name: cost-predict description: 'Use when you need this Cognitive OS skill: Predict task cost from Cognitive OS history, phase routing, and measured model prices.; do not use when a narrower skill directly matches the task.' summary_line: Estimate task cost from historical metrics and default phase routing. version: 1.0.0 last-updated: 2026-04-23 auto-generated: false tech: python routing_intents:

• intent: forecast_task_model_cost

description: User asks for an estimated cost of a planned task using Cognitive OS history, phase routing, and model price data. confidence: 0.87

• intent: budget_before_running_work

description: User wants a pre-execution budget forecast or cost risk estimate, not a retrospective accounting report. confidence: 0.83 triggers:

• cost predict
• predict cost
• estimate cost
• budget forecast
• how much will this cost

audience: project platforms:

• claude-code

prerequisites: []

<!-- SCOPE: both -->

Cost Predict

Purpose

Give a real, bounded cost estimate before starting medium+ work. This skill uses the existing lib/cost_predictor.py engine and the metrics already written by Cognitive OS, so the output is grounded in project history rather than a generic guess.

Invoke

/cost-predict <task-description>

When to Use

• Before starting a medium, large, or critical task
• When deciding whether to split work into smaller slices
• When comparing implementation strategies with different likely costs
• When you need a budget forecast tied to real repository history

Steps

Step 1: Classify the task

Read the task description and choose the closest task_type:

• feature
• bugfix
• refactor
• docs
• research

If the task spans multiple types, choose the dominant cost driver and say so explicitly.

Step 2: Run the predictor

Use the executable wrapper:

python3 scripts/cost_predict.py --type feature "Add codex-first settings projection to bootstrap"

If you need structured output for follow-on automation, use:

python3 scripts/cost_predict.py --json --type feature "Add codex-first settings projection to bootstrap"

Step 3: Read the evidence

Interpret the result in this order:

1. Estimated range (min, mid, max)
2. Confidence level
3. Basis (historical, model_routing, no_data)
4. Similar tasks, if present
5. Recommendation

If the basis is no_data or confidence is low, say that clearly and treat the output as a planning hint, not a commitment.

Step 4: Turn the estimate into a decision

• If cost is low and confidence is decent, proceed normally
• If cost is moderate or high, propose splitting the work
• If confidence is low, recommend a smaller exploratory slice first
• If the estimate is driven mostly by expensive phases, call that out

Output Format

## Cost Prediction
 
Task: Add codex-first settings projection to bootstrap
Task type: feature
Estimate: $0.42-$0.96 (mid: $0.63)
Confidence: MEDIUM (0.58)
Basis: historical
Recommendation: Most expensive phase: design (opus). Consider sonnet for non-reasoning phases.

Contextual Trigger

Use this skill when the user asks how much a task will cost, asks for a budget forecast, or needs a quick estimate before implementation.