@askjo/pi-reflect 1.0.0 → 1.2.0

package/tests/helpers-and-utils.test.ts

@@ -6,8 +6,37 @@ import {
 	escapeRegex,
 	truncateText,
 	projectNameFromDir,
+	resolveModelAuth,
+	buildReflectAnalysisTool,
 } from "../extensions/reflect.js";
 
+describe("resolveModelAuth", () => {
+	const model = { provider: "test", id: "model" };
+
+	it("prefers current getApiKeyAndHeaders and preserves header-only auth", async () => {
+		let legacyCalled = false;
+		const auth = await resolveModelAuth({
+			getApiKeyAndHeaders: async () => ({ ok: true, headers: { "X-Test": "yes" } }),
+			getApiKey: async () => { legacyCalled = true; return "legacy"; },
+		}, model);
+		assert.deepEqual(auth, { ok: true, apiKey: undefined, headers: { "X-Test": "yes" } });
+		assert.equal(legacyCalled, false);
+	});
+
+	it("falls back to the legacy getApiKey API", async () => {
+		assert.deepEqual(await resolveModelAuth({ getApiKey: async () => "legacy" }, model), { ok: true, apiKey: "legacy" });
+	});
+});
+
+describe("reflect analysis tool schema", () => {
+	it("uses scalar JSON Schema types for protobuf-compatible optional fields", () => {
+		const properties = (buildReflectAnalysisTool().parameters.properties.edits.items as any).properties;
+		assert.equal(properties.old_text.type, "string");
+		assert.equal(properties.after_text.type, "string");
+		assert.equal(properties.merge_sources.type, "array");
+	});
+});
+
 describe("resolvePath", () => {
 	it("expands ~ to HOME", () => {
 		const home = process.env.HOME!;

package/tests/prompt-building.test.ts

@@ -27,25 +27,18 @@ describe("buildReflectionPrompt", () => {
 	it("instructs JSON-only output", () => {
 		const prompt = buildReflectionPrompt("/AGENTS.md", "content", "transcripts");
 		assert.ok(prompt.includes("single JSON object"));
-		assert.ok(prompt.includes("No markdown, no explanation, no preamble"));
+		assert.ok(prompt.includes("No markdown, no preamble"));
 	});
 
-	it("describes the strengthen edit type", () => {
+	it("describes all edit types", () => {
 		const prompt = buildReflectionPrompt("/AGENTS.md", "content", "transcripts");
 		assert.ok(prompt.includes('"strengthen"'));
-		assert.ok(prompt.includes("old_text"));
-		assert.ok(prompt.includes("character-for-character"));
-	});
-
-	it("describes the add edit type", () => {
-		const prompt = buildReflectionPrompt("/AGENTS.md", "content", "transcripts");
 		assert.ok(prompt.includes('"add"'));
+		assert.ok(prompt.includes('"remove"'));
+		assert.ok(prompt.includes('"merge"'));
+		assert.ok(prompt.includes("old_text"));
 		assert.ok(prompt.includes("after_text"));
-	});
-
-	it("mentions duplication prevention", () => {
-		const prompt = buildReflectionPrompt("/AGENTS.md", "content", "transcripts");
-		assert.ok(prompt.includes("Never duplicate content"));
+		assert.ok(prompt.includes("merge_sources"));
 	});
 
 	it("requires 2+ occurrences for new rules", () => {
@@ -55,15 +48,14 @@ describe("buildReflectionPrompt", () => {
 
 	it("lists correction signals to look for", () => {
 		const prompt = buildReflectionPrompt("/AGENTS.md", "content", "transcripts");
-		assert.ok(prompt.includes("bro"));
-		assert.ok(prompt.includes("wtf"));
-		assert.ok(prompt.includes("actually"));
+		assert.ok(prompt.includes("frustration"));
+		assert.ok(prompt.includes("correcting"));
 	});
 
