docs(05): create phase plan

Phase 05: Polish & Deploy
- 3 plans in 3 waves
- 1 parallel (Wave 1), 2 sequential (Waves 2-3)
- Ready for execution

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
Lucas Berger
2026-01-31 21:00:51 -05:00
parent 4c09d61943
commit b7cefd931e
4 changed files with 539 additions and 0 deletions
+7
View File
@@ -93,6 +93,13 @@ Plans:
**Delivers:** Production readiness
**Plans:** 3 plans
Plans:
- [ ] 05-01-PLAN.md — Remove NLU nodes and add keyword routing with persistent menu
- [ ] 05-02-PLAN.md — Standardize error messages and migrate credentials
- [ ] 05-03-PLAN.md — Write deployment README and end-to-end testing
**Status:** 🔲 Not started
---
@@ -0,0 +1,172 @@
---
phase: 05-polish-deploy
plan: 01
type: execute
wave: 1
depends_on: []
files_modified: [n8n-workflow.json]
autonomous: true
must_haves:
truths:
- "User sees persistent menu buttons in Telegram"
- "Typing 'status' triggers container list (no Claude API call)"
- "Typing 'start plex' triggers start flow (no Claude API call)"
- "Unknown commands show menu instead of error"
artifacts:
- path: "n8n-workflow.json"
provides: "Keyword Router Switch node"
contains: "Keyword Router"
- path: "n8n-workflow.json"
provides: "Persistent menu reply markup"
contains: "is_persistent"
key_links:
- from: "Telegram Trigger"
to: "Keyword Router"
via: "Route Message path"
pattern: "Keyword Router"
- from: "Keyword Router"
to: "Docker List Containers"
via: "status output"
pattern: "status.*Docker List"
---
<objective>
Replace Claude/NLU nodes with keyword-based routing and add persistent Telegram menu buttons.
Purpose: Remove external Claude API dependency, enable offline-first operation with simple keyword matching.
Output: Working keyword routing with persistent menu, no Claude API calls in workflow.
</objective>
<execution_context>
@/home/luc/.claude/get-shit-done/workflows/execute-plan.md
@/home/luc/.claude/get-shit-done/templates/summary.md
</execution_context>
<context>
@.planning/PROJECT.md
@.planning/ROADMAP.md
@.planning/STATE.md
@.planning/phases/05-polish-deploy/05-CONTEXT.md
@.planning/phases/05-polish-deploy/05-RESEARCH.md
@n8n-workflow.json
</context>
<tasks>
<task type="auto">
<name>Task 1: Remove NLU/Claude nodes and add Keyword Router</name>
<files>n8n-workflow.json</files>
<action>
Remove these nodes from the workflow:
- Prepare Claude Request
- Claude Intent Parser (HTTP Request node calling api.anthropic.com)
- Parse Intent (Code node)
- Intent Router (old Switch node routing on parsed intent)
- Send Unknown Intent
- Send Intent Error
- Remove Anthropic API Key credential reference
Replace with a single Switch node called "Keyword Router" with these rules (case-insensitive matching on message.text):
- Contains "status" -> output "status" -> connect to Docker List Containers
- Contains "start" -> output "start" -> connect to Parse Action Command (with action set to start)
- Contains "stop" -> output "stop" -> connect to Parse Action Command (with action set to stop)
- Contains "restart" -> output "restart" -> connect to Parse Action Command (with action set to restart)
- Contains "update" -> output "update" -> connect to existing update flow entry point
- Contains "logs" -> output "logs" -> connect to existing logs flow entry point
- Fallback (extra) -> connect to new "Show Menu" node
Use the Switch node pattern from RESEARCH.md:
```json
{
"conditions": {
"options": { "caseSensitive": false },
"conditions": [{
"leftValue": "={{ $json.message.text }}",
"rightValue": "status",
"operator": { "type": "string", "operation": "contains" }
}]
}
}
```
Update "Route Message" node to connect directly to "Keyword Router" instead of the old Claude flow.
</action>
<verify>
Open workflow in n8n UI. Send "status" - should trigger Docker List Containers path.
Check that no nodes reference api.anthropic.com or Anthropic API Key credential.
</verify>
<done>
Keyword Router handles all 6 commands (status, start, stop, restart, update, logs) plus fallback.
No Claude/NLU nodes remain in workflow.
