fix(broker): retry Relaycast workspace busy startup admission

.agentworkforce/trajectories/completed/2026-09/traj_q7a7m644yvsi/summary.md

@@ -0,0 +1,37 @@
+# Trajectory: Retry transient Relaycast workspace_busy admission failures
+
+> **Status:** ✅ Completed
+> **Confidence:** 92%
+> **Started:** September 10, 2026 at 12:33 AM
+> **Completed:** September 10, 2026 at 12:42 AM
+
+---
+
+## Summary
+
+Added bounded Relaycast workspace_busy startup/admission retry with diagnostics-preserving exhaustion tests, unrelated-429 terminal coverage, explicit fail-closed session boundary documentation, and a patch changelog entry.
+
+**Approach:** Standard approach
+
+---
+
+## Key Decisions
+
+### Retry only the exact HTTP 429 workspace_busy admission code
+- **Chose:** Retry only the exact HTTP 429 workspace_busy admission code
+- **Reasoning:** Generic 429s may represent credential or quota policy and must remain terminal; RelayError 8.0.0 does not retain Retry-After, so the existing bounded 200ms/400ms startup backoff is the safest available fallback.
+
+### Keep complete startup handshake fail-closed after admission retry exhaustion
+- **Chose:** Keep complete startup handshake fail-closed after admission retry exhaustion
+- **Reasoning:** Replaying startup would repeat unkeyed workspace and agent registration POSTs and could create duplicate or orphaned identities.
+
+---
+
+## Chapters
+
+### 1. Work
+*Agent: default*
+
+- Retry only the exact HTTP 429 workspace_busy admission code: Retry only the exact HTTP 429 workspace_busy admission code
+- Keep complete startup handshake fail-closed after admission retry exhaustion: Keep complete startup handshake fail-closed after admission retry exhaustion
+- The broker now retries exact workspace_busy admission failures at the request boundary, while preserving terminal behavior for unrelated 429s and complete-handshake replay safety. Focused, full broker library tests and clippy are green.

.agentworkforce/trajectories/completed/2026-09/traj_q7a7m644yvsi/trajectory.json

@@ -0,0 +1,77 @@
+{
+  "id": "traj_q7a7m644yvsi",
+  "version": 1,
+  "task": {
+    "title": "Retry transient Relaycast workspace_busy admission failures"
+  },
+  "status": "completed",
+  "startedAt": "2026-09-09T22:33:18.944Z",
+  "completedAt": "2026-09-09T22:42:11.347Z",
+  "agents": [
+    {
+      "name": "default",
+      "role": "lead",
+      "joinedAt": "2026-09-09T22:42:04.687Z"
+    }
+  ],
+  "chapters": [
+    {
+      "id": "chap_73p4he5ty41v",
+      "title": "Work",
+      "agentName": "default",
+      "startedAt": "2026-09-09T22:42:04.687Z",
+      "endedAt": "2026-09-09T22:42:11.347Z",
+      "events": [
+        {
+          "ts": 1788993724688,
+          "type": "decision",
+          "content": "Retry only the exact HTTP 429 workspace_busy admission code: Retry only the exact HTTP 429 workspace_busy admission code",
+          "raw": {
+            "question": "Retry only the exact HTTP 429 workspace_busy admission code",
+            "chosen": "Retry only the exact HTTP 429 workspace_busy admission code",
+            "alternatives": [],
+            "reasoning": "Generic 429s may represent credential or quota policy and must remain terminal; RelayError 8.0.0 does not retain Retry-After, so the existing bounded 200ms/400ms startup backoff is the safest available fallback."
+          },
+          "significance": "high"
+        },
+        {
+          "ts": 1788993726731,
+          "type": "decision",
+          "content": "Keep complete startup handshake fail-closed after admission retry exhaustion: Keep complete startup handshake fail-closed after admission retry exhaustion",
+          "raw": {
+            "question": "Keep complete startup handshake fail-closed after admission retry exhaustion",
+            "chosen": "Keep complete startup handshake fail-closed after admission retry exhaustion",
+            "alternatives": [],
+            "reasoning": "Replaying startup would repeat unkeyed workspace and agent registration POSTs and could create duplicate or orphaned identities."
+          },
+          "significance": "high"
+        },
+        {
+          "ts": 1788993728782,
+          "type": "reflection",
+          "content": "The broker now retries exact workspace_busy admission failures at the request boundary, while preserving terminal behavior for unrelated 429s and complete-handshake replay safety. Focused, full broker library tests and clippy are green.",
+          "raw": {
+            "confidence": 0.92
+          },
+          "significance": "high",
+          "tags": [
+            "confidence:0.92"
+          ]
+        }
+      ]
+    }
+  ],
+  "retrospective": {
+    "summary": "Added bounded Relaycast workspace_busy startup/admission retry with diagnostics-preserving exhaustion tests, unrelated-429 terminal coverage, explicit fail-closed session boundary documentation, and a patch changelog entry.",
+    "approach": "Standard approach",
+    "confidence": 0.92
+  },
+  "commits": [],
+  "filesChanged": [],
+  "projectId": "AgentWorkforce/relay",
+  "tags": [],
+  "_trace": {
+    "startRef": "4306bb29be696ec84a4e3844f86fdc4a91584407",
+    "endRef": "4306bb29be696ec84a4e3844f86fdc4a91584407"
+  }
+}

