flowmade-one/architecture
3
0
Fork 0
Code Issues 15 Pull Requests 1 Actions Packages Projects Releases Wiki Activity

Refactor spawn-issues as orchestrator

Browse Source 
spawn-issues now orchestrates the full workflow:
- Phase 1: Spawn issue-workers in parallel, wait for completion
- Phase 2: Review loop - spawn code-reviewer, if needs work spawn pr-fixer
- Phase 3: Report final status

issue-worker simplified:
- Removed Task tool and review loop
- Just implements, creates PR, cleans up
- Returns structured result for orchestrator to parse

Benefits:
- Better visibility into progress
- Reuses pr-fixer agent
- Clean separation of concerns
- Orchestrator controls review cycle

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
Hugo Nijhuis
2026-01-09 17:33:22 +01:00
parent 7dffdc4e77
commit 3afe930a27
2 changed files with 200 additions and 104 deletions
Download Patch File Download Diff File Expand all files Collapse all files

47
agents/issue-worker/agent.md
View File

@@ -1,13 +1,13 @@
---
name: issue-worker
description: Autonomous agent that works on a single issue in an isolated git worktree
tools: Bash, Read, Write, Edit, Glob, Grep, TodoWrite, Task
description: Autonomous agent that implements a single issue in an isolated git worktree
tools: Bash, Read, Write, Edit, Glob, Grep, TodoWrite
skills: gitea, issue-writing
---
# Issue Worker Agent
Autonomously implements a single issue in an isolated git worktree.
Autonomously implements a single issue in an isolated git worktree. Creates a PR and returns - the orchestrator handles review.
## Input
@@ -86,26 +86,7 @@ Closes #<ISSUE_NUMBER>"
Capture the PR number from the output (e.g., "Pull Request #42 created").
### 6. Review Loop
Spawn the `code-reviewer` agent **synchronously** to review the PR in the current worktree:
```
Task tool with:
- subagent_type: "code-reviewer"
- run_in_background: false
- prompt: "Review PR #<PR_NUMBER>. Working directory: <WORKTREE_PATH>"
```
Based on review feedback:
- **If approved**: Proceed to cleanup
- **If needs work**:
1. Address the review feedback
2. Commit and push the fixes
3. Trigger another review
4. Repeat until approved (max 3 iterations to avoid infinite loops)
### 7. Cleanup Worktree
### 6. Cleanup Worktree
Always clean up, even if earlier steps failed:
@@ -114,20 +95,21 @@ cd <REPO_PATH>
git worktree remove ../<REPO_NAME>-issue-<ISSUE_NUMBER> --force
```
### 8. Final Summary
### 7. Final Summary
**IMPORTANT**: Your final output must be a concise summary (5-10 lines max) for the spawning process:
**IMPORTANT**: Your final output must be a concise summary for the orchestrator:
```
Issue #<NUMBER>: <title>
Status: <completed|partial|blocked>
PR: #<PR_NUMBER> (<url>)
Changes: <1-2 sentence summary>
Review: <approved|needs-work|skipped>
Notes: <any blockers or important details>
ISSUE_WORKER_RESULT
issue: <ISSUE_NUMBER>
pr: <PR_NUMBER>
branch: <branch-name>
status: <success|partial|failed>
title: <issue title>
summary: <1-2 sentence description of changes>
```
Do NOT include verbose logs or intermediate output - only this final summary.
This format is parsed by the orchestrator. Do NOT include verbose logs - only this summary.
## Important Guidelines
@@ -144,3 +126,4 @@ If you encounter an error:
1. Try to recover if possible
2. If unrecoverable, create a PR with partial work and explain the blocker
3. Always run the cleanup step
4. Report status as "partial" or "failed" in summary

257
commands/spawn-issues.md
View File

@@ -1,11 +1,12 @@
---
allowed-tools: Bash, Task, Read
description: Spawn parallel background agents to work on multiple issues simultaneously
allowed-tools: Bash, Task, Read, TaskOutput
description: Orchestrate parallel issue implementation with review cycles
argument-hint: <issue-number> [<issue-number>...]
---
# Spawn Issues
# Spawn Issues (Orchestrator)
Spawn background agents to work on multiple issues in parallel. Each agent works in an isolated git worktree.
Orchestrate parallel issue implementation: spawn workers, review PRs, fix feedback, until all approved.
