Bugbot findings: (#1446 High) the agent loop shipped without the slack
cap, breaking Slack bang commands and LLM replies at this point in the
stack — the slack (full guidance text, including the quiet-reply rule and
the Promise.all acknowledgment pattern, addressing the #1447 Medium) and
agents caps move down from part 3 to where the loop lands, with the
SlackCapability call() adapter and allowlist entries. (#1447 High at its
source) ensureItxContext seeding is now versioned: bump
AGENT_CONTEXT_CAPS_VERSION and existing agents re-define caps on next
wake (defines upsert; capability-noted idempotency keys dedupe).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Bugbot findings: (#1446 High) the agent loop shipped without the slack
cap, breaking Slack bang commands and LLM replies at this point in the
stack — the slack (full guidance text, including the quiet-reply rule and
the Promise.all acknowledgment pattern, addressing the #1447 Medium) and
agents caps move down from part 3 to where the loop lands, with the
SlackCapability call() adapter and allowlist entries. (#1447 High at its
source) ensureItxContext seeding is now versioned: bump
AGENT_CONTEXT_CAPS_VERSION and existing agents re-define caps on next
wake (defines upsert; capability-noted idempotency keys dedupe).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Bugbot findings: (#1446 High) the agent loop shipped without the slack
cap, breaking Slack bang commands and LLM replies at this point in the
stack — the slack (full guidance text, including the quiet-reply rule and
the Promise.all acknowledgment pattern, addressing the #1447 Medium) and
agents caps move down from part 3 to where the loop lands, with the
SlackCapability call() adapter and allowlist entries. (#1447 High at its
source) ensureItxContext seeding is now versioned: bump
AGENT_CONTEXT_CAPS_VERSION and existing agents re-define caps on next
wake (defines upsert; capability-noted idempotency keys dedupe).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Final follow-up to the codemode rip (#1445/#1446/#1447). MCP's
streamable HTTP transport is a stateless protocol — fetch with metadata
— so nothing about an MCP client belongs in a Durable Object.

## What

- **`OutboundMcpFromOurClientCapability` DO deleted** — class, worker
export, alchemy namespace + `OUTBOUND_MCP_FROM_OUR_CLIENT_CAPABILITY`
binding, vitest harness wiring. Nothing has dialed it since the codemode
rip (its only caller was the deleted provider registration).
- **Core helpers promoted** to `src/itx/caps/mcp-client-core.ts` with
clean names: `connectMcp` (custom-fetch aware, closes on failed
connect), `listMcpTools`, `executeMcpToolCall`. `McpClient` dedupes onto
them; the mock-server unit test moved and adapted.
- **The DO allusion is gone from the docs**: `McpClient`'s header now
states statelessness as the design ("connect → call → close, per
invocation; deliberately no Durable Object anywhere in this path")
instead of presenting connection caching as a future optimization.

## Deploy note

This is the repo's first Durable Object class deletion. The DO had no
SQLite (in-memory cache only) and no inbound subscribers, so deletion is
data-safe; alchemy is expected to emit the `deleted_classes` migration
on deploy — **this PR's preview deploy is the proof**. If it refuses,
the fallback is a tombstone class like `CodemodeSession`'s.

## Testing

Repo typecheck / lint / knip / unit suite green locally (incl. the moved
mock-server core test). The `McpClient` e2e remains gated on a reachable
MCP server.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

<!-- CLOUDFLARE_PREVIEW -->
## Environment Config Lease
<!-- CLOUDFLARE_PREVIEW_STATE -->
<!--
{
  "apps": {},
  "environmentConfigLease": {
    "dopplerConfig": "preview_4",
    "leasedUntil": 1781109925654,
    "leaseId": "f7427e44-72e2-4e04-8617-122e174b3c57",
    "slug": "preview-4",
    "type": "environment-config-lease"
  }
}
-->
<!-- /CLOUDFLARE_PREVIEW_STATE -->
Lease: `preview-4`
Doppler config: `preview_4`
Type: `environment-config-lease`
Leased until: 2026-06-10T16:45:25.654Z
<!-- /CLOUDFLARE_PREVIEW -->

