Merge pull request #1459 from link-assistant/issue-1458-95ae87b62e25

fix: interactive mode comment output - shell quoting, queue deadlock, duplicate init

Author: konard

.changeset/fix-interactive-mode-comments.md

@@ -0,0 +1,5 @@
+---
+'@link-assistant/hive-mind': patch
+---
+
+Fix interactive mode PR comment output: use stdin for GitHub API calls to prevent shell quoting corruption, flush comment queue before tool result timeout to prevent stuck "Waiting for result..." comments, and guard against duplicate session started comments from late system.init events

docs/case-studies/issue-1458/analysis.md

@@ -0,0 +1,124 @@
+# Case Study: Interactive Mode Comment Output Issues (#1458)
+
+## Overview
+
+Investigation of 89+ broken PR comments during an interactive Claude session on
+`xlab2016/space_db_private/pull/20`. Three distinct failure modes were identified,
+each with a different root cause.
+
+**Source data:**
+
+- PR: `xlab2016/space_db_private/pull/20` (339 total comments)
+- Solution draft log: 51,196 lines (from gist)
+- Representative samples saved in this directory as JSON files
+
+---
+
+## Issue 1: Empty/Corrupted Markdown Fields (27 comments)
+
+### Symptoms
+
+Comments displayed with literal empty fields like:
+
+```
+**Pattern:** ''
+**File:** ''
+**Path:** ''
+```
+
+Bodies contained literal single-quote characters (char code 39) wrapping field values.
+
+### Root Cause
+
+`command-stream`'s `quote()` function wraps values in shell single quotes:
+`'value'` → which requires a shell to interpret. However, `gh api -f body=...`
+passed via `execFile` (which bypasses the shell) received the quotes as literal
+characters instead of shell metacharacters.
+
+The comment body was correctly constructed in JavaScript, but corrupted during
+the GitHub API call because `command-stream`'s tagged template literal produced
+shell-quoted strings that were never shell-interpreted.
+
+### Fix
+
+Replaced `command-stream`'s `$` template literal calls with direct `execFileAsync`
+using `--input -` to pass the JSON body via stdin. This completely bypasses shell
+quoting and ensures the body arrives at the GitHub API exactly as constructed.
+
+---
+
+## Issue 2: Comments Not Updated With Results (61 comments)
+
+### Symptoms
+
+Tool use comments permanently showed "⏳ Waiting for result..." instead of being
+edited with the tool's output when the result arrived.
+
+### Root Cause
+
+Deadlock between `handleToolResult` and `processQueue`:
+
+1. `handleToolResult` awaits `commentIdPromise` (resolved when the tool_use
+   comment is posted and gets a comment ID)
+2. `commentIdPromise` is resolved by `processQueue` processing the comment queue
+3. `processQueue` runs in `processEvent`'s `finally` block
+4. But `processEvent` is blocked because `handleToolResult` is still awaiting
+
+This created a circular dependency. After 30 seconds, the timeout fired and
+`handleToolResult` posted a separate "result" comment instead of editing the
+original. Evidence from logs:
+
+```
+21:55:38 - Tool use queued for comment
+21:56:08 - "Timeout waiting for tool use comment, posting result separately"
+```
+
+### Fix
+
+Added explicit queue flushing in `handleToolResult` before waiting for the
+comment ID promise. This breaks the deadlock by processing any pending queue
+items (including the tool_use comment) immediately.
+
+---
+
+## Issue 3: Duplicate Session Started Comment (1 comment)
+
+### Symptoms
+
+A second "🚀 Interactive session started" comment appeared at `22:40:53`,
+well after the session was already active (first init at `21:54:53`).
+
+### Root Cause
+
+A `task_notification` event for a background dotnet SDK installation arrived
+at `22:40:52` (after the `result` event at `22:40:37`). This late notification
+triggered Claude CLI to emit a second `system.init` event with the same session
+ID at `22:40:53`.
+
+Timeline from logs:
+
+```
+21:54:53 - First system.init (session starts)
+22:40:37 - result event (session effectively ends)
+22:40:52 - Late task_notification (background dotnet install)
+22:40:53 - Second system.init (same session ID - causes duplicate comment)
+```
+
+### Fix
+
+Added a guard in `handleSystemInit` that checks if `state.sessionId` is already
+set. If a session is already initialized, the duplicate `system.init` event is
+silently ignored (with verbose logging for diagnostics).
+
+---
+
+## Files Changed
+
+- `src/interactive-mode.lib.mjs` — All three fixes applied
+- `tests/test-interactive-mode.mjs` — Updated test mocks for new `execFileAsync` pattern
+- `src/unicode-sanitization.lib.mjs` — No changes (already correct from #1324)
+
+## Testing
+
+All 91 tests pass after the fixes, including existing regression tests for
+unicode sanitization (#1324) and agent task events (#1450).