.agentworkforce/trajectories/completed/2026-09/traj_zylcy1hl0sus/summary.md

@@ -0,0 +1,32 @@
+# Trajectory: Add RelayFlow proof for workspace_busy startup admission
+
+> **Status:** ✅ Completed
+> **Confidence:** 90%
+> **Started:** September 10, 2026 at 12:57 AM
+> **Completed:** September 10, 2026 at 12:57 AM
+
+---
+
+## Summary
+
+Added workspace-busy-startup-retry-429 RelayFlow proof covering baseline failure, head retry success, unrelated 429 terminal behavior, bounded exhaustion, and no whole-handshake replay.
+
+**Approach:** Standard approach
+
+---
+
+## Key Decisions
+
+### Used a dedicated real-broker RelayFlow case with an ephemeral HTTP admission server
+- **Chose:** Used a dedicated real-broker RelayFlow case with an ephemeral HTTP admission server
+- **Reasoning:** This gives deterministic base/head evidence for exact 429 workspace_busy behavior while checking unrelated 429 terminal handling and bounded request-scoped exhaustion without replaying registration.
+
+---
+
+## Chapters
+
+### 1. Work
+*Agent: default*
+
+- Used a dedicated real-broker RelayFlow case with an ephemeral HTTP admission server: Used a dedicated real-broker RelayFlow case with an ephemeral HTTP admission server
+- The cleanroom proof distinguishes the pre-fix terminal 429 from the retrying head and confirms unrelated throttles remain terminal; both compiled binaries passed their intended arm.

.agentworkforce/trajectories/completed/2026-09/traj_zylcy1hl0sus/trajectory.json

@@ -0,0 +1,73 @@
+{
+  "id": "traj_zylcy1hl0sus",
+  "version": 1,
+  "task": {
+    "title": "Add RelayFlow proof for workspace_busy startup admission"
+  },
+  "status": "completed",
+  "startedAt": "2026-09-09T22:57:20.071Z",
+  "completedAt": "2026-09-09T22:57:35.297Z",
+  "agents": [
+    {
+      "name": "default",
+      "role": "lead",
+      "joinedAt": "2026-09-09T22:57:31.069Z"
+    }
+  ],
+  "chapters": [
+    {
+      "id": "chap_rsah9jnq4u84",
+      "title": "Work",
+      "agentName": "default",
+      "startedAt": "2026-09-09T22:57:31.069Z",
+      "endedAt": "2026-09-09T22:57:35.297Z",
+      "events": [
+        {
+          "ts": 1788994651070,
+          "type": "decision",
+          "content": "Used a dedicated real-broker RelayFlow case with an ephemeral HTTP admission server: Used a dedicated real-broker RelayFlow case with an ephemeral HTTP admission server",
+          "raw": {
+            "question": "Used a dedicated real-broker RelayFlow case with an ephemeral HTTP admission server",
+            "chosen": "Used a dedicated real-broker RelayFlow case with an ephemeral HTTP admission server",
+            "alternatives": [],
+            "reasoning": "This gives deterministic base/head evidence for exact 429 workspace_busy behavior while checking unrelated 429 terminal handling and bounded request-scoped exhaustion without replaying registration."
+          },
+          "significance": "high"
+        },
+        {
+          "ts": 1788994653138,
+          "type": "reflection",
+          "content": "The cleanroom proof distinguishes the pre-fix terminal 429 from the retrying head and confirms unrelated throttles remain terminal; both compiled binaries passed their intended arm.",
+          "raw": {
+            "focalPoints": [
+              "base-head proof",
+              "request bounds",
+              "diagnostics"
+            ],
+            "confidence": 0.9
+          },
+          "significance": "high",
+          "tags": [
+            "focal:base-head proof",
+            "focal:request bounds",
+            "focal:diagnostics",
+            "confidence:0.9"
+          ]
+        }
+      ]
+    }
+  ],
+  "retrospective": {
+    "summary": "Added workspace-busy-startup-retry-429 RelayFlow proof covering baseline failure, head retry success, unrelated 429 terminal behavior, bounded exhaustion, and no whole-handshake replay.",
+    "approach": "Standard approach",
+    "confidence": 0.9
+  },
+  "commits": [],
+  "filesChanged": [],
+  "projectId": "AgentWorkforce/relay",
+  "tags": [],
+  "_trace": {
+    "startRef": "8f1f4059716d1a9aad91eaa18f33c4dceadd454a",
+    "endRef": "8f1f4059716d1a9aad91eaa18f33c4dceadd454a"
+  }
+}

CHANGELOG.md

@@ -5,7 +5,11 @@ All notable changes to Agent Relay will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
-## [Unreleased]
+## [Unreleased - Patch]
+
+### Fixed
+
+- `agent-relay node up` now retries the narrowly transient Relaycast `workspace_busy` admission response while keeping unrelated rate limits terminal.
 
 ## [11.10.4] - 2026-09-08