@link-assistant/hive-mind 1.50.10 → 1.50.12

package/CHANGELOG.md

@@ -1,5 +1,19 @@
 # @link-assistant/hive-mind
 
+## 1.50.12
+
+### Patch Changes
+
+- 065deae: ## Summary
+
+  Fix Playwright MCP setup guidance and verification for Codex environments.
+
+## 1.50.11
+
+### Patch Changes
+
+- bf9cf54: Fix prompt template builders crashing when literal `.png` appears in screenshot guidance.
+
 ## 1.50.10
 
 ### Patch Changes

package/README.md

@@ -174,6 +174,10 @@ claude
 # Optionally test Claude connection
 claude -p hi --model haiku
 
+# Verify Playwright MCP is registered for both CLIs in this container image
+claude mcp list | grep playwright
+codex mcp list | grep playwright
+
 # You might need to update hive-mind and agent to latest versions:
 bun install -g @link-assistant/hive-mind
 bun install -g @link-assistant/agent
@@ -195,25 +199,28 @@ docker attach hive-mind
 
 # --- Persisting auth data across restarts ---
 
-# Extract auth data from a running (or stopped) container to the host:
-mkdir -p ~/.hive-mind
-docker cp hive-mind:/workspace/.claude ~/.hive-mind/claude
-docker cp hive-mind:/workspace/.claude.json ~/.hive-mind/claude.json
-docker cp hive-mind:/workspace/.config/gh ~/.hive-mind/gh
-
-# Fix ownership to match the sandbox user inside the container:
-SANDBOX_UID=$(docker exec hive-mind id -u sandbox)
-chown -R $SANDBOX_UID:$SANDBOX_UID ~/.hive-mind/claude ~/.hive-mind/gh
-chown $SANDBOX_UID:$SANDBOX_UID ~/.hive-mind/claude.json
+# On the host, create the directories used by the current Docker workflow:
+mkdir -p /root/.hive-mind/claude /root/.hive-mind/codex /root/.hive-mind/gh
+touch -a /root/.hive-mind/claude.json
 