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

… it in transit (#1462)

Narrowed after #1447 landed: that PR fixed the e2e test's assertions
(asserting the real wire shape, `name === "Error"` + raw `code`), which
unblocks preview e2e on main. What it did NOT fix is the detection
helper itself — `getItxErrorCode` still requires `error.name ===
"ItxError"`, and capnweb 0.8.0 drops custom error names in transit (the
receiver maps only builtin names to classes; the props loop skips `name`
on both ends). So after any capnweb hop, every runtime consumer of the
helper — react hooks, `stream-tail`, `isItxAccessError` retry predicates
— fails to recognize kernel errors and treats NOT_FOUND/FORBIDDEN like
connection noise.

Probed against a live preview deployment:

```
name: Error                    ← dropped by capnweb
code: NOT_FOUND                ← survives (own enumerable prop)
details: {"projectIdOrSlug":"definitely-not-a-project"}  ← survives
```

## Fix

`getItxErrorCode` duck-types on the closed five-code set alone; `name`
is not consulted (it cannot be, after capnweb). Set membership keeps
foreign coded errors (`ECONNREFUSED`, …) out. The unit-test
`simulateCapnwebCrossing` helper no longer reattaches `name` by hand —
that hand-modeled reattachment is exactly how the original bug stayed
green in unit tests while failing on the real wire. Docs in `errors.ts`
updated to match (merged with #1447's wording).

## Verification

- The capnweb-crossed shape (plain `Error` + `code`) is now detected;
unit tests assert it
- `pnpm test:itx-stream-subscribe` (real Workers RPC boundary, which
legitimately preserves `name`) still passes
- Full apps/os suite, typecheck, lint clean; e2e verified earlier
against a live preview deployment (preview_6)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

<!-- CURSOR_SUMMARY -->
---

> [!NOTE]
> **Medium Risk**
> Changes shared error-classification used for retry vs access-failure
handling across capnweb clients; behavior widens slightly (code-only
shapes now match) but is constrained by the five-code allowlist.
> 
> **Overview**
> **`getItxErrorCode` no longer requires `error.name === "ItxError"`** —
it treats any object whose `code` is one of the five itx codes as an
ItxError. That matches errors after a capnweb RPC crossing, where the
receiver rebuilds a plain `Error` and **drops** the name while `code` /
`details` still arrive as own props.
> 
> Docs on `getItxErrorCode` in `errors.ts` now spell out why `name` must
not be used and how the closed code set avoids colliding with other
`code`-bearing errors (e.g. `ECONNREFUSED`).
> 
> Unit tests were aligned with real capnweb: `simulateCapnwebCrossing`
no longer re-attaches `name`, duck-typed cases use plain `Error` +
`code` only, and rejection cases use foreign codes instead of wrong
`name`.
> 
> <sup>Reviewed by [Cursor Bugbot](https://cursor.com/bugbot) for commit
5c3e6d3. Bugbot is set up for automated
code reviews on this repo. Configure
[here](https://www.cursor.com/dashboard/bugbot).</sup>
<!-- /CURSOR_SUMMARY -->

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

Conflict resolutions:
- packages/iterate/src/cli.ts: took this branch's version wholesale.
  Main's only change was the superadmin->admin rename (#1453), and every
  renamed region (auth strategies, impersonation dance, device flow,
  --superadmin login flag) was already deleted here in the OAuth PKCE
  rewrite — there is nothing to port.
- apps/os/e2e/vitest/e2e-test-map.e2e.test.ts: accepted main's deletion
  (codemode rip #1447); this branch had only touched a doc comment.
- event-stream-terminal.tsx (auto-merged, needed follow-up): main's new
  listChildren itx walk gets the authedFetch arg, matching the getState
  adaptation from the previous merge.
- apps/iterate-com skills-registry.ts: took main's regenerated copy;
  this branch carried a stale-formatted version of the generated file.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>