docs/case-studies/issue-1458/duplicate-session.json

@@ -0,0 +1,8 @@
+[
+  {
+    "id": 4104682726,
+    "created_at": "2026-03-21T22:40:53Z",
+    "updated_at": "2026-03-21T22:40:53Z",
+    "body": "## 🚀 Interactive session started\n\n| Property | Value |\n|----------|-------|\n| **Session ID** | `2024f25c-03c6-4eb4-b7ed-0b2847653bba` |\n| **Model** | `claude-sonnet-4-6` |\n| **Claude Code Version** | `2.1.81` |\n| **Permission Mode** | `bypassPermissions` |\n| **Working Directory** | `/tmp/gh-issue-solver-1774130096902` |\n| **Available Tools** | `Task`, `TaskOutput`, `Bash`, `Glob`, `Grep`, `ExitPlanMode`, `Read`, `Edit`, `Write`, `NotebookEdit`, `WebFetch`, `TodoWrite`, `WebSearch`, `TaskStop`, `AskUserQuestion`, `Skill`, `EnterPlanMode`, `EnterWorktree`, `ExitWorktree`, `CronCreate`, `CronDelete`, `CronList`, `ToolSearch` |\n| **MCP Servers** | `claude.ai Google Calendar` (needs-auth), `claude.ai Gmail` (needs-auth) |\n| **Slash Commands** | `/update-config`, `/debug`, `/simplify`, `/batch`, `/loop`, `/claude-api`, `/compact`, `/context`, `/cost`, `/heapdump`, `/init`, `/pr-comments`, `/release-notes`, `/review`, `/security-review`, `/extra-usage`, `/insights` |\n| **Agents** | `general-purpose`, `statusline-setup`, `Explore`, `Plan` |\n\n---\n\n<details>\n<summary>📄 Raw JSON</summary>\n\n```json\n[\n  {\n    \"type\": \"system\",\n    \"subtype\": \"init\",\n    \"cwd\": \"/tmp/gh-issue-solver-1774130096902\",\n    \"session_id\": \"2024f25c-03c6-4eb4-b7ed-0b2847653bba\",\n    \"tools\": [\n      \"Task\",\n      \"TaskOutput\",\n      \"Bash\",\n      \"Glob\",\n      \"Grep\",\n      \"ExitPlanMode\",\n      \"Read\",\n      \"Edit\",\n      \"Write\",\n      \"NotebookEdit\",\n      \"WebFetch\",\n      \"TodoWrite\",\n      \"WebSearch\",\n      \"TaskStop\",\n      \"AskUserQuestion\",\n      \"Skill\",\n      \"EnterPlanMode\",\n      \"EnterWorktree\",\n      \"ExitWorktree\",\n      \"CronCreate\",\n      \"CronDelete\",\n      \"CronList\",\n      \"ToolSearch\"\n    ],\n    \"mcp_servers\": [\n      {\n        \"name\": \"claude.ai Google Calendar\",\n        \"status\": \"needs-auth\"\n      },\n      {\n        \"name\": \"claude.ai Gmail\",\n        \"status\": \"needs-auth\"\n      }\n    ],\n    \"model\": \"claude-sonnet-4-6\",\n    \"permissionMode\": \"bypassPermissions\",\n    \"slash_commands\": [\n      \"update-config\",\n      \"debug\",\n      \"simplify\",\n      \"batch\",\n      \"loop\",\n      \"claude-api\",\n      \"compact\",\n      \"context\",\n      \"cost\",\n      \"heapdump\",\n      \"init\",\n      \"pr-comments\",\n      \"release-notes\",\n      \"review\",\n      \"security-review\",\n      \"extra-usage\",\n      \"insights\"\n    ],\n    \"apiKeySource\": \"none\",\n    \"claude_code_version\": \"2.1.81\",\n    \"output_style\": \"default\",\n    \"agents\": [\n      \"general-purpose\",\n      \"statusline-setup\",\n      \"Explore\",\n      \"Plan\"\n    ],\n    \"skills\": [\n      \"update-config\",\n      \"debug\",\n      \"simplify\",\n      \"batch\",\n      \"loop\",\n      \"claude-api\"\n    ],\n    \"plugins\": [],\n    \"uuid\": \"4af5eeb6-e2b5-46e8-b64f-165455a16779\",\n    \"fast_mode_state\": \"off\"\n  }\n]\n```\n\n</details>"
+  }
+]

