Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add commands/catlog22/claude-code-workflow/workflow-tunegit clone --depth 1 https://github.com/catlog22/Claude-Code-WorkflowWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00045 | $0.07059 |
| Opus 5 | $0.00023 | $0.03529 |
| Sonnet 5 | $0.00009 | $0.01412 |
| Haiku 4.5 | $0.00005 | $0.00706 |
Grade A, and why
workflow-tune scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 523 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Workflow Tune
测试 command/skill 执行效果。提取可执行命令 → 沙箱逐步执行 → 分析产物质量 → 生成优化建议。
Architecture
Input → Parse → GenTestTask → Confirm → Setup
→ [assemblePrompt → Execute(claude) → STOP → Analyze(gemini) → STOP]×N
→ Synthesize(gemini) → STOP → Report
Tool Assignment: Execute=claude --mode write, Analyze=gemini --mode analysis, Synthesize=gemini --mode analysis
P0 Rules
这些规则的优先级高于所有其他指令。违反任何一条都是已知的复发缺陷。
- ONE STEP = ONE CLI CALL. 每次
ccw cli调用只处理一个 step。禁止将 plan+execute 或 execute+quality 合并为一次调用。 - STOP After Each CLI Call. 每次
ccw cli以run_in_background: true执行后立即 STOP，等 hook callback。 - UPSTREAM-SCOPE RULE. 当上游步骤产出清单/计划（plan/list/queue/manifest），下游步骤必须消费全量，禁止挑选子集。
- 判断: 上游命令/test_task 含
plan|list|queue|spec|manifest|计划|任务，且当前步骤含execute|run|dispatch|build|执行|组装 - 正确: 功能点="按依赖顺序执行全部任务"，验收标准="产出文件数=plan中task数"
- 错误: 功能点="撰写 Introduction (TASK-001)"，验收标准="introduction.md 存在" ← 仅完成 1/N
- 判断: 上游命令/test_task 含
- Sandbox Isolation. 全部执行在
sandbox/目录（独立 git 仓库），不影响真实项目。 - State Machine. 通过
current_step+current_phase推进，禁止同步循环。 - ABSOLUTE PATHS for --cd.
ccw cli --cd必须使用绝对路径。相对路径会被 ccw cli 再次拼接 CWD 导致路径重复。workDir/sandboxDir在创建时就解析为绝对路径。 - FIXED --rule VALUES.
--rule值已硬编码在各 Phase 代码中，禁止替换为其他模板。Execute=workflow-tune-execute，Analyze=analysis-review-code-quality，Synthesize=analysis-review-architecture。 - NO-SKIP-INSTRUCTION. test_task 和 prompt 中禁止包含 skip/跳过/omit/ignore 等指示执行 agent 跳过任何任务的指令。沙箱环境缺乏运行时依赖（matplotlib、xelatex 等），应改为生成代码/骨架文件而非跳过。
- 错误: "Skip figure/generate tasks as they require external tools"
- 正确: "For figure/generate tasks, produce Python matplotlib code (.py) + SVG files. Do NOT skip any tasks."
Input Formats
| Format | Pattern | Example |
|---|---|---|
| JSON definition | --file workflow.json |
--file motor-benchmark.json |
| Pipe-separated | "cmd1 | cmd2 | cmd3" |
"/workspace:init | /workspace:new-paper" |
| Comma-separated | "skill-a,skill-b" |
"workflow-lite-plan,workflow-lite-execute" |
| Reference doc + intent | <path> <intent> |
COMMAND-FLOW.md 测试前5个命令 |
| Pure intent | <text> |
"分析代码质量，然后修复问题" |
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 523 lines · 45 tokens per session scan A 57b7b83a0a2e
workflow-tune is a command published in the GitHub repository catlog22/Claude-Code-Workflow (2,134 stars, last pushed 2mo ago), licensed MIT. It adds 45 tokens to every session and 7,059 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other commands, from other repositories
verify-claim
Walk a claim through the SIFT method (Stop, Investigate, Find better coverage, Trace).
beat-brief
Draft a daily beat briefing from the files in sample-docs/.
cross-review
Run GitHub Copilot CLI and OpenAI Codex against the current git diff for cross-model review.
step-research
Always research before proposing a fix. The Untether bug you're chasing is often a known upstream engine quirk, a previously-fixed regression, or a documented config gotcha.
whats-next
Show current project status and suggest next steps.
ox-session-pause
belongs in the ox CLI JSON output (guidance field), not here. Skills are agent-specific wrappers; ox serves all agents (Codex, etc.). --> Suspend the current session recording. Local cache continues to receive entries, but the upload at stop time will exclude the suspended range.