fix(#2072): thread resolved model into routed-agent spawns (assumptions-analyzer, code-reviewer, code-fixer)

model_overrides / models.<phaseType> were silently inert for gsd-assumptions-analyzer, gsd-code-reviewer, and gsd-code-fixer on Claude Code: resolveModelInternal honors them, but the workflows spawned these agents with no model= param, so the resolved value never reached the Agent tool and the agents inherited the session model — no warning. Fix — thread each agent's resolved model at every spawn site (the established plan-phase pattern; the architecture-consistent Claude mechanism, since 13 other agents already thread their model): - discuss-phase-assumptions.md: `resolve-model gsd-assumptions-analyzer --raw` → ANALYZER_MODEL, threaded. - code-review.md + code-review-fix.md (re-review): `resolve-model gsd-code-reviewer --raw` → REVIEWER_MODEL, threaded. - code-review-fix.md (both fixer spawns): `resolve-model gsd-code-fixer --raw` → FIXER_MODEL, threaded (same silently-inert bug, same file — folded in per review). - quick.md review step: was reusing `{executor_model}` for gsd-code-reviewer (so the reviewer's own override was ignored); init.quick now resolves `reviewer_model` (gsd-code-reviewer) and the spawn threads it. resolve-model --raw returns the bare model string (resolve-execution --raw would return effort — wrong). The resolver maps these agents to phaseType discuss / verification / execution, so models.<phaseType> apply too. Scope: the three agents reachable from the two issue-named workflows + quick.md. The wider systemic class (other agents in UNTOUCHED workflows with the same pattern) stays documented on the issue for a maintainer-scoped structural decision (thread-at-source vs embed-at-install like #2256), not widened here. Docs: the stale "discuss — reserved, no subagent today" model-profile tables now list gsd-assumptions-analyzer and the verification row includes gsd-code-reviewer, across the English docs, the shipped gsd-core/references/model-profiles.md reference, and the ja-JP / zh-CN / ko-KR / pt-BR locale mirrors. Tests: - tests/model-resolver.test.cjs: #2072 acceptance — model_overrides and models.discuss/verification/execution resolve for all three agents. - tests/model-routing-spawn-threading.test.cjs: every spawn of the three agents threads a resolved model (fails pre-fix); a header-precise parity guard fails the suite if a new un-threaded spawn of any of them regresses. All 16 golden-install-parity fixtures + the workflow size baseline regenerated for the changed shipped files (4 workflows + the reference doc); bin/lib is excluded from parity. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-07 20:28:15 -04:00
parent cce405c9b7
commit 4483300253
37 changed files with 318 additions and 115 deletions
--- a/tests/model-routing-spawn-threading.test.cjs
+++ b/tests/model-routing-spawn-threading.test.cjs
@@ -0,0 +1,150 @@
+// allow-test-rule: source-text-is-the-product #2072
+// Workflow .md files ARE the deployed orchestration contract the runtime executes;
+// asserting that a spawn threads a resolved model= is asserting the deployed contract.
+
+'use strict';
+
+/**
+ * #2072 — model_overrides / models.<phaseType> were silently inert for
+ * gsd-assumptions-analyzer and gsd-code-reviewer on Claude: the resolver honored
+ * them, but the workflows spawned the agents with NO model= param, so the resolved
+ * value never reached the Agent tool and the agents inherited the session model.
+ *
+ * Fix contract: every spawn of these two catalog agents must thread a resolved
+ * model=, and the workflow must obtain that model (inline `resolve-model` for the
+ * single-agent workflows, or the `reviewer_model` field of the init.quick bundle).
+ */
+
+const { test, describe } = require('node:test');
+const assert = require('node:assert/strict');
+const fs = require('node:fs');
+const path = require('node:path');
+
+const WF = path.join(__dirname, '..', 'gsd-core', 'workflows');
+const read = (rel) => fs.readFileSync(path.join(WF, rel), 'utf-8');
+
+// Every .md under gsd-core/workflows (incl. nested steps/ and modes/).
+function allWorkflowMd() {
+  const out = [];
+  (function walk(dir, rel) {
+    for (const e of fs.readdirSync(dir, { withFileTypes: true })) {
+      if (e.isDirectory()) walk(path.join(dir, e.name), path.join(rel, e.name));
+      else if (e.name.endsWith('.md')) out.push(path.join(rel, e.name));
+    }
+  })(WF, '');
+  return out;
+}
+
+// Return the full text of every `Agent( … )` call in `content`, tracking string
+// state (triple- and single-double-quoted) and paren depth so a prompt body's own
+// parens/quotes don't end the call early. Order-independent: a call's params are
+// captured whether subagent_type= appears before or after the prompt.
