chore(deps): bump ollama/ollama from 0.21.2 to 0.23.0 in the docker-compose group

CHANGELOG.md

@@ -9,6 +9,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [Unreleased]
 
 ### Changed
+- **Bump docker-compose group: ollama/ollama 0.21.2→0.23.0** (#411)
 - **CI badge URL switched from modern `actions/workflows/ci.yml/badge.svg` form to legacy `/workflows/CI/badge.svg`** (the `badge_url` GitHub itself reports for our workflows). The modern form rendered "no status" because it implicitly uses the same broken filter described in the Fixed entry below; the legacy form correctly returns "passing." Closes part of [#406](https://github.com/cmeans/mcp-awareness/issues/406).
 - **Cross-linked intention-firing semantics across the `awareness://briefing` resource and the `get_briefing`, `get_intentions`, and `remind` tool docstrings.** Each surface was previously documented in isolation, leaving an LLM (or human) reading them cold unable to determine the cross-tool relationship — most commonly producing a redundant `get_briefing` followed by `get_intentions(state="fired")`, even though `get_briefing` already both fires any pending intentions whose `deliver_at` has passed and surfaces currently-fired intentions inline. Same shape of discoverability gap as the one closed by [#399](https://github.com/cmeans/mcp-awareness/pull/399) for `remind`'s three `deliver_at` modes — correct behavior already in place; just not visible from the documented surface alone. The `awareness://briefing` resource (which the server's own `instructions.md` directs agents to read at conversation start, and which Claude Desktop reads instead of the tool path) gets the same firing language as `get_briefing` — both call the same `generate_briefing()` and fire identically; `get_briefing` now mentions that it fires due intentions and surfaces fired ones inline (and that a follow-up `get_intentions(state="fired")` is redundant); `get_intentions` positions itself as the drill-in for non-fired states or filtered queries and notes that pending intentions without `deliver_at` are not auto-fired by briefing; `remind` adds a reader-side cross-link making the one-call workflow explicit. Docs-only — no code, tests, or schema changes; behavior identical to v0.18.4. Closes [#404](https://github.com/cmeans/mcp-awareness/issues/404). Surfaced by Claude Desktop during a session walking through an inbound/outbound handoff workflow (awareness logical_key `tool-desc-gap-intention-firing-semantics`).
 

docker-compose.qa.yaml

@@ -74,7 +74,7 @@ services:
       start_period: 10s
 
   ollama-qa:
-    image: ollama/ollama:0.21.2
+    image: ollama/ollama:0.23.0
     container_name: awareness-ollama-qa
     restart: unless-stopped
     ports: