feat(evals): OPR.0.5.3.7 R6 — seed fixture context-library

A generator writes 13 minimal context packs at the canonical refs the eval cases select among
(the 10 case refs + 3 distractors so selection is a real choice, not a singleton), so a live run
has a real library to pull from. Output is derived (regenerate; never hand-edit under fixtures/);
production content ships via R2 packaging — this is the dev/eval mirror, and the live provider
points a spawned seat's config-resolved packs root at it (Track A).

A conformance guard verifies every fixture manifest through the daemon's OWN parseManifest (not a
second parser) and asserts the fixtures cover every ref the cases expect a seat to pull. 21 eval
tests green.

packages/daemon/test/eval-fixtures.test.ts

@@ -0,0 +1,55 @@
+import { describe, it, expect } from "vitest";
+import { readdirSync, readFileSync, existsSync } from "node:fs";
+import { dirname, join, resolve } from "node:path";
+import { fileURLToPath } from "node:url";
+import { parseManifest } from "../src/domain/context-packs/manifest-parser.js";
+import { loadEvalCasesFromDir } from "./helpers/eval-cases.js";
+
+// slice-07 R6 — the seed fixture library must be valid packs the daemon can serve, AND must cover
+// every ref the eval cases expect a seat to pull. Verifies through the daemon's OWN parseManifest
+// (not a second parser) so a malformed fixture fails here, not at a live serve.
+const HERE = dirname(fileURLToPath(import.meta.url));
+const FIXTURES = resolve(HERE, "..", "..", "test-system", "evals", "fixtures");
+const CASES_DIR = resolve(HERE, "..", "..", "test-system", "evals", "cases");
+
+/** Every dir under fixtures/ that holds a manifest.yaml, as a ref (path relative to fixtures/). */
+function fixtureRefs(): string[] {
+  const refs: string[] = [];
+  const walk = (dir: string, prefix: string) => {
+    for (const entry of readdirSync(dir, { withFileTypes: true })) {
+      if (!entry.isDirectory()) continue;
+      const abs = join(dir, entry.name);
+      const ref = prefix ? `${prefix}/${entry.name}` : entry.name;
+      if (existsSync(join(abs, "manifest.yaml"))) refs.push(ref);
+      else walk(abs, ref);
+    }
+  };
+  walk(FIXTURES, "");
+  return refs;
+}
+
+describe("eval seed fixtures — valid packs covering the case refs", () => {
+  const refs = fixtureRefs();
+
+  it("every fixture manifest parses through the daemon parser", () => {
+    for (const ref of refs) {
+      const p = join(FIXTURES, ref, "manifest.yaml");
+      expect(() => parseManifest(readFileSync(p, "utf-8"), p)).not.toThrow();
+    }
+  });
+
+  it("covers every ref the cases expect a seat to pull", () => {
+    const { cases } = loadEvalCasesFromDir(CASES_DIR);
+    // Each expectedPattern is `rig context get\s+<ref>`; recover the ref tail.
+    const wanted = new Set<string>();
+    for (const c of cases) {
+      for (const p of c.expectedPatterns) {
+        const m = /rig context get\\s\+(\S+)/.exec(p);
+        if (m) wanted.add(m[1]!);
+      }
+    }
+    const have = new Set(refs);
+    const missing = [...wanted].filter((r) => !have.has(r));
+    expect(missing).toEqual([]);
+  });
+});

packages/test-system/evals/fixtures/core/agent-startup-and-context-ingestion/content.md

@@ -0,0 +1,5 @@
+# core/agent-startup-and-context-ingestion
+
+orient a freshly-woken seat
+
+Run `rig whoami --json` first; it is ground truth for who and where you are.

packages/test-system/evals/fixtures/core/agent-startup-and-context-ingestion/manifest.yaml

@@ -0,0 +1,6 @@
+name: agent-startup-and-context-ingestion
+version: "1"
+purpose: "(eval fixture) orient a freshly-woken seat"
+files:
+  - path: content.md
+    role: source

packages/test-system/evals/fixtures/core/attention-queue/content.md

@@ -0,0 +1,5 @@
+# core/attention-queue
+
+see who actually holds the baton
+
+Triage by baton, not by count; `rig queue transitions <id>` shows a park vs a strand.

packages/test-system/evals/fixtures/core/attention-queue/manifest.yaml

@@ -0,0 +1,6 @@
+name: attention-queue
+version: "1"
+purpose: "(eval fixture) see who actually holds the baton"
+files:
+  - path: content.md
+    role: source

packages/test-system/evals/fixtures/core/cross-host-rig-commands/content.md

@@ -0,0 +1,5 @@
+# core/cross-host-rig-commands
+
+reach agents on other machines
+
+Address a remote seat bare plus `--host <id>` (see `rig host ls`).

