メニュー
Measure Manciple handoff prompt size for a task by running the local token audit script and reporting its deterministic bucket output.
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。
SOC 職業分類に基づく
Coordinate Manciple task-runner workers efficiently across actionable Manciple work, with each worker using the manciple-mcp-task-runner skill and Manciple MCP tools. Use when the user asks to spawn Manciple workers, process multiple Manciple tasks in parallel, or coordinate Manciple task execution across sub-agents.
Complete Manciple task work using the Manciple MCP server, optionally focused on implementation, docs, test, refactor, review, or hardening work; default to any of those task types when no focus is given. Use when the user asks Claude Code to run, complete, pick up, execute, or work through a Manciple task in this repo, especially when MCP tools named manciple_get_task_packet, manciple_verify, manciple_format_task, manciple_set_status, or manciple_run_log are available.
Review completed Manciple task work by loading the task packet, checking review readiness with manciple_verify --profile review, evaluating acceptance criteria against run log and git diff evidence, and recording a deterministic verdict (approve/request-changes/block). Use when the user asks Claude Code to review, approve, or give feedback on a completed Manciple task.
Plan and create Manciple task specs from a product brief, feature description, GitHub issue, or engineering goal. Use when the user asks Claude Code to plan, decompose, break down, or create tasks for a feature or goal — especially before any implementation starts.
Coordinate Manciple task-runner workers efficiently across actionable Manciple work, with each worker using the manciple-mcp-task-runner skill and Manciple MCP tools. Use when the user asks to spawn Manciple workers, process multiple Manciple tasks in parallel, or coordinate Manciple task execution across sub-agents.
Complete Manciple task work using the Manciple MCP server, optionally focused on implementation, docs, test, refactor, review, or hardening work; default to any of those task types when no focus is given. Use when the user asks Codex to run, complete, pick up, execute, or work through a Manciple task in this repo, especially when MCP tools named manciple_get_task_packet, manciple_verify, manciple_format_task, manciple_set_status, or manciple_run_log are available.
| name | manciple-token-auditor |
| description | Measure Manciple handoff prompt size for a task by running the local token audit script and reporting its deterministic bucket output. |
Use this skill when the user asks Claude Code to audit, estimate, measure, or compare Manciple prompt token size for a task.
node scripts/manciple-token-audit.mjs <task-id>
Pass through any supported estimator options the user requests, such as --budget <tokens>, --include-review, --include-run-log, --include-diff, or --include-git-context.
Report the script output. Preserve the measured buckets, total, budget, and risk line from the receipt.
Do not manually count tokens or invent estimates. The script delegates to Manciple's deterministic local token-estimate command so reports are repeatable and do not require network access.
This audit measures Manciple handoff size only. It does not measure total agent spend from file reads, tool calls, retries, internal reasoning, generated output, or model/provider accounting.