harness: evaluate Kimi Code, Qoder CLI, Continue cn, and SWE-agent targets

[willwashburn]

Summary

Run a discovery spike for lower-confidence or broader-market coding harness targets before splitting implementation issues.

Context

These tools appear to have local CLI, agent, or automation surfaces, but each needs verification before committing to first-class Relay support.

Candidates:

• Kimi Code: https://www.kimi.com/code/docs/en/
• Qoder CLI: https://docs.qoder.com/cli/using-cli
• Continue CLI (cn): https://docs.continue.dev/guides/cli
• SWE-agent / mini-swe-agent: https://swe-agent.com/latest/usage/cli/

Scope

• Confirm each tool has a local spawnable CLI suitable for Relay.
• Identify non-interactive/headless command flags and auth requirements.
• Check whether each supports MCP/config hooks, ACP, JSON output, or another structured integration point.
• Rank candidates by implementation cost and expected user value.
• Split concrete harness issues for candidates that pass the spike.

Acceptance

• A short matrix records viable vs non-viable candidates with source links.
• Follow-up implementation issues are created for viable candidates.
• Non-viable candidates have explicit rationale so they are not rediscovered repeatedly.