 	it("warns about false positives", () => {
 		const prompt = buildReflectionPrompt("/AGENTS.md", "content", "transcripts");
 		assert.ok(prompt.includes("no worries"));
-		assert.ok(prompt.includes("NOT corrections"));
+		assert.ok(prompt.includes("Ignore normal flow"));
 	});
 
 	it("uses basename, not full path in prompt", () => {

package/tests/reflect.test.ts

@@ -0,0 +1,539 @@
+import { describe, it, before, after } from "node:test";
+import assert from "node:assert/strict";
+import * as fs from "node:fs";
+import * as path from "node:path";
+import * as os from "node:os";
+
+import {
+	computeFileMetrics,
+	applyEdits,
+	buildTranscriptBatches,
+	buildReflectionPrompt,
+	buildPromptForTarget,
+	escapeRegex,
+	truncateText,
+	projectNameFromDir,
+	extractTranscript,
+	formatSessionTranscript,
+	loadHistory,
+	saveHistory,
+	loadConfig,
+	saveConfig,
+	resolvePath,
+	type AnalysisEdit,
+	type ReflectTarget,
+	type ReflectRun,
+	type SessionData,
+	type SessionExchange,
+	DEFAULT_TARGET,
+	HISTORY_FILE,
+	CONFIG_FILE,
+} from "../extensions/reflect.js";
+
+function makeTmpDir(): string {
+	return fs.mkdtempSync(path.join(os.tmpdir(), "pi-reflect-test-"));
+}
+
+function rmrf(dir: string): void {
+	fs.rmSync(dir, { recursive: true, force: true });
+}
+
+// --- computeFileMetrics ---
+
+describe("computeFileMetrics", () => {
+	it("counts chars, words, lines, estTokens", () => {
+		const m = computeFileMetrics("hello world\nsecond line\n");
+		assert.equal(m.chars, 24);
+		assert.equal(m.words, 4);
+		assert.equal(m.lines, 3); // trailing newline = extra empty line
+		assert.equal(m.estTokens, Math.ceil(24 / 4));
+	});
+
+	it("handles empty string", () => {
+		const m = computeFileMetrics("");
+		assert.equal(m.chars, 0);
+		assert.equal(m.words, 0);
+		assert.equal(m.lines, 1);
+		assert.equal(m.estTokens, 0);
+	});
+});
+
+// --- escapeRegex ---
+
+describe("escapeRegex", () => {
+	it("escapes regex special chars", () => {
+		assert.equal(escapeRegex("a.b*c?d"), "a\\.b\\*c\\?d");
+		assert.equal(escapeRegex("foo[bar]"), "foo\\[bar\\]");
+		assert.equal(escapeRegex("$100 (USD)"), "\\$100 \\(USD\\)");
+	});
+});
+
+// --- truncateText ---
+
+describe("truncateText", () => {
+	it("returns null for null input", () => {
+		assert.equal(truncateText(null, 10), null);
+	});
+
+	it("truncates long text", () => {
+		const result = truncateText("hello world this is long", 10);
+		assert.ok(result !== null);
+		// truncateText uses "\n[...truncated, N chars omitted]" format
+		assert.ok(result!.includes("truncated") || result!.length <= 24);
+		// Original 24 chars, limit 10 — should be truncated
+		assert.ok(result!.length < 24 || result!.includes("truncated"));
+	});
+
+	it("returns text unchanged if within limit", () => {
+		assert.equal(truncateText("short", 100), "short");
+	});
+});
+
+// --- projectNameFromDir ---
+
+describe("projectNameFromDir", () => {
+	it("extracts project name from worktree path", () => {
+		const name = projectNameFromDir("my-project-feature-branch");
+		assert.equal(typeof name, "string");
+	});
+});
+
+// --- applyEdits ---
+
+describe("applyEdits", () => {
+	const baseContent = `# Rules
+
+## Debugging
+- **Always check logs first** before theorizing.
+- **Don't guess** — verify assumptions with data.
+
+## Execution
+- **Execute first, explain minimally**: Do the work, say what you did.
