English | 简体中文

Easy Agent Harness

Type: Low-Code Harness Owner: study-derived Mode: Public Base

Easy Agent Harness is a study-owned public repo for a beginner-friendly, low-code agent harness.

It is built for ordinary users who want agents to handle everyday requests without first buying into a coding-first CLI workflow.

It favors explicit rules, readable control surfaces, and predictable behavior over prompt magic or framework cleverness.

It is intentionally smaller than the private study workspace, and different from learning-os.

• study canonical private owner
• learning-os public learning harness
• easy-agent-harness public low-code agent harness baseline

Proof first: verify the repo-safe surface, inspect one compact support loop, then read the deeper docs.

Positioning

The public positioning is:

• beginner-friendly and low-code, not code-first
• a rules-first harness for everyday agent use
• public baseline, not canonical doctrine
• local-first and machine-readable where that reduces ambiguity

The repo moved from study-superpowers to study-agent-harness, and now to easy-agent-harness.

The current name keeps the artifact type legible while lowering first-click friction. low-code belongs in the positioning, not in the slug.

Current Harness Core

The current public release is still deliberately small.

It ships the control spine of the harness rather than a giant task catalog:

• repo-safe validation entrypoints
• public-boundary bootstrap
• AI-facing onboarding and machine-readable routing
• explicit packet ownership handoff for ops-shaped work
• repair scope and patch-home classification
• execution receipts for ops-shaped packets

That control loop is:

ops-handoff -> repair-scope -> execution-receipt

This repo is not trying to mirror the full private workspace on day one. The current release is the readable, verifiable control core.

Proof Lane

If you only want to verify that the current public slice is real, run:

pwsh -NoProfile -File ./tools/Test-All.ps1 -RepoOnly

Expected result:

• PowerShell validators parse cleanly
• machine-readable routing files load
• required public surfaces exist
• maintainer-only working state stays out of the repo

Then open:

• docs/demo-flow.md

That demo shows the current harness control loop end to end:

ops-handoff -> repair-scope -> execution-receipt

Why This Repo Exists

Many public agent repos assume strong coding ability, terminal fluency, or a willingness to live inside a developer-first framework.

This repo chooses another tradeoff:

• lower code burden
• clearer rules and boundaries
• smaller readable surfaces
• predictable execution and validation
• honest separation between public result and private maintainer state

The current public core is still narrow, but the direction is broader than a single support slice: it is meant to be the base harness for ordinary rule-driven agent use.

Repo Map

• CHANGELOG.md public release history for the harness repo
• ROADMAP.md near-term public-facing priorities
• CONTRIBUTING.md contribution scope and validation expectations
• AI_CONTEXT.md shortest AI-facing entrypoint
• system.md public identity and operating rules
• docs/public-scope.md what the current slice includes and excludes
• docs/architecture.md relationship to study and learning-os
• docs/packet-ownership.md public-safe handoff contract for ops-shaped packets
• docs/repair-scope.md public-safe contract for scope classification and patch-home choice
• docs/execution-receipts.md public-safe contract for concise operational close-out receipts
• docs/run-with-codex.md minimal Codex execution path
• docs/demo-flow.md compact proof of the first shipped support loop
• agent/README.md public-safe agent layer
• agent/skills/repo-ops-and-boundary/SKILL.md minimal operator skill
• agent/skills/ops-handoff/SKILL.md packet ownership handoff skill for ops-shaped work
• agent/skills/repair-scope/SKILL.md repair-scoping skill for choosing patch home before implementation
• agent/skills/execution-receipt/SKILL.md execution-receipt skill for operational close-out evidence
• tools local validators for this public slice

Next

• Proof first Run the repo-only checks, then open docs/demo-flow.md.
• AI agent entry Start with AI_CONTEXT.md, then task-router.json.
• Beginner-facing direction Use ROADMAP.md to see which ordinary-use proof or onboarding surface should land next.
• Contribution and roadmap Use ROADMAP.md, CONTRIBUTING.md, and CHANGELOG.md.

Boundary

This public repo does not include:

• private doctrine
• learner-specific state
• maintainer-only working files
• private logs or runtime memory
• a mirror of the full private support stack
• a code-first framework posture disguised as beginner tooling

The canonical owner remains study.