<< All versions
Skill v1.0.0
currentAutomated scan100/100ba-calderonmorales/terminal-jarvis/token-budget
──Details
PublishedJune 23, 2026 at 12:27 AM
Content Hashsha256:93c1f0164f26c676...
Git SHAc4b578d74326
──Files
Files (1 file, 2.4 KB)
SKILL.md2.4 KBactive
SKILL.md · 79 lines · 2.4 KB
version: "1.0.0"
Skill: Token Budget Management
Name: token-budget Description: Token efficiency patterns for long AI coding sessions Trigger: Long sessions, complex multi-step tasks, session optimization
Objective
Maximize session duration beyond 12 hours through intelligent orchestration.
Budget Caps
| Agent Type | Model | Max Tokens | Responsibilities | |
|---|---|---|---|---|
| Lead | Sonnet 4.5 | 1000 | Planning, architecture, orchestration, integration, validation | |
| Spawned | Haiku Latest | 750 | File edits, docs, tests, refactoring, specialized tasks |
Lead Orchestrator Pattern
Phase 1: Planning (Lead - 1000 tokens)
- Analyze request scope
- Break into discrete tasks
- Identify parallelization opportunities
- Select appropriate agents
- Define validation criteria
Phase 2: Delegation (Agents - 750 tokens each)
- Spawn agents in parallel for independent tasks
- Sequential for dependencies
- Focused context (only what's needed)
- Clear success criteria
Phase 3: Integration (Lead - 1000 tokens)
- Collect results
- Run quality checks (cargo check, clippy, fmt, test)
- Coordinate commits
- Update documentation
Token Conservation Techniques
| Technique | Good | Bad | |
|---|---|---|---|
| Delegation | Lead plans, 5 agents implement | Lead implements everything | |
| Parallelization | 3 agents simultaneously | Sequential when independent | |
| File reads | Read specific file | Read entire directory tree | |
| Validation | cargo check after each agent | Accumulate, validate at end | |
| Context | "Update lines 45-60 in file X" | "Read file, understand, update" |
Target Metrics
- Agent utilization: >80% (most work by agents)
- Parallelization: >40% (many concurrent)
- Session duration: >12 hours
Efficiency Guidelines
- DO NOT create documentation files unless explicitly requested
- DO leverage specialized agents proactively
- DO use concise responses - skip verbose explanations when action is clear
Proactive Agent Usage
AI assistants MUST invoke specialized agents immediately without waiting to be asked:
| Scenario | Agent | |
|---|---|---|
| Documentation | @documentation-specialist | |
| Testing | @qa-automation-engineer | |
| Code review | @code-reviewer | |
| Security | @security-specialist | |
| Infrastructure | @devops-engineer | |
| Architecture | @software-architect | |
| Performance | @performance-specialist |