docs/case-studies/issue-1458/markdown-formatting.json

@@ -0,0 +1,26 @@
+[
+  {
+    "id": 4104659088,
+    "created_at": "2026-03-21T22:33:51Z",
+    "updated_at": "2026-03-21T22:33:51Z",
+    "body": "## 🤖 Agent task: wget -q https://dot.net/v1/dotnet-install.sh -O /tmp/dotnet-install.sh && chmod +x /tmp/dotnet-install.sh && /tmp/dotnet-install.sh --version 9.0.100 --install-dir /tmp/dotnet9 > /tmp/dotnet_install.log 2>&1; echo \"Exit: $?\"\n\n| Property | Value |\n|----------|-------|\n| **Task ID** | `bk8qj36e3` |\n| **Type** | `local_bash` |\n| **Status** | ⏳ Running... |\n\n\n---\n\n<details>\n<summary>📄 Raw JSON</summary>\n\n```json\n[\n  {\n    \"type\": \"system\",\n    \"subtype\": \"task_started\",\n    \"task_id\": \"bk8qj36e3\",\n    \"tool_use_id\": \"toolu_01WVXmWkiUzVQiNTKHhPYdYp\",\n    \"description\": \"wget -q https://dot.net/v1/dotnet-install.sh -O /tmp/dotnet-install.sh && chmod +x /tmp/dotnet-install.sh && /tmp/dotnet-install.sh --version 9.0.100 --install-dir /tmp/dotnet9 > /tmp/dotnet_install.log 2>&1; echo \\\"Exit: $?\\\"\",\n    \"task_type\": \"local_bash\",\n    \"uuid\": \"b3953022-c072-4943-abe7-b651a56a5eef\",\n    \"session_id\": \"2024f25c-03c6-4eb4-b7ed-0b2847653bba\"\n  }\n]\n```\n\n</details>"
+  },
+  {
+    "id": 4104661257,
+    "created_at": "2026-03-21T22:34:20Z",
+    "updated_at": "2026-03-21T22:34:21Z",
+    "body": "'## 🔎 Grep tool use\n\n**Pattern:** \n**Path:** \n\n<details open>\n<summary>📤 Output (✅ success)</summary>\n\n\n\n</details>\n\n---\n\n<details>\n<summary>📄 Raw JSON</summary>\n\n\n\n</details>'"
+  },
+  {
+    "id": 4104662013,
+    "created_at": "2026-03-21T22:34:32Z",
+    "updated_at": "2026-03-21T22:34:33Z",
+    "body": "'## 🔎 Grep tool use\n\n**Pattern:** \n**Path:** \n\n<details open>\n<summary>📤 Output (✅ success)</summary>\n\n\n\n</details>\n\n---\n\n<details>\n<summary>📄 Raw JSON</summary>\n\n\n\n</details>'"
+  },
+  {
+    "id": 4104663699,
+    "created_at": "2026-03-21T22:35:02Z",
+    "updated_at": "2026-03-21T22:35:03Z",
+    "body": "'## 📖 Read tool use\n\n**File:** \n**Range:** offset=338, limit=60\n\n<details open>\n<summary>📤 Output (✅ success)</summary>\n\n\n\n</details>\n\n---\n\n<details>\n<summary>📄 Raw JSON</summary>\n\n\n\n</details>'"
+  }
+]

