rheinmir Orca Dispatch Reference
Reference doc, not a skill. Do not invoke in loop.
Nguồn chân lý DUY NHẤT cho dispatch. Mọi skill orca-* / orchestration (orca-workflow, orca-onboard, orca-cli, orchestration, council…) chỉ TRỎ về đây — đừng nhân bản roster/syntax ở nơi khác (chống drift).
Agent backends — roster & chọn theo cost (verified 2026-07-01)
Orca drive model bằng cách chạy một agent CLI trong terminal/worktree. Picker UI: OpenCode · Claude · GitHub Copilot · Antigravity · Kiro. Ba hệ tên hay lệch (picker ≠ @-handle ≠ --agent id) → bảng này hoà giải:
| Picker (UI) | --agent <id> |
provider/model | Cost | Dùng cho |
|---|---|---|---|---|
| OpenCode | opencode |
opencode/big-pickle (+free khác) |
$0 | task rẻ, render, seat answer-only |
| Claude | claude |
(phiên hiện tại) | đắt nhất | reasoning khó, chairman — để dành |
| Antigravity | agy |
— | rẻ | render dự phòng |
| GitHub Copilot | copilot |
— | rẻ | dự phòng |
| Kiro | kiro |
— | rẻ | dự phòng |
--agent <id>bịa →"Unknown TUI agent"(validate TRƯỚC khi tạo worktree). Tập recognized cho--inject(live error):claude · codex · gemini · droid. Không có lệnh liệt kê — picker UI là ground truth.- Cost-tier (luật): rẻ / cơ học / answer-only → opencode free; reasoning đắt / tổng hợp / chairman → Claude (đắt nhất, để dành). Khớp split "Claude-nghĩ / CLI-rẻ-render" của
orca-workflow.
OpenCode — cheap dispatch (verified)
Free models ($0, opencode zen): opencode/big-pickle · deepseek-v4-flash-free · mimo-v2.5-free · nemotron-3-ultra-free · north-mini-code-free.
opencode run -m opencode/big-pickle --dir "<worktree-path>" "<task>" # headless, $0
opencode run -m opencode/big-pickle --format json "<task>" # JSON events (parse máy)
opencode run -m <model> -s <session_id> "<task>" # nối session (--continue / --fork)
# ⚠ KHÔNG --dangerously-skip-permissions khi dispatch TỪ Claude Code (classifier DENY — bài học 120626).
# Task answer-only (không đụng file) thì KHÔNG cần quyền → chạy thẳng được.
⚠ Reliability (verified 2026-07-01): big-pickle ngon với câu ngắn/answer-only (toán → vài giây); STALL/timeout trên task NẶNG (đọc nhiều file / sinh dài — đo >560s vẫn chưa xong). Đặt watchdog 60–90s → im lặng thì kill + fallback Claude. Task có file-edit/dependency → đừng giao opencode.
Concurrency — KHÔNG mở nhiều opencode trong CÙNG 1 folder (researched + verified)
Nhiều opencode cùng 1 folder dùng chung SQLite/session (resolve project_id theo working dir) → đè nhau, hỏng git snapshot (opencode issues #31307 / #4251 / #28249). Test "3 song song 1 folder OK" là MISLEADING — chỉ đúng với câu tí hon answer-only.
Pattern ĐÚNG (đã validate): mỗi worker = 1 worktree riêng, 1 opencode/worktree:
orca worktree create --name <w> --setup skip --no-parent --json # tách dir+branch
opencode run --dir "<worktree-path>" -m opencode/big-pickle "<task>" & # N cái song song được
orca worktree rm --worktree name:<w> --force --json # dọn
Hoặc multiplexer: opencode serve --port <P> + opencode run --attach http://localhost:<P> -s <id>.
Orchestration dispatch (qua orca runtime — plumbing đã verify live)
orca orchestration task-create --spec "<task>" --task-title "<t>" --json # → task_id
orca orchestration dispatch --task <id> --to <handle> --inject --json # --inject CẦN agent CLI đang chạy trong terminal đích
orca orchestration check --terminal <handle> --types worker_done --wait --timeout-ms 300000 --json
- Spawn agent vào terminal trước khi
--inject:orca worktree create --agent <id> --prompt "<task>"hoặcorca terminal create --command "<cli>". - Group address:
@all @idle @claude @codex @opencode @gemini @droid @worktree:<id>.
Antigravity (agy)
Binary: agy — %LOCALAPPDATA%\agy\bin\agy.exe. NOT antigravity, NOT ~/.local/bin/agy.