-# On subsequent runs, mount the auth data to keep it between restarts:
-docker run -dit \
-  --name hive-mind \
-  --restart unless-stopped \
+# In our Docker images HOME=/workspace, so Codex stores its data in /workspace/.codex.
+# Mount the full Codex directory so auth.json, config.toml, and sessions survive restarts.
+docker run -dit --user sandbox --name hive-mind --restart unless-stopped \
   -v /root/.hive-mind/claude:/workspace/.claude \
+  -v /root/.hive-mind/codex:/workspace/.codex \
   -v /root/.hive-mind/claude.json:/workspace/.claude.json \
   -v /root/.hive-mind/gh:/workspace/.config/gh \
-  konard/hive-mind:latest
+  konard/hive-mind:latest bash -l -c 'bash /workspace/start-bot.sh'
+
+# After the first start, fix ownership to match the sandbox user inside the container:
+SANDBOX_UID=$(docker exec hive-mind id -u sandbox)
+chown -R $SANDBOX_UID:$SANDBOX_UID /root/.hive-mind/claude /root/.hive-mind/codex /root/.hive-mind/gh
+chown $SANDBOX_UID:$SANDBOX_UID /root/.hive-mind/claude.json
+
+# Important: mounted ~/.codex data overrides the image-baked Codex config.
+# If the host directory was created before Playwright MCP was added to the image,
+# re-register it once inside the running container:
+docker exec -it hive-mind bash -lc 'codex mcp list && codex mcp add playwright -- npx -y @playwright/mcp@latest --isolated --headless --no-sandbox --timeout-action=600000 --viewport-size 1920x1080'
 ```
 
 **Benefits of Docker:**
@@ -224,6 +231,8 @@ docker run -dit \
 - ✅ Easy to run multiple instances with different GitHub accounts
 - ✅ Consistent environment across different machines
 
+The Docker image itself now registers Playwright MCP for both Claude and Codex during build, and CI verifies those registrations in the built container. If `codex mcp list` is still empty in a running container, the usual cause is not the published image itself but a mounted `/workspace/.codex` directory from the host that replaces the image's default Codex configuration.
+
 See [docs/DOCKER.md](./docs/DOCKER.md) for advanced Docker usage.
 
 #### Stoping and removing docker container

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@link-assistant/hive-mind",
-  "version": "1.50.10",
+  "version": "1.50.12",
   "description": "AI-powered issue solver and hive mind for collaborative problem solving",
   "main": "src/hive.mjs",
   "type": "module",
@@ -20,8 +20,8 @@
     "lint": "eslint 'src/**/*.{js,mjs,cjs}'",
     "lint:fix": "eslint 'src/**/*.{js,mjs,cjs}' --fix",
     "check:duplication": "jscpd .",
-    "format": "prettier --write \"**/*.{js,mjs,json,md}\"",
-    "format:check": "prettier --check \"**/*.{js,mjs,json,md}\"",
+    "format": "prettier --write \"**/*.{js,mjs,json,md}\" --ignore-path .prettierignore",
+    "format:check": "prettier --check \"**/*.{js,mjs,json,md}\" --ignore-path .prettierignore",
     "changeset": "changeset",
     "changeset:version": "changeset version",
     "changeset:publish": "npm run build:pre && changeset publish",

package/pr-1607-body.md

@@ -0,0 +1,34 @@
+## Summary
+
+Fixes #1606 by documenting and verifying Playwright MCP registration for Codex in addition to Claude Code, including the Docker-specific case where persisted Codex state overrides the image defaults.
+
+## Root Cause
+
+The reported environment had `@playwright/mcp` installed and registered in Claude, but `codex mcp list` had no configured servers. The immediate cause was missing Codex MCP registration. Local reproduction confirmed that `/workspace/.codex/config.toml` can exist without a Playwright MCP entry and that `codex mcp add playwright ...` fixes the state immediately. In Docker deployments, the most likely explanation is that a host-mounted `/workspace/.codex` directory preserved an older Codex config and replaced the image-baked MCP registration. Existing docs and helper scripts also focused mainly on Claude setup, so the mismatch was easy to miss even though `/version` already reported it correctly.
+
+## Changes
+
+- added regression coverage for the mixed MCP state where Claude is connected and Codex is not
+- updated Playwright MCP verification and integration scripts to check Codex MCP registration explicitly
+- updated Docker verification to fail if Claude or Codex is missing the Playwright MCP registration
+- updated configuration and Docker docs to include both `claude mcp add ...` and `codex mcp add ...`
+- documented that mounting `/workspace/.codex` can override the image defaults and reintroduce the problem
+- added the investigation record and collected evidence under `docs/case-studies/issue-1606`
+
+## Reproduction
+
+1. Install `@playwright/mcp` and register it only with Claude.
+2. Run `claude mcp list` and confirm `playwright` is present.
+3. Run `codex mcp list` and observe `No MCP servers configured yet`.
+4. Run `/version` and observe `Playwright MCP: <version> | Claude Code: connected | Codex: not connected`.
+
+## Verification
+
+- `node tests/test-version-info.mjs`
+- `node tests/test-version-parsing.mjs`
+- `bash scripts/verify-docker-image.sh`
+
+## Evidence
+
+- case study: `docs/case-studies/issue-1606/README.md`
+- PR: https://github.com/link-assistant/hive-mind/pull/1607

package/src/claude.prompts.lib.mjs

@@ -160,7 +160,7 @@ Initial research.
