Audit token and prompt-cache diagnostics parity

[shiny-code-bot]

Finish Line

Token, rate-limit, and prompt-cache diagnostics promised by Every Code are audited against the Codex-fork token pipeline and protected by fixtures before UI changes.

Current Status

State: Created from #397 parity ledger.

Deleted evidence from the pre-#390 tree includes old token/account usage modules, limits overlay/view code, rate-limit history cells, and limits snapshots. Current main has Codex token usage replay, app-server token events, login token data, and TUI token usage, but prompt-cache hit-rate parity is not proven.

Acceptance Criteria

• Audit current ThreadTokenUsageUpdatedNotification and token replay behavior against Every Code docs/release promises.
• Classify old token/rate-limit/limits snapshots as port, rewrite, covered, defer, or retire.
• Add missing fixtures for prompt-cache hit-rate and rate-limit display semantics before UI changes.
• Do not change token event schemas or replay processing until the prompt-cache gap is inventoried.
• End code-bearing slices with ./build-fast.sh passing cleanly.

Relationships

• Child of Stabilize Every Code parity on Codex CLI fork baseline #397.
• Related to Plan Every Code token efficiency and local LLM sandboxing #43 token efficiency work.

[cbusillo]

Landed first token diagnostics parity slice in #403 (merge commit 6ff1ffa).

Evidence:

• Restored prompt-cache hit-rate display in /status when cached input tokens are non-zero.
• Added configurable footer status-line item: prompt-cache-hit-rate.
• Added docs/token-diagnostics-parity.md and updated the parity ledger row.
• Narrowed release-note wording so it no longer claims telemetry-present schema semantics or rolling average support.

Validation:

• cargo test -p codex-tui prompt_cache
• INSTA_UPDATE=always cargo test -p codex-tui status_snapshot
• ./build-fast.sh passed; note it printed the existing unlocked-build/Cargo.lock warning but completed successfully.
• Blob size policy passed on PR feat: restore prompt-cache hit-rate diagnostics #403.

Remaining #401 gaps:

• schema/provider path still collapses absent cached telemetry and true 0 cached input into cached_input_tokens = 0.
• no rolling prompt-cache average restored yet.
• old /limits overlay and local account-usage history remain separate parity decisions.