agent-harness-construction

Installation
Summary

Design and optimize AI agent action spaces, tool definitions, and observation formatting for higher completion rates.

  • Action space quality depends on stable tool names, narrow input schemas, deterministic outputs, and avoiding catch-all tools; granularity should match operation risk (micro-tools for high-risk, macro-tools only when round-trip cost dominates.
  • Observation design requires every tool response to include status, summary, next_actions, and artifacts; error paths must provide root cause hints, safe retry instructions, and explicit stop conditions.
  • Context budgeting keeps system prompts minimal, moves large guidance into on-demand skills, and prefers file references over inlining; compaction happens at phase boundaries, not arbitrary token thresholds.
  • Recommends hybrid architecture combining ReAct planning with typed tool execution; track completion rate, retries per task, pass@1/pass@3, and cost per successful task to measure effectiveness.
SKILL.md

Agent Harness Construction

Use this skill when you are improving how an agent plans, calls tools, recovers from errors, and converges on completion.

Core Model

Agent output quality is constrained by:

  1. Action space quality
  2. Observation quality
  3. Recovery quality
  4. Context budget quality

Action Space Design

  1. Use stable, explicit tool names.
  2. Keep inputs schema-first and narrow.
  3. Return deterministic output shapes.
  4. Avoid catch-all tools unless isolation is impossible.
Related skills
Installs
3.6K
GitHub Stars
180.7K
First Seen
Mar 5, 2026