* Initial work on postgres state store
* Run fix
* Replace pre-written CHANGELOG with proper changeset
The CHANGELOG was manually written with a version entry. This repo uses
Changesets to manage versioning, so add a proper changeset file instead.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Add missing @vitest/coverage-v8 dev dependency
The vitest config specifies coverage provider "v8" but the package
was missing from devDependencies.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Drop drizzle-orm, use raw postgres queries
The adapter only needs 3 simple tables with basic CRUD. drizzle-orm is
a full ORM that adds significant dependency weight for no real benefit
here. The ensureSchema() method was already using raw queries. This
halves the dependency surface to just the postgres driver.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Add comprehensive unit tests for postgres state adapter
Test factory function edge cases (missing URL, env var fallbacks,
custom keyPrefix) and ensureConnected guard for all state operations.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Document expired row cleanup limitation for postgres adapter
Unlike Redis, Postgres doesn't auto-delete expired rows. Document the
opportunistic cleanup behavior and provide SQL for periodic cleanup.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Remove stale drizzle keyword from package.json
drizzle-orm was removed in a prior commit but the keyword remained.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Fix expiresAt parsing in acquireLock
The postgres library returns timestamptz columns as JavaScript Date
objects, not strings. Wrapping in new Date() was unnecessary overhead.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Document lock atomicity difference vs Redis adapters
The Postgres ON CONFLICT approach relies on row-level locking rather
than a single atomic SET NX PX like Redis. Note this for users who
need high-contention distributed locking.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Use crypto.randomUUID() for lock token generation
Math.random() is not cryptographically secure and has a higher
collision risk in distributed environments. crypto.randomUUID() is
available in Node 16+ and provides better uniqueness guarantees.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Delete CHANGELOG.md
* Add comprehensive unit tests for postgres state adapter
Mock the postgres module and SQL client to achieve 100% coverage
across statements, branches, functions, and lines. Tests cover
factory function, connection lifecycle, subscriptions, locking,
cache operations, and the owned-client disconnect path.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Add setIfNotExists method and merge with main
Implement the setIfNotExists method added to the StateAdapter
interface since this branch diverged. Uses INSERT ... ON CONFLICT
DO NOTHING with RETURNING to atomically check-and-set.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Add postgres to state docs navigation and memory adapter callout
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Migrate postgres state adapter from postgres to pg (node-postgres)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Rename state-postgres to state-pg and align version to 4.17.0
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Hayden Bleasel <hello@haydenbleasel.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* Fix JSX issues
* fix: add null guards for event.thread in onAction handlers
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* feat(slack): support streaming task_update, plan_update, markdown_text chunks
- Add StreamChunk type for structured streaming content
- Extend stream() to accept AsyncIterable<string | StreamChunk>
- Add taskDisplayMode option to StreamOptions
- Slack adapter passes structured chunks directly to Slack's streaming API
- Handle url_verification before signature check to prevent Slack from
silently disabling event delivery when signing secret is misconfigured
- Add tests for StreamChunk handling in thread.post()
* review fixes: type safety, security, dedup
- Replace loose StreamChunk interface (index signature) with
discriminated union (MarkdownTextChunk | TaskUpdateChunk | PlanUpdateChunk)
- Move url_verification back after signature check to prevent
unauthenticated challenge responses
- Remove redundant AsyncIterable<string> from PostableMessage union
- Extract pushTextAndFlush helper to deduplicate markdown flush logic
- Remove unnecessary `as string` casts after type guards
- Export individual chunk types from package index
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* add changeset
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Hayden Bleasel <hello@haydenbleasel.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix openModal crash when action has no thread context
Home tab actions (e.g. Change Team) have no thread → openModal closure
crashed accessing thread.channel on null. Guard thread access in the
openModal closure and pass undefined to storeModalContext.
Add test for empty threadId action → openModal works with null thread.
* add changeset
* make ActionEvent.thread nullable, remove unsafe cast
- Change `ActionEvent.thread` type from `Thread` to `Thread | null`
- Remove `null as unknown as Thread` unsafe cast in handleActionEvent
- Update test files to use null guards and optional chaining
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Hayden Bleasel <hello@haydenbleasel.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: preserve nesting depth when rendering lists in all chat adapters
All format converters (Slack, Discord, Teams, Google Chat) were flattening
nested lists during Markdown→platform conversion. When a listItem had a
paragraph followed by a child list, the child list output was concatenated
directly onto the parent item's text with no indentation or separating
newline, producing garbled output like "• parent• child 1\n• child 2".