# CHECK trước khi dùng:
agy --version
Hook: Orca v1.4.21 fix Windows hook quoting — no manual antigravity-hook.cmd edit.
Dispatch status (post v1.4.21):
| Method | Status |
|---|---|
dispatch --inject |
retest |
terminal send |
OK |
| Tool calls | OK |
worker_done callback |
retest |
Fallback if --inject fails:
orca terminal send --title "Antigravity" --text "<task>"
orca terminal wait --for tui-idle
orca terminal read --title "Antigravity"
Skill Installation per Agent CLI
Skills: llmwiki/skills/<category>/<name>.md.
Claude Code
# CHECK:
ls .claude/commands/
# Install:
cp llmwiki/skills/dev-loop/propose.md .claude/commands/propose.md
OpenCode / Antigravity
# CHECK:
ls ~/.agents/skills/
# Install (skill = folder with SKILL.md):
mkdir -p ~/.agents/skills/propose/
cp llmwiki/skills/dev-loop/propose.md ~/.agents/skills/propose/SKILL.md
# Restart OpenCode after install.
AgentMemory
BASE="https://agentmemory.giatbh.io.vn"
TOKEN="${AGENTMEMORY_TOKEN}"
# CHECK health:
curl -sk -H "Authorization: Bearer $TOKEN" "$BASE/agentmemory/health"
# Ghi:
curl -sk -X POST "$BASE/agentmemory/remember" \
-H "Authorization: Bearer $TOKEN" -H "Content-Type: application/json" \
-d '{"content":"<text>","category":"fact|preference|decision|context"}'
# Tìm:
curl -sk -H "Authorization: Bearer $TOKEN" "$BASE/agentmemory/search?query=<keyword>"
RTK — Token Proxy
RTK (Rust Token Killer): auto-filter CLI output before context — 60-90% token reduction.
# CHECK:
rtk --version
# Install hook (once):
rtk hooks install --agent claude
rtk hooks install --agent opencode
# All agent commands auto-route through RTK.
# Savings:
rtk stats --today
Xem [[concepts/RTK]] để biết chi tiết.
Output Report
After all main skill tasks complete, write a propose draft to the wiki.
Steps
1. Build the filename:
- Format:
DDMMYY-<ten>.md DDMMYY= today (e.g.,020626for 2 June 2026)<ten>= 2–4 kebab-case words summarising what was done (e.g.,landing-page-coteccons,brand-kit-fintech,ingest-auth-spec)
2. Write llmwiki/wiki/draft/orca/DDMMYY-<ten>.md:
# DDMMYY-<ten>
**Type:** draft
**Status:** proposed
**Tags:** <skill-name>, output-report
**Proposed:** YYYY-MM-DD
## Agent Task Assignment
| Task | Agent | Status |
|------|-------|--------|
| <mô tả task 1> | <tên agent> | pending / in-progress / done |
| <mô tả task 2> | <tên agent> | pending / in-progress / done |
## What
<One sentence — what this skill invocation produced or decided>
## Output
<Key artefacts, files created/modified, or decisions made>
## Files
| File | Action |
|------|--------|
| `path/to/file` | created / modified |
## Notes
- Invoked via: `/<skill-name>` skill
## Origin
- **Draft:** `wiki/draft/orca/DDMMYY-<ten>.md`
- **Commit:** _(filled by verify-before-commit)_
- **Date promoted:** _(filled by verify-before-commit)_
3. Update wiki index & log:
llmwiki/wiki/index.md— append one row:| [DDMMYY-<ten>](draft/orca/DDMMYY-<ten>.md) | draft | YYYY-MM-DD |llmwiki/wiki/log.md— append:## YYYY-MM-DD — <skill-name> — <ten>
4. Update agent statuses & sync push — BẮT BUỘC, không bỏ qua:
- Mở lại file
llmwiki/wiki/draft/orca/DDMMYY-<ten>.md - Cập nhật cột Status trong bảng
## Agent Task Assignmenttheo trạng thái thực tế của từng agent (pending → in-progress → done) - Clone
rheinmir/setupnhánhorca, copy các skill file đã sửa, rồi push ngược lên:git clone git@github.com:rheinmir/setup.git /tmp/rheinmir-setup-sync -b orca --depth 1 cp /path/to/skill.md /tmp/rheinmir-setup-sync/skills/<skill-name>/SKILL.md cd /tmp/rheinmir-setup-sync git add . git commit -m "skill: sync update — DDMMYY-<ten>" git push origin orca rm -rf /tmp/rheinmir-setup-sync
Skip chỉ khi skill không tạo ra artifact hoặc quyết định nào.