メニュー
Deterministic multi-agent task assignment via GitHub labels — classify, label, and generate queue views from `agent:` labels instead of manual queue files
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/vamseeachanta/workspace-hub --skill agent-label-routingコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
SOC 職業分類に基づく
Write outbound email and external messages in Vamsee Achanta's voice — a subtle offer to help, never bold or rash claims. Load before drafting ANY email, LinkedIn/Collide reply, proposal note, or outreach sent under his name.
Save/publish analysis or computation results from ANY ecosystem repo to Hugging Face as a queryable, viewer-renderable dataset. Use when the user wants to "save results to hugging face", "publish dataset to HF", "hugging face data saving", "save analysis results", "hf dataset", "make results queryable", or "render via datasets-server API". Reshapes nested results into flat parquet tables, writes a dataset card with a viewer `configs:` block and provenance, applies license/public-vs-private routing, enforces a domain data-quality gate (faithful-to-source != correct), publishes to `aceengineer/<repo>-<projection>`, and verifies via the datasets-server API.
Clone, create, fork, configure, and manage GitHub repositories. Manage remotes, secrets, releases, and workflows. Works with gh CLI or falls back to git + GitHub REST API via curl.
SKILL.md を表示中
| name | agent-label-routing |
| description | Deterministic multi-agent task assignment via GitHub labels — classify, label, and generate queue views from `agent:` labels instead of manual queue files |
| version | 1.0.0 |
| category | coordination |
| type | skill |
| trigger | manual |
| auto_execute | false |
| capabilities | ["agent_classification","batch_labeling","queue_generation","overnight_planning"] |
| tools | ["terminal","gh"] |
| scripts | ["refresh-agent-work-queue.sh"] |
Deterministic agent task assignment using GitHub agent: labels. Labels ARE the queue — separate markdown files are just filtered views.
| Label | Color | Role |
|---|---|---|
agent:gemini | F5A623 | Research, prep, large-doc ingestion, standards mapping |
agent:claude | DA552F | Heavy coding, architecture, orchestration, complex TDD |
agent:codex | 3182CE | Bounded implementation, test writing, review, refactoring |
agent:any | 8E54E9 | No strong preference — whichever agent has capacity |
gh label create "agent:gemini" --color "F5A623" --description "Research, prep, large-doc ingestion"
gh label create "agent:claude" --color "DA552F" --description "Heavy coding, architecture, orchestration"
gh label create "agent:codex" --color "3182CE" --description "Bounded implementation, tests, review"
gh label create "agent:any" --color "8E54E9" --description "No strong preference"
Use rule-based scoring with keyword matching + category boosts:
# GEMINI signals (+2 each)
keywords: literature, review, research, catalog, triage, summarize,
standards gap, standards mapping, migrate literature, acquire, index,
scrape, dedup, job market
category boost: cat:document-intelligence (+3), cat:data-pipeline + dark-intelligence (+2)
# CLAUDE signals (+2 each)
keywords: architecture, orchestration, governance, credit utilization,
work queue, model switching, dispatch, integration, concept selection,
capex/opex, facility sizing, production profile, floating platform,
stability, gyradius, trim ballast, epic
category boost: cat:ai-orchestration (+), cat:engineering + dark-intelligence (+)
keywords: test coverage, bug fix, solver queue, extract metadata,
batch-at-stop, config-protection hook, convert agent to skill,
deduplicate skill, provider config
category boost: cat:engineering-calculations + domain:code-promotion (+)
Never use sequential gh issue edit — it is SLOW and hits rate limits.
Parallel background shell (recommended):
GEMINI=(1863 1862 1860 ... 68 issues)
for n in "${GEMINI[@]}"; do
gh issue edit "$n" --add-label "agent:gemini" 2>/dev/null &
done
wait # blocks until all complete, then move to next agent
175 issues labeled in ~15s vs 300s+ sequential.
Alternative: one-liner per batch
Same approach — background all gh issue edit calls for one agent, wait, then next.
# Query live from GitHub — never manually maintain
gh issue list -L 100 --label "agent:gemini,priority:high" --json number,title
gh issue list -L 100 --label "agent:claude,priority:high" --json number,title
gh issue list -L 100 --label "agent:codex,priority:high" --json number,title
scripts/refresh-agent-work-queue.sh regenerates notes/agent-work-queue.md from live label queries. Run weekly on Sunday via cron.
# Move issue from Gemini to Codex
gh issue edit 1234 --remove-label "agent:gemini" --add-label "agent:codex"
When a routed agent cannot complete the assigned lane after work has started, preserve traceability and reroute instead of abandoning or silently switching providers.
Use references/provider-reroute-traceability.md for the detailed pattern. Minimum steps:
agent: routing label (--remove-label agent:<old> --add-label agent:<new>); do not mutate lifecycle labels unless the plan gate actually changed..planning/quick/.After labeling, execute research tasks in batches of 5-6 per Gemini session:
h-router-gemini -t terminal,file -q "You are ACE Engineer advance scout.
Working directory: /mnt/local-analysis/workspace-hub.
Execute ALL 5 tasks. Commit after each. Do NOT push. Close each issue.
TASK 1: <issue-title> (#number)
- Use search_files or terminal to gather data
- Create: <output path>
- Commit: git add <file> && git commit -m '<msg>'
- Close: gh issue close <number> -c '<close comment>'
TASK 2-5: same pattern...
"
-t terminal,file for filesystem writes, add web for web searchgh issue list --json outputs a JSON array, not newline-delimited JSON. Use --jq for clean output.gh issue edit hits API rate limits and is very slow. Always use background parallelism for bulk operations.gh CLI may return exit 1 on error even for individual failures in parallel. Use 2>/dev/null per call and wait to sync.ls -la since sandbox isolation can sometimes prevent writes from persisting.h-nemotron, h-qwen) timeout after 5 min — too short for 5+ task batches. Gemini via OpenRouter takes ~2 min per session of 5 tasks.When the user asks to review related GitHub issues and prepare a board for a flow such as data layer → execution layer → result/output layer, do not treat agent: labels as the only source of truth. Build a dependency-aware Kanban using issue metadata, parent/child links, plan state, and GitHub Project fields.
Use references/layered-kanban-flow-routing.md for the detailed pattern. Key guardrails:
docs/reports/YYYY-MM-DD-<topic>-kanban.md artifact when the work is coordination/planning.status:plan-approved, close issues, or mutate lifecycle labels just to populate the board.agent: labels applied, when label routing is explicitly part of the tasknotes/agent-work-queue.md (auto-generated view)docs/reports/YYYY-MM-DD-<topic>-kanban.md (layered Kanban flow report)docs/plans/overnight-prompts-YYYY-MM-DD.md (weekly overnight plan)scripts/refresh-agent-work-queue.sh (cron-ready refresh script)6 live Gemini sessions + 4 cron batches = 30+ research documents, 14 issues closed, ~12 min total compute, ~$0 from $20/mo Gemini Pro subscription. Sprint issues created for Claude (#1897 field dev, #1898 naval arch, #1899 governance) and Codex (#1908 test coverage).