+function agentCalls(content) {
+  const calls = [];
+  const re = /Agent\(/g;
+  let m;
+  while ((m = re.exec(content)) !== null) {
+    let i = m.index + m[0].length;
+    let depth = 1;
+    let tq = false; // inside """ … """
+    let sq = false; // inside " … "
+    while (i < content.length && depth > 0) {
+      if (tq) {
+        if (content.startsWith('"""', i)) { tq = false; i += 3; continue; }
+        i++; continue;
+      }
+      if (sq) {
+        if (content[i] === '\\') { i += 2; continue; }
+        if (content[i] === '"') { sq = false; }
+        i++; continue;
+      }
+      if (content.startsWith('"""', i)) { tq = true; i += 3; continue; }
+      if (content[i] === '"') { sq = true; i++; continue; }
+      if (content[i] === '(') { depth++; }
+      else if (content[i] === ')') { depth--; }
+      i++;
+    }
+    calls.push(content.slice(m.index, i));
+    re.lastIndex = i; // don't re-scan inside this call
+  }
+  return calls;
+}
+
+describe('#2072: routed-agent spawns thread the resolved model', () => {
+  test('discuss-phase-assumptions.md resolves + threads gsd-assumptions-analyzer model', () => {
+    const c = read('discuss-phase-assumptions.md');
+    assert.match(c, /resolve-model gsd-assumptions-analyzer/, 'must resolve the routed model');
+    assert.match(
+      c,
+      /subagent_type="gsd-assumptions-analyzer",\s*model="\{ANALYZER_MODEL\}"/,
+      'spawn must thread the resolved model=',
+    );
+  });
+
+  test('code-review.md resolves + threads gsd-code-reviewer model', () => {
+    const c = read('code-review.md');
+    assert.match(c, /resolve-model gsd-code-reviewer/);
+    assert.match(c, /subagent_type="gsd-code-reviewer",\s*model="\{REVIEWER_MODEL\}"/);
+  });
+
+  test('code-review-fix.md re-review resolves + threads gsd-code-reviewer model', () => {
+    const c = read('code-review-fix.md');
+    assert.match(c, /resolve-model gsd-code-reviewer/);
+    assert.match(c, /subagent_type="gsd-code-reviewer",\s*model="\{REVIEWER_MODEL\}"/);
+  });
+
+  test('quick.md review step threads gsd-code-reviewer own model (not executor_model)', () => {
+    const c = read('quick.md');
+    // reviewer_model comes from the init.quick bundle; the spawn must use it.
+    assert.match(c, /subagent_type="gsd-code-reviewer",\s*\n\s*model="\{reviewer_model\}"/);
+    assert.doesNotMatch(
+      c,
+      /subagent_type="gsd-code-reviewer",\s*\n\s*model="\{executor_model\}"/,
+      'reviewer must not reuse the executor model (own model_overrides would be ignored)',
+    );
+  });
+
+  test('code-review-fix.md threads gsd-code-fixer model at both fixer spawns', () => {
+    const c = read('code-review-fix.md');
+    assert.match(c, /resolve-model gsd-code-fixer/, 'must resolve the fixer model');
+    const fixerSpawns = c.match(/subagent_type="gsd-code-fixer", model="\{FIXER_MODEL\}"/g) || [];
+    assert.strictEqual(fixerSpawns.length, 2, 'both gsd-code-fixer spawns must thread FIXER_MODEL');
+  });
+
+  // Parity guard: EVERY spawn of the fixed routed agents across ALL workflows must
+  // carry a model= in its Agent(...) call, so a new silently-inert spawn cannot
+  // regress. Uses a quote/paren-aware scan of the WHOLE Agent(...) call, so it holds
+  // for single-line and multi-line calls, multiple spawns per file, and either param
+  // ordering (subagent_type= before OR after the prompt body).
+  test('no spawn of the fixed routed agents is missing a model= (parity guard)', () => {
+    const ROUTED = /subagent_type="(gsd-code-reviewer|gsd-assumptions-analyzer|gsd-code-fixer)"/;
+    const offenders = [];
+    for (const rel of allWorkflowMd()) {
+      for (const call of agentCalls(read(rel))) {
+        const routed = call.match(ROUTED);
+        if (routed && !/\bmodel\s*=/.test(call)) {
+          offenders.push(`${rel}: ${routed[1]} spawn missing model=`);
+        }
+      }
+    }
+    assert.deepEqual(offenders, [], `routed-agent spawn(s) missing model=:\n${offenders.join('\n')}`);
+  });
+
+  // The scanner itself must catch a prompt-first, un-threaded spawn (the exact blind
+  // spot a naive "params before prompt=" heuristic misses) — otherwise the guard
+  // above could pass vacuously.
+  test('parity guard detects a prompt-first spawn that omits model=', () => {
+    const synthetic = [
+      'Agent(',
+      '  prompt="""do the thing (with parens) and a " quote""",',
+      '  subagent_type="gsd-code-reviewer"',
+      ')',
+    ].join('\n');
+    const [call] = agentCalls(synthetic);
+    assert.ok(/subagent_type="gsd-code-reviewer"/.test(call), 'scanner must capture the prompt-first subagent_type');
+    assert.ok(!/\bmodel\s*=/.test(call), 'and correctly see that model= is absent');
+    // A well-formed prompt-first spawn WITH model= must be accepted.
+    const ok = agentCalls(synthetic.replace(')', '  model="{reviewer_model}"\n)'))[0];
+    assert.ok(/\bmodel\s*=/.test(ok));
+  });
+});