</done>
</task>
<task type="auto">
<name>Task 2: Add persistent Telegram menu</name>
<files>n8n-workflow.json</files>
<action>
Create a new "Show Menu" node (HTTP Request type, not native Telegram node - per project pattern for keyboards).
Configure HTTP Request:
- URL: `https://api.telegram.org/bot{{ $credentials.telegramApi.token }}/sendMessage`
- Method: POST
- Body (JSON):
```json
{
"chat_id": "={{ $json.message.chat.id }}",
"text": "Use buttons below or type commands:",
"parse_mode": "HTML",
"reply_markup": {
"keyboard": [
[{"text": "Status"}],
[{"text": "Start"}, {"text": "Stop"}],
[{"text": "Restart"}, {"text": "Update"}],
[{"text": "Logs"}]
],
"is_persistent": true,
"resize_keyboard": true,
"one_time_keyboard": false
}
}
```
Connect Keyword Router fallback output to Show Menu node.
NOTE: Research suggests emojis optional (Claude's discretion from CONTEXT.md). Start without emojis for cleaner keyword matching - button text "Status" matches "status" keyword.
Also wire /start command to Show Menu:
- In Keyword Router, add rule: Contains "/start" -> output "menu" -> connect to Show Menu
</action>
<verify>
Send any unknown command (e.g., "hello") - should receive menu with persistent keyboard.
Send "/start" - should receive same menu.
Keyboard should persist after sending other messages.
</verify>
<done>
Persistent menu visible in Telegram chat.
Menu buttons trigger corresponding actions when tapped.
/start and unknown commands show menu.
</done>
</task>
</tasks>
<verification>
1. All Claude/NLU nodes removed (grep for "anthropic", "Claude", "Intent" in workflow JSON)
2. Keyword Router handles: status, start, stop, restart, update, logs, /start, fallback
3. Persistent menu shows 6 buttons in grouped layout
4. Button taps trigger corresponding keyword routes
5. No errors in n8n execution logs for any command
</verification>
<success_criteria>
- Zero references to Anthropic API or Claude in workflow
- All 6 container commands work via typed keywords
- Persistent menu visible and functional
- Unknown input shows menu (not error)
</success_criteria>
<output>
After completion, create `.planning/phases/05-polish-deploy/05-01-SUMMARY.md`
</output>
@@ -0,0 +1,157 @@
---
phase: 05-polish-deploy
plan: 02
type: execute
wave: 2
depends_on: [05-01]
files_modified: [n8n-workflow.json]
autonomous: true
must_haves:
truths:
- "Docker socket errors show 'Cannot connect to Docker' (not stack trace)"
- "Failed actions show 'Failed to X' format (not verbose details)"
- "User ID stored in n8n credentials (not hardcoded in workflow JSON)"
artifacts:
- path: "n8n-workflow.json"
provides: "Terse error messages in response nodes"
contains: "Cannot connect to Docker"
- path: "n8n-workflow.json"
provides: "Credential reference for user auth"
contains: "$credentials"
key_links:
- from: "IF User Authenticated"
to: "n8n credentials"
via: "credential reference expression"
pattern: "\\$credentials"
---
<objective>
Harden error handling with minimal user-facing messages and migrate user ID to n8n credentials system.
Purpose: Production-ready error UX and secure credential storage for workflow sharing.
Output: Terse error messages, credential-based auth, exportable workflow without sensitive data.
</objective>
<execution_context>
@/home/luc/.claude/get-shit-done/workflows/execute-plan.md
@/home/luc/.claude/get-shit-done/templates/summary.md
</execution_context>
<context>
@.planning/PROJECT.md
@.planning/ROADMAP.md
@.planning/STATE.md
@.planning/phases/05-polish-deploy/05-CONTEXT.md
@.planning/phases/05-polish-deploy/05-RESEARCH.md
@.planning/phases/05-polish-deploy/05-01-SUMMARY.md
@n8n-workflow.json
</context>
<tasks>
<task type="auto">
<name>Task 1: Standardize error messages</name>
<files>n8n-workflow.json</files>
<action>
Find all error response nodes and update their message text to follow minimal format:
**Docker/infrastructure errors (detect "docker.sock" or "ECONNREFUSED" in error):**
- Message: "Cannot connect to Docker"
**Action failures:**
- Pattern: "Failed to {action} {container}"
- Examples: "Failed to start plex", "Failed to stop nginx"
**No match errors:**
- Keep existing "No container matching 'X'" format (already terse)
Update these nodes to use terse format:
- Send Docker Error -> "Cannot connect to Docker"
- Send Action Result (error case) -> "Failed to {action} {container}"
- Send Update Error -> "Failed to update {container}"
- Send Logs Error -> "Failed to get logs for {container}"
Do NOT include:
- Stack traces
- HTTP status codes
- Docker API error details
- Technical debugging info
Per CONTEXT.md: "Minimal error messages - 'Failed to start plex' without verbose details"
</action>
<verify>
Grep workflow JSON for error messages - should be terse, no technical details.
