Analysis
The best Opencode harness is already here; the better business is the control plane above it
Verdict
If the question is "what is the best OpenCode harness available right now for serious power users?" the answer is oh-my-openagent.
It has the clearest combination of:
- OpenCode-native positioning
- large and current user pull (57k+ GitHub stars; ~122k combined weekly npm downloads across
oh-my-opencode and oh-my-openagent during the rename transition)
- strong opinionated workflow surface (autonomous deep worker,
team_* tool family, security research workflow)
- active maintenance as of today
Why this is not the business to build
The generic harness layer is already crowded and increasingly commoditized.
Evidence:
- OpenCode itself is huge: 159k+ GitHub stars, 1.07M weekly npm downloads, and a 1,274-point / 619-comment Hacker News launch signal.
- oh-my-openagent is already the breakout OpenCode-native harness.
- opencode-workspace already covers team roles, MCPs, permission boundaries, and a curated multi-agent bundle.
- everything-claude-code covers the cross-harness meta-layer: memory persistence, evals, orchestration, dashboards, and security guidance.
- cc-sdd covers spec-driven workflow portability and already ships OpenCode skills in beta.
That means a new entrant that just says "better harness for OpenCode" is entering a noisy open-source knife fight.
Best buildable B2B angle
The better opportunity is an OpenCode harness control plane for teams.
Not a replacement for OpenCode or oh-my-openagent. A layer above them.
ICP
Best initial ICP:
- platform engineering / developer productivity leads at 20-300 engineer software companies
- agencies and consultancies running repeated AI-assisted delivery workflows
- security-conscious startups that want open/local coding agents but need governance before wider rollout
Core pain
These teams no longer mainly ask "can the agent code?"
They ask:
- which harness should this repo or team use?
- what permissions should be allowed by default?
- which workflows are safe to automate fully vs require approval?
- how do we compare success, review load, cost, and failure modes across harnesses?
- how do we keep prompts/skills from turning into ungoverned tribal knowledge?
The source repos themselves point at this pain by converging on the same primitives: permissions, role boundaries, orchestration, skills, memory, evals, dashboards, and security workflows.
Willingness to pay
Willingness to pay looks strongest where agent usage is already happening but rollout is bottlenecked by risk and coordination.
This is not a consumer or indie-dev purchase first. It is a team productivity / governance budget line.
The buyer pays when you can credibly reduce:
- duplicated harness tinkering across teams
Market density and competition gap
Market density is high at the OSS harness layer, but there is still a gap at the team operations layer.
Existing projects are strong at creating agent capability. They are weaker at turning that capability into a repeatable management surface for organizations:
- policy packs by repo type
- comparative run analytics
- recommended harness profile by task type
- reusable workflow scorecards
- lightweight admin UX for non-tinkerers
That is the gap.
Weekend MVP
A credible weekend MVP is:
- ingest session artifacts/logs from OpenCode and one or two popular harnesses
- classify runs by repo, workflow, and policy profile
- show timeline, cost, tools used, approvals requested, failure reason, and outcome
- let a team define simple guardrails per repo type
- rank prompts/skills/workflows by success rate and review burden
In plain terms: start with "post-run observability + policy recommendations," not full orchestration.
Bottom line
Best current harness: oh-my-openagent.
Best company to build: the analytics/governance/workflow-ops layer that helps teams safely standardize around OpenCode harnesses instead of building yet another harness from scratch.