Plan mode rewrites prompt history and breaks prefix caching

OpenCode 1.18.5 rewrites model-visible history when resuming a plan session.

Captured two consecutive OpenAI-compatible requests:

The immediate follow-up got 0 cached tokens and re-prefilled all 45,222 tokens in 143.3s.

Historical messages remain byte-identical. Only the new turn is appended.

macOS, no plugins, custom OpenAI-compatible provider.

This issue might be a duplicate of or closely related to existing issues. Please check:

[FEATURE]:Performance issue: Context cache invalidation when switching modes or during compaction #37489: Performance issue — context cache invalidation when switching modes or during compaction (describes the same Plan/Build mode tool-list change and history modification invalidating the cache prefix)
[BUG] Prompt caching breaks when a plugin appends a message #43507: Prompt caching breaks when a plugin appends a message (same general mechanism — history instability between turns causing full cache misses)
Compaction reuses the session's prompt_cache_key while sending a maximally divergent prefix — a guaranteed full cache write that can never be read #43249: Compaction reuses the session's prompt_cache_key with a maximally divergent prefix (related cache-key misuse during compaction)

(#26749 and #32202 are already referenced in your report.)

Thanks for the report! OpenCode 2.0 is now out and development is focused on V2, so we're closing issues filed against OpenCode 1. Please upgrade (opencode upgrade --major, details: https://anoma.ly/notes/opencode-2-0), and if this is still relevant on 2.0, reopen this issue or open a new one with your 2.0 version.

— from 𝕺𝖕𝖊𝖓𝕮𝖔𝖉𝖊

Navigation Menu