Manually test by stopping n8n's Docker socket access and sending command - should see "Cannot connect to Docker".
</verify>
<done>
All user-facing error messages follow terse format.
Infrastructure errors show "Cannot connect to Docker".
Action errors show "Failed to X Y" pattern.
</done>
</task>
<task type="auto">
<name>Task 2: Migrate user ID to n8n credentials</name>
<files>n8n-workflow.json</files>
<action>
Currently the workflow has hardcoded user ID in IF nodes for auth check. Per CONTEXT.md and RESEARCH.md, migrate to n8n credentials system.
**Step 1: Create credential type reference in workflow**
n8n credentials will be created manually by user during deployment, but workflow must reference them.
**Step 2: Update auth check nodes**
Find "IF User Authenticated" and "IF Callback Authenticated" nodes.
Change the condition from hardcoded ID comparison to credential reference:
```
// Old (hardcoded):
$json.message.from.id === 123456789
// New (credential reference):
$json.message.from.id === parseInt($credentials.telegramAuth.userId)
```
Note: The credential name "telegramAuth" with field "userId" follows RESEARCH.md pattern.
User will create this credential during deployment (documented in README from Plan 03).
**Step 3: Clean up sensitive data**
Search workflow JSON for any hardcoded numeric IDs that look like Telegram user IDs (8+ digit numbers).
Remove or replace with credential references.
Per CONTEXT.md: "Sensitive values (Telegram user ID) moved to n8n credentials system - not hardcoded in workflow JSON"
</action>
<verify>
Grep workflow JSON for 8+ digit numbers - should find none (except node position coordinates).
Auth check nodes should reference $credentials.telegramAuth.userId.
</verify>
<done>
No hardcoded user ID in workflow JSON.
Auth nodes use credential reference.
Workflow can be safely exported/shared.
</done>
</task>
</tasks>
<verification>
1. All error messages terse (no stack traces, no verbose details)
2. Docker socket errors produce "Cannot connect to Docker"
3. Action failures produce "Failed to X Y" format
4. No hardcoded Telegram user ID in workflow JSON
5. Auth nodes reference $credentials.telegramAuth.userId
6. Workflow exports cleanly without embedded secrets
</verification>
<success_criteria>
- Error messages fit in single Telegram line (no scrolling needed)
- Zero hardcoded sensitive values in workflow JSON
- grep -E '[0-9]{9,}' workflow.json returns only position coordinates
</success_criteria>
<output>
After completion, create `.planning/phases/05-polish-deploy/05-02-SUMMARY.md`
</output>
@@ -0,0 +1,203 @@
---
phase: 05-polish-deploy
plan: 03
type: execute
wave: 3
depends_on: [05-02]
files_modified: [README.md, n8n-workflow.json]
autonomous: false
must_haves:
truths:
- "README has step-by-step deployment instructions"
- "README documents credential creation in n8n"
- "README documents Docker socket setup for n8n container"
- "All 6 commands tested end-to-end via Telegram"
artifacts:
- path: "README.md"
provides: "Deployment guide"
min_lines: 50
contains: "Installation"
- path: "n8n-workflow.json"
provides: "Production-ready workflow"
key_links:
- from: "README.md"
to: "n8n credentials"
via: "documentation"
pattern: "telegramAuth"
---
<objective>
Write deployment README and perform end-to-end testing of complete bot functionality.
Purpose: Enable users to deploy the bot on their own Unraid servers with clear instructions.
Output: Complete README, verified workflow, production-ready deployment package.
