0.1.0a11

Release 0.1.0a11.

An ENC:v3 fix release. Every fix here was found by verifying the browser
extension's new ENC:v3 support against a real install (Slack #my-hermes and
Discord #general) rather than by unit testing — on a real install, no v3
command could authenticate at all before this.

Fixes

• Resolve a v3 channel key stored under the extension's composite id, so
  existing installs work with no state migration. The extension pushes
  channel_key_set keyed by slack:{team}:{cid} / discord:{guild}:{cid}
  and save_channel_key stores that verbatim, while a gateway event only
  carries the platform's native chat_id — so the exact-match lookup missed
  every stored key and refused the command as key_not_bound_to_channel. The
  composite is accepted only when its last segment is exactly the channel
  id and its first segment is the event's platform, so the binding
  stays as tight as the exact match it replaces.
• Log the fixed InvalidEncryptedEnvelope reason identifier on refusal (never
  message text, keys or plaintext) — every refusal previously logged the same
  sentence while needing completely different fixes, which made a live
  rejection undiagnosable without replaying the wire by hand.
• Authenticate a top-level Slack command under its real context. The Slack
  adapter's default reply_in_thread session keying stamps thread_id with a
  synthetic root equal to a top-level message's own ts; Slack assigns that
  ts only after the send, so the extension provably encrypted with
  thread_root=None and every genuine top-level v3 command was refused as
  authentication_failed. The command AAD now canonicalizes to
  thread_root=None when the routed root IS the command's own message_id
  (mirror of the documented Discord auto-thread canonicalization), while reply
  routing keeps the synthetic thread. Canonicalizing requires two
  adapter-stamped markers to agree, so a captured top-level token still cannot
  be replayed into a genuine thread.

Changes

• Bump the pinned version surfaces to 0.1.0a11.
• Document the Slack synthetic-thread-root canonicalization in
  docs/dev/SLACK_E2E.md §2.

Test plan

• uv run pytest — full suite green; 11 new tests in
  tests/extension/test_extension_e2e_v3.py (composite-id resolution,
  platform-mismatch and partial-suffix refusals, synthetic-root top-level
  acceptance, genuine-thread acceptance, top-level-token-in-thread
  refusal, thread-token-at-top-level refusal, non-Slack no-canonicalize,
  degenerate message_id fail-closed, replay through the dispatch hook,
  reply_in_thread=false parity)
• ruff check / ruff format --check / mypy --strict src tools
• Adversarial review (independent agent): verdict SOUND, no
  CRITICAL/HIGH/MEDIUM findings; its two hardening suggestions and five
  coverage gaps were all landed
• Live gateway round trip on the production install (2026-08-01): a
  top-level ENC:v3 command from the real Slack client decrypted by the
  gateway, agent reply re-encrypted in-thread with the inbound kid; same
  round trip verified on Discord including the auto-thread path
• TestPyPI dry run + fresh-venv discovery check (release workflow)

Pairs with mordredagent/mordred-extension#11 and #12 — the extension
side of the same wire format.