[BadhriDO]

Summary

Connection create for Harness Runtime is now strict (AR-3404 / MARSOHS-2480). If the actor already has an active, pending, or expired connection for the provider, POST /v2/agents/auth/{provider}/connections returns 409 instead of resuming it. The spec still describes the old behaviour, so the generated clients (dots, pydo) document it wrongly.

• Create (agents_create_provider_connection.yml): the description no longer says "Creates, or resumes". It explains the one-connection-per-actor rule and points to reconnect, and the 409 is now documented.
• Reconnect (agents_reconnect_provider_connection.yml): the description now says what reconnect does. It resumes a pending authorization, or restarts an expired or revoked connection. The connection keeps its id, user_id and scopes. An active connection is returned unchanged, and only the owning member can reconnect.

No paths, parameters or schemas change. The 409 reuses the file's existing rate-limit headers and the error schema, the same way delete documents its 409.

Production behaviour

Checked against production on 2026-11-11: calling create for an actor that already has an active connection returns

HTTP 409
{"id":"conflict","message":"already exists: a connection already exists for this provider and actor; use reconnect or update, or delete it before changing credentials", ...}

The duplicate check runs before any scope or credential check, so with strict create this is the only conflict the public create returns.

The reconnect route itself was documented and tested against production in #1301. The SDKs already have it: dots since v1.22.0, pydo through both its hand-written client and its generated client, godo since v1.234.0, and doctl through a PR in review.

Test plan

• Both changed files parse as YAML, and the 409's *id001 header alias resolves.
• Spectral lint. I didn't run it locally and am leaving it to this PR's CI.

🤖 Generated with Claude Code