Skip to content

auto-discover and integrate upstream agent harnesses / agent APIs #186

@WiggidyW

Description

@WiggidyW

ObjectiveAI today wraps a fixed set of upstreams — OpenRouter (chat-style LLM API), the claude-agent-sdk, and the codex-sdk. Each new upstream costs human effort to integrate. Make this scale.

Proposal

Build a pipeline of search agents that continuously discover other people's agent harnesses and agent APIs (open-source projects, hosted services, SDKs) and automatically generate the ObjectiveAI integration layer for each: upstream client, prompt/decoding parameter mapping, output-mode coverage, and provider-preference plumbing — the same surface area that agent::completions / objectiveai-claude-agent-sdk-runner / objectiveai-codex-sdk-runner cover today.

End state: ObjectiveAI is the meta-harness — every agent harness that exists becomes one more callable upstream in a Swarm, without any of us writing the glue by hand.

Sketch

  • Discovery agents: crawl GitHub topics / package registries / blog posts / arXiv references for "agent harness", "agent SDK", "agent API", etc. Score candidates by activity, license, and surface fit.
  • Integration generator: given a candidate, generate the runner crate (or library bindings) + the Agent config schema mapping. Treat existing runners as exemplars.
  • Verification: round-trip a small benchmark Swarm through each newly-added upstream to confirm it actually answers.
  • Whitelist gate: human-in-the-loop approval before a generated integration ships, mirroring the plugin-install whitelist pattern.

Why

Every harness we don't integrate is a Swarm we can't run. The integrations themselves are mechanically similar — there's no reason a human should be writing the 20th one by hand.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions