Context engineering and token management. Compaction strategies, progressive disclosure, structured note-taking via AgentDB. Triggers: tokens, compaction, memory, handoff, summarize, context window.
From kernelnpx claudepluginhub ariaxhan/kernel-claude --plugin kernelThis skill is limited to using the following tools:
reference/context-research.mdEnables AI agents to execute x402 payments with per-task budgets, spending controls, and non-custodial wallets via MCP tools. Use when agents pay for APIs, services, or other agents.
Compares coding agents like Claude Code and Aider on custom YAML-defined codebase tasks using git worktrees, measuring pass rate, cost, time, and consistency.
Designs and optimizes AI agent action spaces, tool definitions, observation formats, error recovery, and context for higher task completion rates.
<core_principles>
<token_budget>
<compaction_protocol> Before compaction:
After compaction:
<compaction_strategies>
Before: [full exploration, multiple tool calls, iterations, dead ends] After: "Explored X → Found Y at path/to/file:line → Key insight: Z"
Before: [read 5 files, searched patterns, traced imports] After: "Traced dependency: A→B→C. Issue in B:42. Fix: add null check."
When context grows heavy:
agentdb learn pattern "finding" "evidence"| Transition | Compact? | Why |
|---|---|---|
| Research → Planning | Yes | Research bulk served its purpose; plan is distilled output |
| Planning → Implementation | Yes | Plan is in contract/file; free context for code |
| Implementation → Testing | Maybe | Keep if tests reference recent code |
| Debugging → Next feature | Yes | Debug traces pollute unrelated work |
| Mid-implementation | No | Losing file paths, variable names, partial state is costly |
| After failed approach | Yes | Clear dead-end reasoning before new approach |
</compaction_strategies>
<failure_modes>
Agent tries too much at once → context exhausted mid-work Fix: Incremental progress with commits at each step
Later agent assumes done prematurely → work incomplete Fix: Explicit done-when criteria with evidence required </failure_modes>