92 lines
3.0 KiB
Markdown
92 lines
3.0 KiB
Markdown
---
|
|
id: 45c21973-7fae-4cf7-b247-b40a757d54b0
|
|
type: decision
|
|
title: "Issue Worker Agent: Phased Workflow Architecture"
|
|
tags: [claude-configs, issue-worker, agent-design, gitea, automation, workflow, cognitive-memory]
|
|
importance: 0.7
|
|
confidence: 0.8
|
|
created: "2026-03-02T20:19:30.456720+00:00"
|
|
updated: "2026-03-02T20:20:08.207657+00:00"
|
|
relations:
|
|
- target: b27fb831-811e-4103-b5f1-3c36d5b1ad20
|
|
type: RELATED_TO
|
|
direction: outgoing
|
|
strength: 0.9
|
|
edge_id: a1813354-3f35-46b8-9689-9d259413bf6e
|
|
- target: 73646e80-a640-4316-b7c6-54a4330f1b28
|
|
type: RELATED_TO
|
|
direction: outgoing
|
|
strength: 0.87
|
|
edge_id: c078c5fc-c046-411b-8807-d52f1a4c399f
|
|
- target: 1052e91d-58ed-4308-87e8-e01d1143a146
|
|
type: RELATED_TO
|
|
direction: outgoing
|
|
strength: 0.77
|
|
edge_id: 21f44795-4134-4ebf-b242-0d233ab82a17
|
|
- target: 4321bee8-105e-4fc8-b645-964d1234c966
|
|
type: BUILDS_ON
|
|
direction: outgoing
|
|
strength: 0.95
|
|
edge_id: 0924ac34-ddaa-4eba-ad0f-743077be7354
|
|
- target: c4b2f42d-b443-447f-8a05-bba099e06036
|
|
type: RELATED_TO
|
|
direction: outgoing
|
|
strength: 0.9
|
|
edge_id: 6b20858b-0402-4dfb-ae93-6cf4a0afb86d
|
|
- target: d326094f-63eb-4b42-8aa6-f2d53d6f2628
|
|
type: REQUIRES
|
|
direction: outgoing
|
|
strength: 0.85
|
|
edge_id: 89529f58-d709-46e1-bbdd-c0636ee41b21
|
|
- target: 6ce2055a-b51b-45e4-8ff0-d34b385f7552
|
|
type: RELATED_TO
|
|
direction: outgoing
|
|
strength: 0.7
|
|
edge_id: 6f390811-ecf9-4a2a-a28c-849d585fe9a6
|
|
---
|
|
|
|
# Issue Worker Agent: Phased Workflow Architecture
|
|
|
|
## Context
|
|
Rewrote the issue-worker agent at `~/.claude/agents/issue-worker.md` from a flat 10-step workflow into 4 distinct phases. Config synced to Gitea at commit `a62f16d`.
|
|
|
|
## Architecture
|
|
|
|
### Phase 1: Understand
|
|
- Read issue from Gitea
|
|
- Explicitly read `CLAUDE.md` before any code work
|
|
- Assess feasibility
|
|
- Add `status/in-progress` label to issue
|
|
|
|
### Phase 2: Implement
|
|
- Explore relevant code
|
|
- Create branch using `-B` flag for retry safety
|
|
- Implement fix
|
|
- Run tests with auto-discovery: `CLAUDE.md` → `Makefile` → `pyproject.toml` → `package.json` → `Cargo.toml`
|
|
|
|
### Phase 3: Review & Ship
|
|
- Self-review via `git diff` before committing (avoids needing a second cleanup commit)
|
|
- Commit and push
|
|
- Create PR via `gitea-mcp`
|
|
- Swap labels: remove `status/in-progress`, add `status/pr-open`
|
|
- Comment on original issue with PR link
|
|
|
|
### Phase 4: Remember
|
|
- Store cognitive memory of fix with proper tags
|
|
- Create graph edges to related memories using `memory_relate`
|
|
|
|
## Key Decisions
|
|
|
|
### Branch Naming
|
|
- Changed from `ai/<repo>#<issue_number>` to `ai/<repo>-<issue_number>`
|
|
- Reason: `#` causes shell interpretation issues and URL encoding problems
|
|
|
|
### MCP Tools Added
|
|
- `edit_issue` — for swapping labels during phase transitions
|
|
- `memory_search` — for finding related existing memories in Phase 4
|
|
- `memory_relate` — for creating graph edges in Phase 4
|
|
|
|
### Self-Review Placement
|
|
- `git diff` review placed before commit, not after
|
|
- Avoids needing a follow-up fixup commit for issues caught post-commit
|