Feat/harness wave0

.github/workflows/e2e.yml

@@ -0,0 +1,46 @@
+name: E2E Tests (Real LLM)
+
+on:
+  push:
+    branches:
+      - 'main'
+      - 'agent-diva-pro/**'
+  pull_request:
+    branches:
+      - 'main'
+      - 'agent-diva-pro/**'
+
+env:
+  CARGO_TERM_COLOR: always
+
+jobs:
+  e2e:
+    runs-on: ubuntu-latest
+    if: ${{ secrets.DEEPSEEK_API_KEY != '' }}
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install Rust stable
+        uses: actions-rust-lang/setup-rust-toolchain@v1
+        with:
+          toolchain: stable
+
+      - name: Install just
+        uses: taiki-e/install-action@v2
+        with:
+          tool: just
+
+      - name: Run E2E tests
+        env:
+          DEEPSEEK_API_KEY: ${{ secrets.DEEPSEEK_API_KEY }}
+        run: just e2e
+        working-directory: agent-diva-pro
+
+      - name: Upload trace artifacts
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: e2e-traces
+          path: agent-diva-pro/target/e2e-traces/
+          retention-days: 7

AGENTS.md

@@ -110,7 +110,9 @@ Write focused unit tests near the code with `#[cfg(test)]`. Add integration test
 
 ## Commit & Pull Request Guidelines
 
-Recent history follows Conventional Commit prefixes (`feat:`, `fix:`, `docs:`); keep using that style with concise imperative summaries. Before PRs, run `just ci`, describe behavioral impact, link related issues, and update docs when interfaces/channels/providers change. Keep PRs focused to a single concern for easier review.
+Recent history follows Conventional Commit prefixes (`feat:`, `fix:`, `docs:`); keep using that style with concise imperative summaries. Commit each completed, self-contained update automatically instead of accumulating a large dirty working tree. Do not push unless the user explicitly requests it. Before PRs, run `just ci`, describe behavioral impact, link related issues, and update docs when interfaces/channels/providers change. Keep PRs focused to a single concern for easier review.
+
+Before committing, clean up generated scratch artifacts, temporary scripts, stale archives, local test outputs, and other non-deliverable dirty-work files created during the task. Do not remove or revert unrelated user changes. Stage only the files for the current focused update, keep unrelated pre-existing changes out of the commit, and record deferred validation or follow-up items in the commit body/report and `TODOLIST.md` when applicable.
 
 **Recommended PR checklist:**
 
@@ -130,6 +132,20 @@ Recent history follows Conventional Commit prefixes (`feat:`, `fix:`, `docs:`);
   - `acceptance.md`: Acceptance steps from user/product perspective.
 - Optional documentation: `prd.md`, `notes.md` (discussion records), `rollback.md` (rollback plan).
 
+## TODOLIST Protocol
+
+- `TODOLIST.md` at the repository root is the canonical backlog for discovered bugs, gaps, deferred work, and unfinished implementation or UX items.
+- When an issue is found during code review, implementation, validation, or documentation work, add it to `TODOLIST.md` unless it is fixed in the same iteration.
+- Entries should include enough context to recover the issue later: status checkbox, short title, reason, expected behavior, and related files or docs when available.
+- When a TODO is completed, move or mark it under the done section instead of silently deleting it.
+
+## COMMIT Rule
+
+- Commits must use English Conventional Commit prefixes.
+- Each commit should cover one focused concern only; avoid mixing unrelated cleanup, docs, and feature work.
+- Before committing, ensure the changed set is clean of temporary artifacts and any non-deliverable scratch files.
+- For non-trivial commits, include a short validation note in the commit message body or accompanying report when applicable.
+
 ## Command Mechanism
 
 - New commands are recorded in `commands/commands.md` and maintain an index in this section (repository does not have this file — create it when adding the first command).
@@ -200,11 +216,11 @@ By default, all rules are mandatory; if exceptions are needed, they must be expl
   - Execution Method: Include "update command index" in change list and acceptance items.
   - Maintainer: Current assistant.
 
-- **no-self-commit-without-request**:
-  - Constraints/Range of applicability: Do not commit or push code without user's explicit request.
-  - Example: Commit only after the user explicitly says "help me commit."
-  - Counterexample: Commit code without authorization.
-  - Execution Method: Confirm explicit user instruction before committing.
+- **auto-commit-each-completed-update**:
+  - Constraints/Range of applicability: After each completed, self-contained update, automatically create one git commit for the files changed by that update. Do not push unless the user explicitly requests it.
+  - Example: After updating project rules in `AGENTS.md`, stage and commit only `AGENTS.md`; after finishing one focused bugfix, commit only that bugfix.
+  - Counterexample: Leave a completed update uncommitted, include unrelated pre-existing workspace changes in the commit, mix multiple concerns into one giant commit, or push without authorization.
+  - Execution Method: Before committing, inspect `git status --short --untracked-files=all`; clean or exclude current-task scratch artifacts; stage explicit paths for the current update only; verify the staged diff is one focused concern; run relevant validation when practical, or explicitly note deferred validation in the commit body/report; update `TODOLIST.md` for discovered but unfixed issues; commit with a concise English Conventional Commit message.
   - Maintainer: Current assistant.
 
 - **use-chinese-when-communicating**:
@@ -214,6 +230,15 @@ By default, all rules are mandatory; if exceptions are needed, they must be expl
   - Execution Method: Use unison Chinese output.
   - Maintainer: Current assistant.
 
+- **todolist-capture-required**:
+  - Constraints/Range of applicability: Any discovered bug, unfinished work, known limitation, or deferred improvement must be recorded in root `TODOLIST.md` unless it is completed in the same iteration.
+  - Example: Discover that GUI image paste is not implemented; add an open TODO with context and expected behavior.
+  - Counterexample: Mention a future fix in chat but leave no durable project backlog entry.
+  - Execution Method: Update `TODOLIST.md` before final response or commit; include related docs/files when available.
+  - Maintainer: Current assistant.
+
+
+
 ---
 
 ## Project Rulebook
@@ -247,4 +272,4 @@ By default, all rules are mandatory; if exceptions are needed, they must be expl
 - **Example**: `[I strictly follow the rules] Modification completed.`
 - **Counterexample**: Replies without the prefix or only partially including the prefix.
 - **Execution Method**: All outputs must include this prefix at the beginning.
-- **Maintainer**: Current assistant.
+- **Maintainer**: Current assistant.

Cargo.toml

@@ -9,6 +9,7 @@ members = [
     "agent-diva-channels",
     "agent-diva-tools",
     "agent-diva-files",
+    "agent-diva-e2e",
     "agent-diva-cli",
     "agent-diva-service",
     "agent-diva-migration",
@@ -74,6 +75,7 @@ regex = "1.10"
 lazy_static = "1.4"
 once_cell = "1.19"
 bytes = "1.5"
+base64 = "0.22"
 which = "6.0"
 windows-service = "0.7"