }; cargo --version 2>&1",
+        "language": "shell"
+      },
+      "script_timing": null,
+      "parallel_results": null,
+      "output": null,
+      "started_at": "2026-05-23T15:20:11.504411Z",
+      "handler": "command",
+      "usage": {
+        "input_tokens": 0,
+        "output_tokens": 0,
+        "total_tokens": 0,
+        "reasoning_tokens": 0,
+        "cache_read_tokens": 0,
+        "cache_write_tokens": 0
+      },
+      "state": "running"
+    }
+  }
 }
\ No newline at end of file
diff --git a/stages/001-start@1/status.json b/stages/001-start@1/status.json
new file mode 100644
index 000000000..b31411130
--- /dev/null
+++ b/stages/001-start@1/status.json
@@ -0,0 +1,6 @@
+{
+  "outcome": "succeeded",
+  "notes": null,
+  "failure_reason": null,
+  "timestamp": "2026-05-23T15:20:11.504080Z"
+}
\ No newline at end of file
diff --git a/stages/002-toolchain@1/script_invocation.json b/stages/002-toolchain@1/script_invocation.json
new file mode 100644
index 000000000..92c244949
--- /dev/null
+++ b/stages/002-toolchain@1/script_invocation.json
@@ -0,0 +1,5 @@
+{
+  "script": "command -v cargo >/dev/null || { curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y && sudo ln -sf $HOME/.cargo/bin/* /usr/local/bin/; }; cargo --version 2>&1",
+  "command": "exec 2>&1\ncommand -v cargo >/dev/null || { curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y && sudo ln -sf $HOME/.cargo/bin/* /usr/local/bin/; }; cargo --version 2>&1",
+  "language": "shell"
+}
\ No newline at end of file