docs/case-studies/issue-1458/not-updated-with-results.json

@@ -0,0 +1,14 @@
+[
+  {
+    "id": 4104659844,
+    "created_at": "2026-03-21T22:34:03Z",
+    "updated_at": "2026-03-21T22:34:03Z",
+    "body": "## 📖 Read tool use\n\n**File:** `/tmp/gh-issue-solver-1774130096902/src/Libs/Magic.Kernel/Devices/Streams/ClawStreamDevice.cs`\n\n_⏳ Waiting for result..._\n\n---\n\n<details>\n<summary>📄 Raw JSON</summary>\n\n```json\n[\n  {\n    \"type\": \"assistant\",\n    \"message\": {\n      \"model\": \"claude-sonnet-4-6\",\n      \"id\": \"msg_01AAfDNY2uhUojtFyWK53J1y\",\n      \"type\": \"message\",\n      \"role\": \"assistant\",\n      \"content\": [\n        {\n          \"type\": \"tool_use\",\n          \"id\": \"toolu_01Ka24aQ9hDWQnyicRJUJQLr\",\n          \"name\": \"Read\",\n          \"input\": {\n            \"file_path\": \"/tmp/gh-issue-solver-1774130096902/src/Libs/Magic.Kernel/Devices/Streams/ClawStreamDevice.cs\"\n          },\n          \"caller\": {\n            \"type\": \"direct\"\n          }\n        }\n      ],\n      \"stop_reason\": null,\n      \"stop_sequence\": null,\n      \"usage\": {\n        \"input_tokens\": 1,\n        \"cache_creation_input_tokens\": 297,\n        \"cache_read_input_tokens\": 67583,\n        \"cache_creation\": {\n          \"ephemeral_5m_input_tokens\": 0,\n          \"ephemeral_1h_input_tokens\": 297\n        },\n        \"output_tokens\": 1,\n        \"service_tier\": \"standard\",\n        \"inference_geo\": \"not_available\"\n      },\n      \"context_management\": null\n    },\n    \"parent_tool_use_id\": null,\n    \"session_id\": \"2024f25c-03c6-4eb4-b7ed-0b2847653bba\",\n    \"uuid\": \"0e64635d-204c-4123-a0cd-050359faf057\"\n  }\n]\n```\n\n</details>"
+  },
+  {
+    "id": 4104662688,
+    "created_at": "2026-03-21T22:34:44Z",
+    "updated_at": "2026-03-21T22:34:44Z",
+    "body": "## 🔎 Grep tool use\n\n**Pattern:** `case Opcodes.GetObj|GetObjAsync|ExecuteGetObj`\n**Path:** `/tmp/gh-issue-solver-1774130096902/src/Libs/Magic.Kernel/Interpretation/Interpreter.cs`\n\n_⏳ Waiting for result..._\n\n---\n\n<details>\n<summary>📄 Raw JSON</summary>\n\n```json\n[\n  {\n    \"type\": \"assistant\",\n    \"message\": {\n      \"model\": \"claude-sonnet-4-6\",\n      \"id\": \"msg_01J2Rnbj3F2YBaQwJTrPVhXQ\",\n      \"type\": \"message\",\n      \"role\": \"assistant\",\n      \"content\": [\n        {\n          \"type\": \"tool_use\",\n          \"id\": \"toolu_01RqbZy4ZLu81R4rsHbhNPio\",\n          \"name\": \"Grep\",\n          \"input\": {\n            \"pattern\": \"case Opcodes.GetObj|GetObjAsync|ExecuteGetObj\",\n            \"path\": \"/tmp/gh-issue-solver-1774130096902/src/Libs/Magic.Kernel/Interpretation/Interpreter.cs\",\n            \"output_mode\": \"content\",\n            \"context\": 15,\n            \"head_limit\": 50\n          },\n          \"caller\": {\n            \"type\": \"direct\"\n          }\n        }\n      ],\n      \"stop_reason\": null,\n      \"stop_sequence\": null,\n      \"usage\": {\n        \"input_tokens\": 1,\n        \"cache_creation_input_tokens\": 1364,\n        \"cache_read_input_tokens\": 78387,\n        \"cache_creation\": {\n          \"ephemeral_5m_input_tokens\": 0,\n          \"ephemeral_1h_input_tokens\": 1364\n        },\n        \"output_tokens\": 1,\n        \"service_tier\": \"standard\",\n        \"inference_geo\": \"not_available\"\n      },\n      \"context_management\": null\n    },\n    \"parent_tool_use_id\": null,\n    \"session_id\": \"2024f25c-03c6-4eb4-b7ed-0b2847653bba\",\n    \"uuid\": \"3924e669-9666-4db0-b2af-2c11f9d89a05\"\n  }\n]\n```\n\n</details>"
+  }
+]