yoi/.yoi/tickets/open/20260601-021104-tui-composer-history-persistence/thread.md

Created

Created by tickets.sh create.

---

Decision

Updated based on user direction:

• keep this as the existing tui-composer-history-persistence ticket rather than creating a duplicate;
• default user-data shape should be like ~/.yoi/<path-to-pwd>/composer-history.json using a path-safe/stable workspace key;
• do not create composer history under workspace ./.yoi/;
• bound persisted recall history to about 30 entries per workspace instead of the older 100-entry note;
• keep typed Segment storage, non-destructive recall semantics, and no Pod/session transcript mutation.

---

Intake summary

既存 Ticket を確認し、同目的の open Ticket は他にないことを確認した。要件は、TUI composer の Up/Down recall 履歴を user data dir 配下に workspace-scoped で永続化し、workspace ./.yoi/、Pod protocol、transcript、Worker/session history、model context には影響させないこと。保存対象は typed Segment vector、non-blank、連続重複抑止、workspace ごと最大 30 件程度。破損ファイルは startup を致命的に壊さず fallback する。受け入れ基準と validation は Ticket 本文に十分記録済みで、Orchestrator が routing 可能なため ready とする。

---

State changed

Intake 確認により、既存 Ticket は実装 routing 可能な状態です。未解決の blocking open question はありません。主な reviewer focus は user-data persistence 境界、private/secret 混入前提の非露出、workspace 分離、既存 TUI recall semantics の回帰防止です。

---

State changed

Ticket を workspace-panel が queued にしました。

---

Decision

Routing decision: defer_parallel_tui_work

This Ticket is queued and implementation-ready, but I am intentionally not accepting it to inprogress yet.

Reason:

• There is already an active queued-route TUI implementation in progress: tui-chat-markdown-table-rendering, with Coder Pod coder-tui-chat-markdown-table-rendering running in .worktree/tui-chat-markdown-table-rendering.
• tui-composer-history-persistence also touches TUI composer/input behavior and likely adjacent tests/helpers.
• Running both TUI changes concurrently would raise conflict/review risk without a clear need; this Ticket can follow immediately after the table-rendering Ticket reaches merge-ready or lands.

State decision:

• Leave this Ticket queued.
• Do not create a worktree or spawn a coder for it yet.

Next action:

• Re-read this Ticket after tui-chat-markdown-table-rendering finishes review/merge routing, then accept it from queued if still unblocked.

---

State changed

Accepted queued implementation after re-reading the Ticket and current workspace state. The previously active TUI table-rendering work has been reviewed, merged, closed, and cleaned up, so the earlier parallel-TUI-work deferral no longer applies.

---

Decision

Routing decision: implementation_ready

Reason:

• This Ticket has clear storage, privacy, workspace-separation, and recall-semantics requirements.
• The earlier deferral behind tui-chat-markdown-table-rendering is resolved: that Ticket has been merged, closed, and cleaned up.
• Main workspace is clean and there are no active implementation worktrees, so implementation side effects can proceed safely.

IntentPacket:

Intent:

• Persist TUI composer input recall history per workspace in user data, while preserving the existing TUI-local/non-destructive recall semantics.

Binding decisions / invariants:

• Store composer history under the existing user data dir, not under workspace ./.yoi/.
• Workspace histories must be separated by a stable/path-safe workspace key and include metadata sufficient to identify the source workspace/display name.
• Store typed Segment vectors, not lossy flattened strings, so structured input recall remains possible.
• Persist only non-blank entries, suppress consecutive duplicates, and bound stored history to about 30 entries per workspace.
• Recalled input must not affect conversation state until the user explicitly submits it.
• Do not mutate Pod protocol, worker history, transcript/session logs, model context, or stored message history for recall operations.
• Treat history content as potentially private/secret: avoid diagnostics/logs/snapshots/tickets/model context exposure of saved entries.
• Corrupt history files must not make TUI startup fatal; fall back to empty history with a bounded warning if needed.
• Preserve existing multiline cursor navigation, Up/Down browse, draft restore, and edit-on-recall behavior.

Implementation latitude:

• Coder may choose the exact on-disk user-data path/key encoding if it is stable, path-safe, data-dir aware, and workspace-scoped.
• Coder may add a small persistence module near existing TUI composer history code or shared user-data helpers if present.
• Coder should prefer focused unit tests with temporary data dirs/workspaces for persistence, separation, bounding, duplicate suppression, corrupt-file fallback, and no workspace-file creation.
• If interactive confirmation is still useful, include exact manual steps in the implementation report, but do not rely on manual-only validation if focused tests are practical.

Escalate if:

• The implementation requires changing Pod protocol/history/session serialization.
• The implementation would store history under workspace .yoi/.
• Existing composer history representation cannot preserve typed Segment vectors without a larger input model redesign.
• Secret/private content would be emitted into logs/diagnostics/tests/tickets/model context.

Validation:

• Focused TUI composer history persistence tests for same-workspace recall after reload, workspace separation, max-size bound, consecutive duplicate suppression, corrupt-file fallback, and no workspace .yoi output.
• Existing TUI composer history/navigation tests if present.
• cargo fmt --check.
• git diff --check.
• cargo check --workspace.
• nix build .#yoi before final completion because TUI behavior/packaging changes.

---