Agent助手
- 作者仓库星标 3,406
- 作者更新于 实时读取
- 作者仓库 claude-octopus
- 领域
- AI 智能
- 兼容 Agent
-
- Claude Code
- Cursor
- Cline
- Codex
- Windsurf
- Gemini CLI
- +20
- 信任分
- 88 / 100 · 社区维护
- 作者 / 版本 / 许可
- @nyldn · 未声明 license
- Token 消耗评级
- 低消耗
- 接入复杂程度
- 需简单配置
- 是否需要外部 API Key
- 不需要
- 兼容的系统
- 未声明(默认跨平台)
- 底层运行要求
- 无特殊要求
- 文件与系统权限
-
- 只读
- Shell 执行
- 读取环境变量
- 允许写入 / 修改
- 网络行为
- 仅限本地
- 安装命令数
- 26 条
档案由构建时根据 SKILL.md 与安装命令自动衍生,可能与作者实际意图存在差异。
需要注意: 未限定 allowed-tools,默认拥有全部工具权限。
---
name: skill-cost-projections
description: Project remaining workflow cost from per-phase averages — warns on budget ceiling overruns Proje…
category: AI 智能
runtime: 无特殊运行时
---
# skill-cost-projections 输出预览
## PART A: 任务判断
- 适用问题:提示词、Agent 工作流、模型评估或自动化推理。
- 输入要求:目标材料、限制条件、期望输出和验收方式。
- 证据边界:围绕“Overview / Step 1: Collect Completed Phase Costs / Step 2: Compute Average Cost Per Step”读取原文规则,不把推断写成作者承诺。
## PART B: 执行结果
- **01** 任务判断:确认你的需求是否属于提示词、Agent 工作流、模型评估或自动化推理,并标出输入、限制和预期结果。
- **02** 执行计划:优先按“Overview / Step 1: Collect Completed Phase Costs / Step 2: Compute Average Cost Per Step”拆成步骤,说明每一步会读取什么、修改什么、产出什么。
- **03** 交付结果:给出可复制的命令、文件改动、检查清单或内容草稿,并说明如何继续迭代。
- **04** 风险边界:结合 读取文件、执行终端命令、读取环境变量、写入/修改文件、主要在本地完成、通常不需要额外 API Key 给出执行前确认项。
## Running Rules
- 读取文件、执行终端命令、读取环境变量、写入/修改文件;主要在本地完成;通常不需要额外 API Key。
- 先小样例验证,再放大到真实任务。
- 交付时同时给结果、检查口径和下一步迭代建议。 原文出现了 `/octo` 这类斜杠命令;如果你的 Agent 支持命令触发,优先用命令开场,再补充目标和边界。
告诉 Agent 目标文件或材料、期望结果、不可改范围、是否允许联网或执行命令。本 Skill 的权限画像是:读取文件、执行终端命令、读取环境变量、写入/修改文件。
先用一个小任务确认它会围绕“Overview / Step 1: Collect Completed Phase Costs / Step 2: Compute Average Cost Per Step”工作;涉及文件或命令时,先看 diff、日志、预览或测试结果。
检查最终产物是否包含明确结果、必要证据和下一步动作;如果输出泛泛而谈,就补充输入、边界和验收标准后重跑。
---
name: skill-cost-projections
description: Project remaining workflow cost from per-phase averages — warns on budget ceiling overruns Proje…
category: AI 智能
source: nyldn/claude-octopus
---
# skill-cost-projections
## 什么时候使用
- 把 AI / Agent方向的常用动作沉淀成 Agent 可调用的技能 适合处理AI Agent、提示词、模型评估与自动化推理,核心价值是把输入、判断、执行、验证和交付边界固定下来,避免 Agent 泛泛回答。 把任务拆成可执行、可检查…
- 面向提示词、Agent 工作流、模型评估或自动化推理,优先处理能明确输入、步骤和验收标准的工作。
## 需要提供什么
- 目标材料、目录范围、期望结果和不可改动内容。
- 是否允许联网、执行命令、读写文件或调用外部服务。
## 执行规则
- 围绕「Overview / Step 1: Collect Completed Phase Costs / Step 2: Compute Average Cost Per Step」组织步骤,不把推断写成作者事实。
- 读取文件、执行终端命令、读取环境变量、写入/修改文件;主要在本地完成;通常不需要额外 API Key。
- 先跑小样例,确认结果可检查后再扩大任务范围。
## 输出要求
- 给出最终产物、关键证据、验证方式和下一步动作。
- 信息不足时标记 unknown,不编造命令、平台或依赖。 作者原文负责流程事实;仓库文件负责来源和命令;流狐只补充适用场景、限制和质量判断。
skill "skill-cost-projections" {
输入层 -> 用户目标 + 目标文件 + 禁止范围 + 验收标准
上下文层 -> Overview / Step 1: Collect Completed Phase Costs / Step 2: Compute Average Cost Per Step
规则层 -> SKILL.md 触发条件 / 执行顺序 / 输出格式
运行层 -> 无特殊运行时 | 读取文件、执行终端命令、读取环境变量、写入/修改文件 | 主要在本地完成
安全层 -> 通常不需要额外 API Key + 小任务验证 + diff / 日志复核
输出层 -> 可复制结果 + 检查清单 + 下一步迭代