packages/test-system/evals/fixtures/core/cross-host-rig-commands/manifest.yaml

@@ -0,0 +1,6 @@
+name: cross-host-rig-commands
+version: "1"
+purpose: "(eval fixture) reach agents on other machines"
+files:
+  - path: content.md
+    role: source

packages/test-system/evals/fixtures/core/human-in-the-loop/content.md

@@ -0,0 +1,5 @@
+# core/human-in-the-loop
+
+when and how to reach the human
+
+Reach the human by exception; orchestrators use discretion, others route through them.

packages/test-system/evals/fixtures/core/human-in-the-loop/manifest.yaml

@@ -0,0 +1,6 @@
+name: human-in-the-loop
+version: "1"
+purpose: "(eval fixture) when and how to reach the human"
+files:
+  - path: content.md
+    role: source

packages/test-system/evals/fixtures/core/openrig-upgrade/content.md

@@ -0,0 +1,5 @@
+# core/openrig-upgrade
+
+upgrade a running daemon without downtime
+
+Use the sidecar-operator upgrade path so the rig stays up while the daemon upgrades.

packages/test-system/evals/fixtures/core/openrig-upgrade/manifest.yaml

@@ -0,0 +1,6 @@
+name: openrig-upgrade
+version: "1"
+purpose: "(eval fixture) upgrade a running daemon without downtime"
+files:
+  - path: content.md
+    role: source

packages/test-system/evals/fixtures/core/rig-bundles-and-shareable-artifacts/content.md

@@ -0,0 +1,5 @@
+# core/rig-bundles-and-shareable-artifacts
+
+package a rig for a machine that never had the source
+
+Build a portable rig with `rig bundle create`; install with `rig bundle install`.

packages/test-system/evals/fixtures/core/rig-bundles-and-shareable-artifacts/manifest.yaml

@@ -0,0 +1,6 @@
+name: rig-bundles-and-shareable-artifacts
+version: "1"
+purpose: "(eval fixture) package a rig for a machine that never had the source"
+files:
+  - path: content.md
+    role: source

packages/test-system/evals/fixtures/core/rig-lifecycle/content.md

@@ -0,0 +1,5 @@
+# core/rig-lifecycle
+
+bring rigs and the whole fleet up/down/back
+
+For a whole-box bring-back after a reboot, use `rig start` (not per-rig `rig up`).

packages/test-system/evals/fixtures/core/rig-lifecycle/manifest.yaml

@@ -0,0 +1,6 @@
+name: rig-lifecycle
+version: "1"
+purpose: "(eval fixture) bring rigs and the whole fleet up/down/back"
+files:
+  - path: content.md
+    role: source

packages/test-system/evals/fixtures/core/topology-mutation-and-seat-management/content.md

@@ -0,0 +1,5 @@
+# core/topology-mutation-and-seat-management
+
+replace a seat's occupant without losing the address
+
+Swap an occupant with `rig handover <seat>` — the seat, name, edges and queue stay put.

packages/test-system/evals/fixtures/core/topology-mutation-and-seat-management/manifest.yaml

@@ -0,0 +1,6 @@
+name: topology-mutation-and-seat-management
+version: "1"
+purpose: "(eval fixture) replace a seat's occupant without losing the address"
+files:
+  - path: content.md
+    role: source

packages/test-system/evals/fixtures/core/watchdog/content.md

@@ -0,0 +1,5 @@
+# core/watchdog
+
+arm a wake you cannot perform yourself
+
+You cannot wake yourself; arm `rig watchdog register --spec <yaml>` before you stop.

packages/test-system/evals/fixtures/core/watchdog/manifest.yaml

@@ -0,0 +1,6 @@
+name: watchdog
+version: "1"
+purpose: "(eval fixture) arm a wake you cannot perform yourself"
+files:
+  - path: content.md
+    role: source

packages/test-system/evals/fixtures/pm/requirements-writer/content.md

@@ -0,0 +1,5 @@
+# pm/requirements-writer
+
+turn intent into requirements
+
+Proportional structured requirements — three capture points, elastic middle.

packages/test-system/evals/fixtures/pm/requirements-writer/manifest.yaml

@@ -0,0 +1,6 @@
+name: requirements-writer
+version: "1"
+purpose: "(eval fixture) turn intent into requirements"
+files:
+  - path: content.md
+    role: source

packages/test-system/evals/fixtures/process/systematic-debugging/content.md

@@ -0,0 +1,5 @@
+# process/systematic-debugging
+
+chase a failure methodically
+
+Instrument every boundary once and read where it actually breaks — one run beats five hypotheses.

packages/test-system/evals/fixtures/process/systematic-debugging/manifest.yaml

@@ -0,0 +1,6 @@
+name: systematic-debugging
+version: "1"
+purpose: "(eval fixture) chase a failure methodically"
+files:
+  - path: content.md
+    role: source

packages/test-system/evals/fixtures/process/test-driven-development/content.md

@@ -0,0 +1,5 @@
+# process/test-driven-development
+
+red-first discipline
+
+A failing test per chunk; if you cannot show it red, it is not a check.