Add an optional `depth` parameter to each adapter's private `nodeToX()`
method. The `isListNode` handler now builds lines explicitly: paragraph
content is prefixed with the bullet/number at the correct indent level
(`" ".repeat(depth)`), and nested list nodes are rendered recursively
at `depth + 1`. Sibling items at the same depth are unaffected.
Add regression tests covering:
- nested unordered lists (2 levels)
- nested ordered lists (2 levels)
- deeply nested lists (3 levels)
- flat sibling items (no regressions)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* refactor: extract renderList() helper to eliminate dead depth parameter
The previous fix added `depth = 0` to each adapter's general nodeToX()
method, but only the isListNode branch ever used it — a dead parameter
across ~10 other branches that made the signature misleading.
Extract a private renderList(node: List, depth: number) method in each
adapter. nodeToX() stays as a clean single-argument node converter;
all depth-aware list logic lives exclusively inside renderList(), which
recurses into itself for nested lists and calls nodeToX() for leaf nodes.
Import the List type from the chat package to give renderList() an
explicit, narrowed parameter type.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* Deduplicate renderList into BaseFormatConverter, fix start offset, add mixed list tests
- Extract shared renderList() into BaseFormatConverter with configurable bullet char
- Respect List.start property for ordered lists starting at non-1 values
- Remove dead isListItemNode branches from all adapters
- Add mixed ordered/unordered nesting test to all adapters
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: Hayden Bleasel <hello@haydenbleasel.com>
* Add Telegram auto mode polling and init username fallback
* Expose Telegram resetWebhook API
* Refine Telegram API to longPolling and safer auto mode
* Fix: Environment variable cleanup sets `process.env.VERCEL` to the string `"undefined"` (truthy) instead of removing it, leaking a truthy VERCEL env var into subsequent tests.
This commit fixes the issue reported at packages/adapter-telegram/src/index.test.ts:661
**Bug explanation:**
In the test "auto mode stays in webhook mode on serverless runtime" (line 622), the cleanup code in the `finally` block at line 662 uses `process.env.VERCEL = undefined` when the env var wasn't previously set. In Node.js, `process.env` coerces all values to strings, so `process.env.VERCEL = undefined` actually sets `process.env.VERCEL` to the string `"undefined"`. This is truthy (`Boolean("undefined")` === `true`).
This was verified empirically:
```
process.env.TEST_VAR = undefined;
typeof process.env.TEST_VAR // "string"
process.env.TEST_VAR // "undefined"
Boolean(process.env.TEST_VAR) // true
```
The consequence is that after this test runs, `process.env.VERCEL` remains set to the truthy string `"undefined"`, causing `isLikelyServerlessRuntime()` to return `true` for all subsequent tests that run in the same process. This could cause any auto-mode tests that follow to incorrectly detect a serverless runtime and behave differently than expected.
**Fix explanation:**
Changed `process.env.VERCEL = undefined` to `delete process.env.VERCEL`, which properly removes the environment variable from `process.env`. After `delete`, `process.env.VERCEL` is `undefined` (the actual undefined value, not the string), and `Boolean(process.env.VERCEL)` correctly returns `false`. This matches the standard pattern for cleaning up environment variables in Node.js tests.
Co-authored-by: Vercel <vercel[bot]@users.noreply.github.com>
Co-authored-by: timolins <me@timo.sh>
* Fix lint: replace delete env cleanup with Reflect.deleteProperty
* Harden Telegram polling tests and polish polling docs
* Improve Telegram polling docs with clearer motivation and examples
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Vercel <vercel[bot]@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* feat(chat, adapter-gchat): add disabled prop to Button
Google Chat Cards v2 natively supports `disabled` on buttons, but
`ButtonOptions` / `ButtonElement` didn't expose it. Add the prop to
the core type and thread it through the gchat adapter serializer.
Closes#149
Made-with: Cursor
* feat(adapter-discord): thread disabled prop through Discord button serializer
DiscordButton already had `disabled` in its type but convertButtonElement
never set it from ButtonElement. Wire it through so disabled buttons
render correctly on Discord too.
Made-with: Cursor
github-actions[bot]