feat(swift): add SummarizingChatStorage for automatic history compression

[cornelcroi]

Issue Link

Closes #584

Summary

• Adds SummarizingChatStorage, a ChatStorage wrapper that automatically compresses conversation history when it exceeds a configurable threshold
• Compression is triggered when message count exceeds triggerAt * 2; user supplies a ChatSummarizer closure (@Sendable, async throws) that receives the full history and keepLast
• Compressed result is written back to the base store via saveMessages (replace semantics) and cached in a private SummarizationCache actor
• fetchAllChats is never intercepted — classifier sees raw history
• Cache is invalidated on any save or saveMessages call
• 13 tests added, all actor-safe (Swift 6 strict concurrency)

Changes

• swift/Sources/AgentSquad/Storage/SummarizingChatStorage.swift — new file
• swift/Tests/AgentSquadTests/SummarizingChatStorageTests.swift — 13 tests

User experience

let storage = SummarizingChatStorage(
    wrapping: InMemoryChatStorage(),
    summarizer: { history, keepLast in
        // call an LLM, return compressed [ConversationMessage]
    },
    triggerAt: 20,
    keepLast: 3
)
let squad = AgentSquad(storage: storage)

Checklist

• Tests added/updated
• Backward compatible (new struct, no existing API changed)
• Swift 6 strict concurrency compliant
• No new dependencies

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b6ab5685bc

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]

Replace history instead of appending summaries

When this wrapper is used with any bundled store (FileChatStorage, InMemoryChatStorage, or DeviceChatStorage), saveMessages is append-only rather than replacement. After the cache is invalidated or the app restarts, fetch reads the original long history plus the summary/recent messages, so the history is not actually compressed and can grow with duplicated content. This needs a true replace/delete path or a storage API that supports replacement.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Trim cached fetches to the requested maxMessages

Once a compressed result is cached, later fetch calls return it before applying the caller's maxMessages. If one caller caches a larger compressed history and another asks for a tighter window, this violates the ChatStorage.fetch contract that all bundled stores honor by trimming to the requested budget. Return a trimmed view of the cached messages or otherwise account for maxMessages in the cache path.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Use a collision-free cache key

Because the cache key concatenates raw IDs with #, distinct scopes collide whenever any ID contains that separator, e.g. (userId: "a#b", sessionId: "c", agentId: "d") and (userId: "a", sessionId: "b#c", agentId: "d"). After one scope is summarized, the other can receive its cached messages without reading the base store, leaking or mixing chat history across users/sessions/agents. Use a structured key type or escaping rather than raw concatenation.

Useful? React with 👍 / 👎.