} Host: Codex CLI — This skill was designed for Claude Code and adapted for Codex. Cross-reference commands use installed skill names in Codex rather than
/octo:*slash commands. Use the active Codex shell and subagent tools. Do not claim a provider, model, or host subagent is available until the current session exposes it. For host tool equivalents, seeskills/blocks/codex-host-adapter.md.
Cost Projections Skill
Overview
Projects remaining workflow cost based on per-phase averages from completed work. Displays cost data in the HUD/statusline and warns when projected totals exceed a configured budget ceiling.
Minimum data requirement: Needs 2+ completed steps before projecting. With fewer than 2 data points, the average is unreliable — display actual spend only and skip projection.
Step 1: Collect Completed Phase Costs
Read cost data from ~/.claude-octopus/metrics/ or from metrics-tracker.sh output. Each completed phase/step should have an associated cost entry.
# Read metrics from the session metrics directory
METRICS_DIR="$HOME/.claude-octopus/metrics"
SESSION_METRICS="$METRICS_DIR/session-$(date +%Y%m%d).jsonl"
# Each line contains: {"phase": "discover", "cost": 0.42, "tokens_in": 12000, "tokens_out": 3400, "timestamp": "..."}
# Sum costs from completed steps
completed_costs=$(jq -s '[.[].cost]' "$SESSION_METRICS")
completed_count=$(jq -s 'length' "$SESSION_METRICS")
total_spent=$(jq -s '[.[].cost] | add' "$SESSION_METRICS")
Data sources (in priority order):
~/.claude-octopus/metrics/session-*.jsonl— structured per-phase cost entriesmetrics-tracker.shoutput — fallback for legacy sessions- HUD accumulated cost counters — last resort
If fewer than 2 completed steps are available, display only actual spend:
💰 Spent: $0.42 (1 step complete — need 2+ for projection)
Step 2: Compute Average Cost Per Step
Calculate the mean cost across all completed steps.
avg_cost=$(echo "$total_spent / $completed_count" | bc -l)
Formula:
avg_cost = total_cost / completed_steps
Example: If 3 steps cost $0.30, $0.50, $0.40 → avg = $1.20 / 3 = $0.40/step
Step 3: Project Remaining Cost
Multiply the average cost per step by the number of remaining steps.
remaining_steps=$((total_steps - completed_count))
projected_remaining=$(echo "$avg_cost * $remaining_steps" | bc -l)
projected_total=$(echo "$total_spent + $projected_remaining" | bc -l)
Formula:
projected_remaining = avg_cost × remaining_steps
projected_total = total_spent + projected_remaining
Step counts by workflow type:
| Workflow | Total Steps | Example Phases |
|---|---|---|
| embrace | 4 | Discover, Define, Develop, Deliver |
| research | 3-7 | Per-agent probe count |
| review | 3 | Fleet, Verifier, Synthesis |
| debate | 3 | Opening, Rebuttal, Synthesis |
Step 4: Display in HUD
Format the cost projection for the HUD/statusline display.
Standard display:
💰 Spent: $2.40 | Est. remaining: $3.60 | Total: ~$6.00
Format rules:
- All dollar amounts to 2 decimal places
- Projected total prefixed with
~to indicate estimate - Use
💰prefix for the cost line - Show in a single compact line for statusline integration
When insufficient data (< 2 steps):
💰 Spent: $0.42 (need 2+ steps for projection)
Integration: This display line is emitted by the octopus-hud hook and rendered in the statusline alongside phase progress and provider status.
Step 5: Budget Ceiling Warning
If the OCTO_BUDGET_CEILING environment variable is set, compare the projected total against it and warn on overrun.
if [[ -n "${OCTO_BUDGET_CEILING:-}" ]]; then
ceiling="$OCTO_BUDGET_CEILING"
if (( $(echo "$projected_total > $ceiling" | bc -l) )); then
overage=$(echo "$projected_total - $ceiling" | bc -l)
echo "⚠️ Budget ceiling: \$$ceiling — projected to exceed by \$$overage"
fi
fi
Display when over budget:
💰 Spent: $2.40 | Est. remaining: $3.60 | Total: ~$6.00
⚠️ Budget ceiling: $5.00 — projected to exceed by $1.00
Display when within budget:
💰 Spent: $2.40 | Est. remaining: $3.60 | Total: ~$6.00
✅ Within budget ceiling: $10.00
Note: OCTO_BUDGET_CEILING is optional. When unset, no ceiling check is performed.
Step 6: Profile Suggestion
If the projected total is high relative to the task, suggest switching to the budget profile to reduce costs.
# Suggest budget profile when projected total exceeds $5.00 (configurable)
COST_THRESHOLD="${OCTO_COST_THRESHOLD:-5.00}"
if (( $(echo "$projected_total > $COST_THRESHOLD" | bc -l) )); then
echo "💡 Tip: Switch to OCTO_PROFILE=budget to reduce costs"
fi
Display:
💰 Spent: $2.40 | Est. remaining: $3.60 | Total: ~$6.00
💡 Tip: Switch to OCTO_PROFILE=budget to reduce costs
The budget profile (OCTO_PROFILE=budget) routes to cheaper models and reduces agent count to minimize spend.
Complete Display Examples
Normal — within budget, moderate cost:
💰 Spent: $2.40 | Est. remaining: $3.60 | Total: ~$6.00
Over budget ceiling:
💰 Spent: $2.40 | Est. remaining: $3.60 | Total: ~$6.00
⚠️ Budget ceiling: $5.00 — projected to exceed by $1.00
💡 Tip: Switch to OCTO_PROFILE=budget to reduce costs
Early in workflow (insufficient data):
💰 Spent: $0.42 (need 2+ steps for projection)
Low cost — no warnings:
💰 Spent: $0.80 | Est. remaining: $0.80 | Total: ~$1.60
Integration
With octopus-hud hook
The cost projection line is rendered as part of the HUD statusline output, updated after each phase completes.
With metrics-tracker.sh
Reads per-phase cost data written by the metrics tracker. Each completed phase logs cost, token counts, and provider details.
With OCTO_PROFILE
When OCTO_PROFILE=budget is active, the projection accounts for reduced per-step costs from cheaper model routing.
Error Handling
No metrics directory:
💰 Cost projection unavailable — no metrics data found
Corrupted metrics data: Skip malformed entries and project from valid data only. If no valid entries remain, show unavailable message.
Zero remaining steps: When all steps are complete, show final total only:
💰 Final cost: $4.80 (4 steps)
先判断是否适合
作者设计意图
作者的方法与取舍
边界和复核