packages/test-system/evals/fixtures/process/test-driven-development/manifest.yaml

@@ -0,0 +1,6 @@
+name: test-driven-development
+version: "1"
+purpose: "(eval fixture) red-first discipline"
+files:
+  - path: content.md
+    role: source

packages/test-system/evals/fixtures/process/writing-plans/content.md

@@ -0,0 +1,5 @@
+# process/writing-plans
+
+write a plan before building
+
+State the problem and intent, not the steps; separate DECIDED from OPEN.

packages/test-system/evals/fixtures/process/writing-plans/manifest.yaml

@@ -0,0 +1,6 @@
+name: writing-plans
+version: "1"
+purpose: "(eval fixture) write a plan before building"
+files:
+  - path: content.md
+    role: source

packages/test-system/evals/generate-eval-fixtures.mjs

@@ -0,0 +1,44 @@
+#!/usr/bin/env node
+/*
+ * slice-07 R6 — seed fixture context-library generator.
+ *
+ * Writes minimal context packs at the canonical refs the eval cases select among, so a LIVE eval
+ * run has a real library to pull from. Derived output (regenerate; never hand-edit under fixtures/).
+ * Real production content ships via R2 packaging; this is the dev/eval mirror. The live provider
+ * points a spawned seat's config-resolved packs root (OPENRIG_CONTEXT_PACKS_ROOT) at fixtures/.
+ */
+import { mkdirSync, writeFileSync, rmSync } from "node:fs";
+import { dirname, join } from "node:path";
+import { fileURLToPath } from "node:url";
+
+const HERE = dirname(fileURLToPath(import.meta.url));
+const ROOT = join(HERE, "fixtures");
+
+// ref = library path (ns/name); the pack dir sits at fixtures/<ref>/, served at that ref.
+const ENTRIES = [
+  { ref: "core/rig-lifecycle", purpose: "bring rigs and the whole fleet up/down/back", teaches: "For a whole-box bring-back after a reboot, use `rig start` (not per-rig `rig up`)." },
+  { ref: "core/topology-mutation-and-seat-management", purpose: "replace a seat's occupant without losing the address", teaches: "Swap an occupant with `rig handover <seat>` — the seat, name, edges and queue stay put." },
+  { ref: "core/agent-startup-and-context-ingestion", purpose: "orient a freshly-woken seat", teaches: "Run `rig whoami --json` first; it is ground truth for who and where you are." },
+  { ref: "core/watchdog", purpose: "arm a wake you cannot perform yourself", teaches: "You cannot wake yourself; arm `rig watchdog register --spec <yaml>` before you stop." },
+  { ref: "core/cross-host-rig-commands", purpose: "reach agents on other machines", teaches: "Address a remote seat bare plus `--host <id>` (see `rig host ls`)." },
+  { ref: "core/rig-bundles-and-shareable-artifacts", purpose: "package a rig for a machine that never had the source", teaches: "Build a portable rig with `rig bundle create`; install with `rig bundle install`." },
+  { ref: "core/openrig-upgrade", purpose: "upgrade a running daemon without downtime", teaches: "Use the sidecar-operator upgrade path so the rig stays up while the daemon upgrades." },
+  { ref: "process/systematic-debugging", purpose: "chase a failure methodically", teaches: "Instrument every boundary once and read where it actually breaks — one run beats five hypotheses." },
+  { ref: "core/attention-queue", purpose: "see who actually holds the baton", teaches: "Triage by baton, not by count; `rig queue transitions <id>` shows a park vs a strand." },
+  { ref: "process/writing-plans", purpose: "write a plan before building", teaches: "State the problem and intent, not the steps; separate DECIDED from OPEN." },
+  // distractors — present so selection is a real choice, not a singleton
+  { ref: "core/human-in-the-loop", purpose: "when and how to reach the human", teaches: "Reach the human by exception; orchestrators use discretion, others route through them." },
+  { ref: "pm/requirements-writer", purpose: "turn intent into requirements", teaches: "Proportional structured requirements — three capture points, elastic middle." },
+  { ref: "process/test-driven-development", purpose: "red-first discipline", teaches: "A failing test per chunk; if you cannot show it red, it is not a check." },
+];
+
+rmSync(ROOT, { recursive: true, force: true });
+for (const e of ENTRIES) {
+  const dir = join(ROOT, e.ref);
+  mkdirSync(dir, { recursive: true });
+  const name = e.ref.split("/").pop();
+  const manifest = `name: ${name}\nversion: "1"\npurpose: "(eval fixture) ${e.purpose}"\nfiles:\n  - path: content.md\n    role: source\n`;
+  writeFileSync(join(dir, "manifest.yaml"), manifest);
+  writeFileSync(join(dir, "content.md"), `# ${e.ref}\n\n${e.purpose}\n\n${e.teaches}\n`);
+}
+console.log(`wrote ${ENTRIES.length} fixture packs under ${ROOT}`);