Skill: token-cost-estimation

What I do

I estimate token costs BEFORE work sessions begin, enabling informed decisions about workflow optimisation. I provide structured cost breakdowns, identify savings opportunities, and track actual vs estimated usage via memory-keeper for continuous improvement.

When to use me

• Always-active: Load with every session automatically
• At the START of any work session before executing tasks
• When planning complex multi-step tasks
• When token budget is constrained
• During retrospectives to compare estimates vs actuals

Core principles

1. Estimate upfront - Never start work without understanding expected cost
2. Break down costs - Show components: investigation, implementation, verification
3. Identify savings - Recommend optimisations before starting
4. Track accuracy - Store estimates and actuals to improve over time
5. Integrate with workflow - Use parallel-execution, scope-management to reduce costs

Estimation Framework

Task Complexity Tiers

Duration Multipliers

Resource Factors

• Files involved: +100 tokens per file read/modified
• Codebase familiarity: New (2x), Familiar (1x), Expert (0.7x)
• Tool usage: Each tool call ~50-100 tokens overhead
• Verification: Tests add 30-50% to implementation cost

Cost Breakdown Template

## Token Cost Estimate

**Session Goal**: [state objective]
**Complexity Tier**: [Simple/Moderate/Complex/Major]
**Estimated Duration**: [time]

### Breakdown
| Phase | Estimated Tokens | Notes |
|-------|------------------|-------|
| Investigation | X | File reads, search, context |
| Implementation | Y | Edits, writes, iterations |
| Verification | Z | Tests, checks, validation |
| **Total** | **X+Y+Z** | |

### Optimisation Opportunities
- [ ] Parallel investigation (save ~X tokens)
- [ ] Scope reduction (save ~Y tokens)
- [ ] Efficient prompting (save ~Z tokens)

### Estimated vs Budget
- Estimate: X tokens
- Budget: Y tokens (if applicable)
- Difference: +/- Z

Savings Strategies

From parallel-execution

• Fan-out investigation: Read multiple files simultaneously
• Parallel verification: Run lint/test/check in parallel
• Estimated savings: 20-40% on investigation phase

From scope-management

• Reduce scope to essential deliverables
• Defer nice-to-haves to separate sessions
• Estimated savings: Variable (scope-dependent)

From token-efficiency

• Structure prompts clearly
• Provide focused context
• Use examples over descriptions
• Estimated savings: 10-30%

Post-Session Tracking

After session completion:

1. Record actual token usage
2. Compare to estimate
3. Store in memory-keeper:
   Copy

   ESTIMATE: [prediction]
   ACTUAL: [result]
   VARIANCE: [difference]
   FACTORS: [what caused variance]
   → Update estimation heuristics
   

KB Reference

~/vaults/baphled/3. Resources/Knowledge Base/AI Development System/Skills/Workflow-Orchestration/Token Cost Estimation.md

Related skills

• pre-action - Clarify scope before estimating
• memory-keeper - Store estimates and actuals
• estimation - Task complexity evaluation
• time-management - Duration estimation
• task-tracker - Progress and complexity tracking
• scope-management - Resource identification
• token-efficiency - Cost reduction techniques
• parallel-execution - Efficiency through parallelism