## Arguments
@@ -13,13 +14,29 @@ One or more issue numbers separated by spaces: `$ARGUMENTS`
Example: `/spawn-issues 42 43 44`
## Process
## Orchestration Flow
1. **Validate arguments**: Ensure at least one issue number is provided
2. **Get repo info**: Determine repository path and name
3. **Fetch issue titles**: Get title for each issue (for display)
4. **Spawn agents**: For each issue, spawn a background `issue-worker` agent
5. **Report**: Display summary table with agent IDs
```
┌─────────────────────────────────────────────────────────────┐
│ spawn-issues │
│ (orchestrator) │
├─────────────────────────────────────────────────────────────┤
│ Phase 1: Implementation │
│ ├── Spawn issue-worker agents (background, parallel) │
│ ├── Wait for all workers to complete │
│ └── Collect PR numbers from results │
├─────────────────────────────────────────────────────────────┤
│ Phase 2: Review Loop (for each PR) │
│ ├── Spawn code-reviewer (background) │
│ ├── Wait for review │
│ ├── If needs work → spawn pr-fixer (background) │
│ ├── Wait for fix │
│ └── Repeat until approved (max 3 iterations per PR) │
├─────────────────────────────────────────────────────────────┤
│ Phase 3: Report │
│ └── Display final status table │
└─────────────────────────────────────────────────────────────┘
```
## Implementation
@@ -38,93 +55,189 @@ REPO_PATH=$(pwd)
REPO_NAME=$(basename $REPO_PATH)
```
### Step 3: For Each Issue
### Step 3: Phase 1 - Spawn Issue Workers
For each issue number in the arguments:
1. Fetch issue title using `tea issues <number>` (just to verify it exists and get the title)
2. Spawn background agent using Task tool:
```
Task tool with:
- subagent_type: "general-purpose"
- run_in_background: true
- prompt: See agent prompt below
```
### Agent Prompt
For each issue, use this prompt:
For each issue number, spawn a background issue-worker agent:
```
You are an issue-worker agent. Work on issue #<NUMBER> autonomously.
Task tool with:
- subagent_type: "general-purpose"
- run_in_background: true
- prompt: <issue-worker prompt below>
```
**Issue Worker Prompt:**
```
You are an issue-worker agent. Implement issue #<NUMBER> autonomously.
Context:
- Repository path: <REPO_PATH>
- Repository name: <REPO_NAME>
- Issue number: <NUMBER>
Instructions from @agents/issue-worker/agent.md:
Process:
1. Setup worktree:
cd <REPO_PATH>
git fetch origin
cd <REPO_PATH> && git fetch origin
git worktree add ../<REPO_NAME>-issue-<NUMBER> -b issue-<NUMBER>-<short-title> origin/main
cd ../<REPO_NAME>-issue-<NUMBER>
2. Get issue details:
tea issues <NUMBER> --comments
2. Get issue: tea issues <NUMBER> --comments
3. Plan with TodoWrite, then implement the changes
3. Plan with TodoWrite, implement the changes
4. Commit:
git add -A && git commit with message "...\n\nCloses #<NUMBER>\n\nCo-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>"
4. Commit: git add -A && git commit -m "...\n\nCloses #<NUMBER>\n\nCo-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>"
5. Push:
git push -u origin <branch-name>
5. Push: git push -u origin <branch-name>
6. Create PR:
tea pulls create --title "[Issue #<NUMBER>] <title>" --description "Closes #<NUMBER>\n\n<summary of changes>"
Capture the PR number from the output.
6. Create PR: tea pulls create --title "[Issue #<NUMBER>] <title>" --description "Closes #<NUMBER>\n\n..."
Capture the PR number.
7. Review loop:
Spawn code-reviewer agent SYNCHRONOUSLY (run_in_background: false) to review the PR:
Task tool with subagent_type: "code-reviewer", prompt: "Review PR #<PR_NUMBER>. Working directory: <WORKTREE_PATH>"
7. Cleanup: cd <REPO_PATH> && git worktree remove ../<REPO_NAME>-issue-<NUMBER> --force
If review says "needs work":
- Address the feedback
- Commit and push fixes
- Re-run the review
- Repeat until approved (max 3 iterations)
8. Output EXACTLY this format (orchestrator parses it):
ISSUE_WORKER_RESULT
issue: <NUMBER>
pr: <PR_NUMBER>
branch: <branch-name>
status: <success|partial|failed>
title: <issue title>
summary: <1-2 sentence description>
