smart-context
by joe3112
AI & LLMsBrief
Token-efficient agent behavior — response sizing, context pruning, tool efficiency, and delegation.
Description
What This Skill Does
Instructs agents to match response length to question complexity, batch tool calls, and avoid loading files unnecessarily. Covers response sizing rules, context pruning, vision call costs, and delegating bulk work to cheaper sub-agent models.
Instead of applying uniform verbosity to every task, it uses a cost-aware decision tree that trims tokens on simple work while preserving depth where complexity demands it.
When to Use It
- Cutting API costs on high-volume agent pipelines
- Batching parallel tool calls to reduce round-trips
- Skipping memory lookups for simple greetings or status checks
- Delegating bulk file formatting to a lighter model
- Getting direct answers without preamble or restatements
Example Workflow
User asks: Is the config file present?
- 1
Classifies question as yes/no type
- 2
Determines a single targeted check is sufficient — no full file read needed
- 3
Makes one tool call with a specific path lookup
- 4
Skips confirmation read after the check
Yes, config.json exists in the project root.
Installation
clawhub install joe3112/smart-contextRun the CLI command, or paste the prompt into Claude Code, Cursor, Codex or whichever coding agent you use.