</objective>
<execution_context>
@/home/luc/.claude/get-shit-done/workflows/execute-plan.md
@/home/luc/.claude/get-shit-done/templates/summary.md
</execution_context>
<context>
@.planning/PROJECT.md
@.planning/ROADMAP.md
@.planning/STATE.md
@.planning/phases/05-polish-deploy/05-CONTEXT.md
@.planning/phases/05-polish-deploy/05-RESEARCH.md
@.planning/phases/05-polish-deploy/05-01-SUMMARY.md
@.planning/phases/05-polish-deploy/05-02-SUMMARY.md
@n8n-workflow.json
</context>
<tasks>
<task type="auto">
<name>Task 1: Write deployment README</name>
<files>README.md</files>
<action>
Replace the stub README with a complete deployment guide. Per CONTEXT.md: "README only - step-by-step instructions in markdown" and "No troubleshooting section - focused on initial setup only".
Structure (following RESEARCH.md template):
# Docker Manager Bot
One-line description: Telegram bot for managing Docker containers on Unraid.
## Prerequisites
- Unraid server with Docker enabled
- n8n container running on Unraid
- Telegram Bot Token (from @BotFather)
- Your Telegram User ID (from @userinfobot)
## Installation
### 1. Configure n8n Container
Document the Docker run flags needed:
```bash
docker run -d \
--name n8n \
--group-add 281 \
-v /var/run/docker.sock:/var/run/docker.sock \
-v /path/to/curl:/usr/bin/curl:ro \
n8nio/n8n
```
Explain:
- `--group-add 281` for Docker socket access
- Socket mount requirement
- Static curl binary mount (hardened n8n image)
### 2. Create n8n Credentials
**Telegram API credential:**
- Type: Telegram API
- Access Token: your bot token from @BotFather
**Telegram Auth credential:**
- Type: Header Auth (or custom)
- Field: userId = your Telegram user ID
### 3. Import Workflow
- Copy n8n-workflow.json to server
- In n8n: Workflows -> Import from File
- Map credentials when prompted
### 4. Activate Workflow
- Open workflow
- Click Active toggle
- Test with "status" message
## Usage
List the 6 commands:
- status - View all containers
- start <name> - Start container
- stop <name> - Stop container
- restart <name> - Restart container
- update <name> - Pull and recreate container
- logs <name> [lines] - View container logs
Mention persistent menu buttons available.
Per CONTEXT.md: No troubleshooting section.
</action>
<verify>
README exists at root.
Contains all 4 installation sections.
No troubleshooting section.
Markdown renders correctly.
</verify>
<done>
Complete deployment guide in README.md.
Step-by-step instructions for fresh Unraid installation.
</done>
</task>
<task type="checkpoint:human-verify" gate="blocking">
<name>Task 2: End-to-end testing</name>
<what-built>
Complete Docker Manager Bot:
- Keyword routing (no Claude dependency)
- Persistent Telegram menu
- All 6 container commands
- Terse error messages
- Credential-based auth
</what-built>
<how-to-verify>
Test each command via Telegram bot:
1. **Menu:** Send /start or any unknown text
- Expected: Persistent keyboard appears with 6 buttons
2. **Status:** Tap Status button or type "status"
- Expected: List of containers with status indicators
3. **Start:** Type "start <stopped-container-name>"
- Expected: Container starts, confirmation message
4. **Stop:** Type "stop <running-container-name>"
- Expected: Container stops, confirmation message
5. **Restart:** Type "restart <container-name>"
- Expected: Container restarts, confirmation message
6. **Update:** Type "update <container-name>"
- Expected: Image pulled, container recreated (or silent if no update)
7. **Logs:** Type "logs <container-name>"
- Expected: Last 50 log lines displayed
8. **Error handling:** Stop n8n's Docker socket access briefly
- Expected: "Cannot connect to Docker" (not stack trace)
9. **Auth:** Message bot from different Telegram account
- Expected: No response (silent ignore per Phase 1 decision)
</how-to-verify>
<resume-signal>Type "approved" if all tests pass, or describe which tests failed</resume-signal>
</task>
</tasks>
<verification>
1. README exists and has all required sections
2. All 6 commands work end-to-end
3. Persistent menu functions correctly
4. Error messages are terse
5. Unauthorized users get no response
6. Workflow exports without sensitive data
</verification>
<success_criteria>
- Fresh user can follow README to deploy bot
- All container management commands functional
- Bot ready for production use on Unraid
</success_criteria>
<output>
After completion, create `.planning/phases/05-polish-deploy/05-03-SUMMARY.md`
</output>