8. Cleanup (ALWAYS do this, even if earlier steps failed):
cd <REPO_PATH> && git worktree remove ../<REPO_NAME>-issue-<NUMBER> --force
9. Final output - ONLY output this concise summary (5-10 lines max):
Issue #<NUMBER>: <title>
Status: <completed|partial|blocked>
PR: #<PR_NUMBER>
Changes: <1-2 sentence summary>
Review: <approved|needs-work|skipped>
Notes: <any blockers or important details>
Work autonomously. Make judgment calls on ambiguous requirements. If blocked, note it in the PR description.
Work autonomously. If blocked, note it in PR description and report status as partial/failed.
```
### Step 4: Report
Display progress:
```
Phase 1: Implementation
Spawning issue-worker agents...
- Issue #42: spawned (task_id: xxx)
- Issue #43: spawned (task_id: xxx)
- Issue #44: spawned (task_id: xxx)
```
After spawning all agents, display:
### Step 4: Wait for Workers
Use TaskOutput to wait for each worker to complete. Parse the ISSUE_WORKER_RESULT from each output to extract:
- issue number
- pr number
- status
Build a tracking table:
```
| Issue | PR | Status |
|-------|-----|---------|
| #42 | #55 | success |
| #43 | #56 | success |
| #44 | - | failed |
```
### Step 5: Phase 2 - Review Loop
For each PR that was created successfully, run a review loop (max 3 iterations):
```
Spawned <N> issue-worker agents:
for each PR:
iteration = 0
while iteration < 3:
# Spawn code-reviewer in background
Task tool with:
- subagent_type: "code-reviewer"
- run_in_background: true
- prompt: "Review PR #<PR_NUMBER> in <REPO_PATH>. Output EXACTLY:
REVIEW_RESULT
pr: <PR_NUMBER>
verdict: <approved|needs-work>
summary: <1-2 sentences>"
| Issue | Title | Status |
|-------|--------------------------|------------|
| #42 | Add /commit command | spawned |
| #43 | Add /pr command | spawned |
| #44 | Add CI status | spawned |
# Wait for review
Use TaskOutput to get result, parse verdict
Agents working in background. Monitor with:
- Check PR list: tea pulls
- Check worktrees: git worktree list
if verdict == "approved":
mark PR as approved, break
if verdict == "needs-work":
# Spawn pr-fixer in background
Task tool with:
- subagent_type: "general-purpose"
- run_in_background: true
- prompt: <pr-fixer prompt - address feedback, commit, push>
# Wait for fix
Use TaskOutput to get result
iteration++
```
Display progress:
```
Phase 2: Review
PR #55: reviewing...
PR #55: needs work → fixing...
PR #55: re-reviewing...
PR #55: approved ✓
PR #56: reviewing...
PR #56: approved ✓
```
### Step 6: Phase 3 - Final Report
Display final status:
```
Completed: 3 issues processed
| Issue | PR | Implementation | Review |
|-------|-----|----------------|----------|
| #42 | #55 | success | approved |
| #43 | #56 | success | approved |
| #44 | - | failed | - |
Summary:
- 2 PRs created and approved
- 1 issue failed (see PR description for details)
```
## PR Fixer Prompt
When spawning pr-fixer for a PR that needs work:
```
You are a pr-fixer agent. Address review feedback on PR #<NUMBER>.
Context:
- Repository path: <REPO_PATH>
- Repository name: <REPO_NAME>
- PR number: <NUMBER>
Process:
1. Get feedback: tea pulls <NUMBER> --comments
2. Setup worktree from PR branch:
cd <REPO_PATH> && git fetch origin
git worktree add ../<REPO_NAME>-pr-<NUMBER> origin/<branch-name>
cd ../<REPO_NAME>-pr-<NUMBER>
git checkout <branch-name>
3. Address each piece of feedback
4. Commit and push:
git add -A && git commit -m "Address review feedback\n\nCo-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>"
git push
5. Cleanup: cd <REPO_PATH> && git worktree remove ../<REPO_NAME>-pr-<NUMBER> --force
6. Output EXACTLY:
PR_FIXER_RESULT
pr: <NUMBER>
status: <fixed|partial|failed>
changes: <summary of fixes>
Work autonomously. If feedback is unclear, make reasonable judgment calls.
```
## Error Handling
- If an issue-worker fails, continue with others
- If a review fails, mark as "review-failed" and continue
- If pr-fixer fails after 3 iterations, mark as "needs-manual-review"
- Always report final status even if some items failed