feat: add typed agent progress events

[steipete]

Summary

• export typed agent-progress line and payload contracts from the TypeScript SDK
• make the ephemeral input union match server behavior: typing and progress require exactly one target; presence may remain workspace-wide or targeted
• add agent.progress to the OpenAPI event enum and regenerate SDK declarations
• add compile-time accepted/rejected input-shape coverage and integration docs

Contributor credit: this extracts the generic SDK value surfaced by @arcabotai in #78. The commit retains Cad's co-author credit; the Unreleased changelog thanks @arcabotai.

Proof

• pnpm --filter @clickclack/sdk-ts typecheck && pnpm build:sdk — passed, including positive and negative type-contract assertions.
• pnpm check && pnpm coverage && pnpm audit --audit-level=high && govulncheck ./... — passed; 85.4% coverage; no known JavaScript or Go vulnerabilities.
• pnpm test:e2e — 98 passed in 43.8s.
• Real built server plus built SDK: targetless presence.changed returned 202; a bot-token agent.progress frame with one channel target returned 202. Observed: {"ownerKind":"human","presenceType":"presence.changed","presenceTargetless":true,"progressType":"agent.progress","progressChannelBound":true,"progressTurn":"live-turn-1"}.
• AutoReview local and exact branch review — clean; no accepted/actionable findings.

Not merged.

[steipete]

Exact-head proof for 068ce38bf1e93e8caec8090dcbc573fb4a48bf45:

• pnpm --filter @clickclack/sdk-ts typecheck && pnpm build:sdk — passed, including positive and negative compile-time assertions for targetless presence, targeted typing/progress, dual targets, missing progress targets, and missing progress lines.
• pnpm check && pnpm coverage && pnpm audit --audit-level=high && govulncheck ./... — passed; 85.4% coverage; no known JavaScript or Go vulnerabilities.
• pnpm test:e2e — 98 passed in 43.8s.
• go build -o /private/tmp/clickclack-progress-live ./apps/api/cmd/clickclack, then the real binary with a disposable SQLite database and node /private/tmp/clickclack-progress-live.mjs against the built SDK — targetless workspace presence and targeted bot progress both returned 202. Observed: {"ownerKind":"human","presenceType":"presence.changed","presenceTargetless":true,"progressType":"agent.progress","progressChannelBound":true,"progressTurn":"live-turn-1"}.
• AutoReview local and exact branch review — clean; no accepted/actionable findings.
• Hosted exact-head CI — all seven jobs passed: Go, TypeScript, Playwright E2E, Docker Image, Linux Desktop, macOS Desktop, Windows Desktop.

Land-ready. Not merged.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 068ce38bf1

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Keep progress line types aligned with supported frames

When SDK users publish the progress frames already supported elsewhere in the app, this new public type rejects them: AgentProgress.svelte renders line kinds like plan, patch, and command_output, and the HTTP agent-progress tests still send line.toolName, but AgentProgressLine only allows four kinds and only the snake_case tool_name. Fresh object literals using those supported shapes now fail to type-check through publishEphemeral, so integrations cannot use the typed SDK for existing progress updates without casts.

Useful? React with 👍 / 👎.