    - When you start, create a detailed plan for yourself and follow your todo list step by step. Add as many relevant points from these guidelines to the todo list as practical so you can track the work clearly.
    - When the user mentions CI failures or asks to investigate logs, consider adding these todos to track the investigation: (1) list recent CI runs with timestamps, (2) download logs from failed runs to the ci-logs/ directory, (3) analyze error messages and identify the root cause, (4) implement a fix, (5) verify that the fix resolves the specific errors found in the logs.
    - When you read the issue, read all details and comments thoroughly.
-   - When you see screenshots or images in issue descriptions, pull request descriptions, comments, or discussions, download the image to a local file first, then use the Read tool to view and analyze it. Before reading downloaded images with the Read tool, verify that the file is a valid image rather than HTML by using a CLI tool such as the 'file' command. When corrupted or non-image files, such as GitHub "Not Found" pages saved as `.png`, are read, they can cause "Could not process image" errors and crash the AI solver process. When the file command shows "HTML", "text", or "ASCII text", the download failed, so do not call Read on that file. Instead: (1) when images are from GitHub issues or PRs, such as URLs containing "github.com/user-attachments", retry with: curl -L -H "Authorization: token $(gh auth token)" -o <filename> "<url>" (2) when the retry still fails, skip the image and note that it was unavailable.
+   - When you see screenshots or images in issue descriptions, pull request descriptions, comments, or discussions, download the image to a local file first, then use the Read tool to view and analyze it. Before reading downloaded images with the Read tool, verify that the file is a valid image rather than HTML by using a CLI tool such as the 'file' command. When corrupted or non-image files, such as GitHub "Not Found" pages saved as \`.png\`, are read, they can cause "Could not process image" errors and crash the AI solver process. When the file command shows "HTML", "text", or "ASCII text", the download failed, so do not call Read on that file. Instead: (1) when images are from GitHub issues or PRs, such as URLs containing "github.com/user-attachments", retry with: curl -L -H "Authorization: token $(gh auth token)" -o <filename> "<url>" (2) when the retry still fails, skip the image and note that it was unavailable.
    - When you need issue details, use gh issue view https://github.com/${owner}/${repo}/issues/${issueNumber}.
    - When you need related code, use gh search code --owner ${owner} [keywords].
    - When you need repo context, read files in your working directory.${

package/src/opencode.prompts.lib.mjs

@@ -137,7 +137,7 @@ ${workspaceInstructions}
 Initial research.
    - When you start, create a detailed plan for yourself and follow your todo list step by step. Add as many relevant points from these guidelines to the todo list as practical so you can track the work clearly.
    - When you read the issue, read all details and comments thoroughly.
-   - When you see screenshots or images in issue descriptions, pull request descriptions, comments, or discussions, download the image to a local file first, then use the Read tool to view and analyze it. Before reading downloaded images with the Read tool, verify that the file is a valid image rather than HTML by using a CLI tool such as the 'file' command. When corrupted or non-image files, such as GitHub "Not Found" pages saved as `.png`, are read, they can cause "Could not process image" errors and crash the AI solver process. When the file command shows "HTML", "text", or "ASCII text", the download failed, so do not call Read on that file. Instead: (1) when images are from GitHub issues or PRs, such as URLs containing "github.com/user-attachments", retry with: curl -L -H "Authorization: token $(gh auth token)" -o <filename> "<url>" (2) when the retry still fails, skip the image and note that it was unavailable.
+   - When you see screenshots or images in issue descriptions, pull request descriptions, comments, or discussions, download the image to a local file first, then use the Read tool to view and analyze it. Before reading downloaded images with the Read tool, verify that the file is a valid image rather than HTML by using a CLI tool such as the 'file' command. When corrupted or non-image files, such as GitHub "Not Found" pages saved as \`.png\`, are read, they can cause "Could not process image" errors and crash the AI solver process. When the file command shows "HTML", "text", or "ASCII text", the download failed, so do not call Read on that file. Instead: (1) when images are from GitHub issues or PRs, such as URLs containing "github.com/user-attachments", retry with: curl -L -H "Authorization: token $(gh auth token)" -o <filename> "<url>" (2) when the retry still fails, skip the image and note that it was unavailable.
    - When you need issue details, use gh issue view https://github.com/${owner}/${repo}/issues/${issueNumber}.
    - When you need related code, use gh search code --owner ${owner} [keywords].
    - When you need repo context, read files in your working directory.${