P0-1 Reopened — Design Question, Not a 1-File Fix

Date: 2026-04-23 Context: coordinator attempted to land P0-1 (TC-S-17 duplicate-path rejection) solo per user's standing "continue推进" instruction. Stopped before touching code because a product-behavior decision is needed.

The conflict

The post-pilot list P0-1 frames this as a "small-diff fix": backend's POST /api/skills/external-paths should reject duplicate path with a 4xx so the renderer's handleAddCustomPath keeps the Add-Path modal open.

Inspection of aionui-backend/crates/aionui-extension/src/external_paths.rs reveals the current Rust implementation is designed as upsert:

• Line 66 doc: "If a path with the same value already exists, it is updated with the new name."
• Function body (lines 75-82): find(|p| p.path == path) → update name; else append.
• Existing test add_duplicate_path_updates_name (line 194-209) asserts the upsert behavior.
• enable_skills_market (line 104) depends on upsert for its idempotency: calling enable twice should not grow the list. Test enable_market_idempotent (line 301) relies on this.

So reject-on-duplicate cannot be a blanket behavior change — it would break the skills-market enable flow.

The product question

What SHOULD the contract be?

Option A — Preserve upsert, make e2e reflect reality.

• Keep add_custom_external_path as upsert.
• Update TC-S-17 to expect the modal to close and the entry's name to update silently. UI changes: none.
• Revise the post-pilot P0-1 framing: there is no migration regression; the TS baseline likely also upserted, and the e2e test was authored against an incorrect assumption.
• Cheapest. Treats "duplicate path" as a legitimate user intent to rename.

Option B — Reject on duplicate for user-driven adds, keep upsert for internal callers.

• Split the function: add_custom_external_path_strict(name, path) → returns DuplicatePath error; kept existing add_custom_external_path for enable_skills_market internal use.
• HTTP handler at POST /api/skills/external-paths calls the strict form. Market enable continues to use the upsert form.
• Add ExtensionError::DuplicatePath(String) and map to StatusCode::CONFLICT (409).
• Renderer's handleAddCustomPath already catches errors and leaves modal open — no renderer change beyond ensuring the toast message is user-friendly for 409.
• TC-S-17 passes as currently written.
• Costs: ~30 lines of Rust + 2 tests + a thin renderer toast update.

Option C — Reject globally, refactor market enable to check-first.

• Make the only add_custom_external_path reject duplicates.
• Rewrite enable_skills_market to if !already_enabled { add... }.
• Breaks the existing enable_market_idempotent test assertion that only 1 entry exists after 2 enable calls — actually it still passes (check-first also yields 1 entry). Let me re-read... yes, it still passes with the check-first pattern.
• More invasive than B, no clear benefit over B.

My recommendation (if forced to pick): Option B

• Preserves existing internal callers' semantics without special casing.
• Matches the user's product intuition that "add duplicate path" is probably an error from the UI's perspective.
• TC-S-17 was authored by the e2e-coverage team and presumably reflects the intended user contract; honoring it is the default stance.
• Smallest blast radius.

Why I stopped instead of landing Option B

Three reasons:

1. The post-pilot list framed this as "1-file small fix" — that framing was wrong. The actual fix touches backend (new function, new error, error mapping, 2 tests) AND a design call on whether to split the function. That's a decision, not a routine.

2. The current Rust code has explicit documentation and a test that say upsert is intentional. Overriding an intentional design without user confirmation violates the "investigate before deleting / overwriting" safety principle.

3. The user signed off for the day and specifically said "明天我希望 收到成果". They got the pilot and the Assistant verification — the significant deliverables. Reopening a product decision and committing code alone before they wake risks pushing a direction they didn't approve.

What needs decision from user

Pick A / B / C above (or a 4th option), then re-spawn a small backend-dev

• frontend-dev team to land it (~1 hour) and e2e-tester to confirm TC-S-17 flips to PASS.

Pointers

• Code: aionui-backend/crates/aionui-extension/src/external_paths.rs
• Error def: aionui-backend/crates/aionui-extension/src/error.rs
• Renderer: src/renderer/pages/settings/SkillsHubSettings.tsx:209-223 (handleAddCustomPath)
• Test case to flip: tests/e2e/features/settings/skills/edge-cases.e2e.ts (or wherever TC-S-17 lives — grep TC-S-17 under tests/e2e/)
• Upstream reference: post-pilot list §P0-1.