+- **Use the oracle subagent early**: For complex problems, use the oracle BEFORE spending 10+ minutes.
+`;
+
+	it("strengthen: replaces matching text", () => {
+		const edits: AnalysisEdit[] = [{
+			type: "strengthen",
+			old_text: "- **Always check logs first** before theorizing.",
+			new_text: "- **Check logs first** — no theorizing without evidence.",
+		}];
+		const { result, applied, skipped } = applyEdits(baseContent, edits);
+		assert.equal(applied, 1);
+		assert.equal(skipped.length, 0);
+		assert.ok(result.includes("Check logs first"));
+		assert.ok(!result.includes("before theorizing."));
+	});
+
+	it("strengthen: skips when old_text not found", () => {
+		const edits: AnalysisEdit[] = [{
+			type: "strengthen",
+			old_text: "- This text does not exist in the file.",
+			new_text: "- Replacement text.",
+		}];
+		const { applied, skipped } = applyEdits(baseContent, edits);
+		assert.equal(applied, 0);
+		assert.equal(skipped.length, 1);
+		assert.ok(skipped[0].includes("Could not find"));
+	});
+
+	it("strengthen: skips ambiguous matches", () => {
+		const content = "rule A\nrule A\n";
+		const edits: AnalysisEdit[] = [{
+			type: "strengthen",
+			old_text: "rule A",
+			new_text: "rule B",
+		}];
+		const { applied, skipped } = applyEdits(content, edits);
+		assert.equal(applied, 0);
+		assert.equal(skipped.length, 1);
+		assert.ok(skipped[0].includes("Ambiguous"));
+	});
+
+	it("strengthen: detects duplication in replacement text", () => {
+		// old_text is >50 chars so duplication detection kicks in
+		const longRule = "- **When the user pushes back or says you're wrong, pivot immediately**: After 1 pushback, stop.";
+		const content = `# Rules\n${longRule}\n`;
+		const edits: AnalysisEdit[] = [{
+			type: "strengthen",
+			old_text: longRule,
+			// new_text contains the old text twice (LLM duplication bug)
+			new_text: `${longRule}\n${longRule}`,
+		}];
+		const { applied, skipped } = applyEdits(content, edits);
+		assert.equal(applied, 0);
+		assert.equal(skipped.length, 1);
+		assert.ok(skipped[0].includes("Duplication"));
+	});
+
+	it("add: inserts after matching text", () => {
+		const edits: AnalysisEdit[] = [{
+			type: "add",
+			new_text: "- **New rule**: Always verify the fix works.",
+			after_text: "- **Don't guess** — verify assumptions with data.",
+		}];
+		const { result, applied } = applyEdits(baseContent, edits);
+		assert.equal(applied, 1);
+		assert.ok(result.includes("New rule"));
+		// Verify insertion is after the anchor
+		const anchorIdx = result.indexOf("verify assumptions with data.");
+		const newIdx = result.indexOf("New rule");
+		assert.ok(newIdx > anchorIdx);
+	});
+
+	it("add: skips if text already exists", () => {
+		const edits: AnalysisEdit[] = [{
+			type: "add",
+			new_text: "- **Don't guess** — verify assumptions with data.",
+			after_text: "- **Always check logs first** before theorizing.",
+		}];
+		const { applied, skipped } = applyEdits(baseContent, edits);
+		assert.equal(applied, 0);
+		assert.equal(skipped.length, 1);
+		assert.ok(skipped[0].includes("already exists"));
+	});
+
+	it("add: skips ambiguous insertion point", () => {
+		const content = "anchor\nstuff\nanchor\n";
+		const edits: AnalysisEdit[] = [{
+			type: "add",
+			new_text: "new line",
+			after_text: "anchor",
+		}];
+		const { applied, skipped } = applyEdits(content, edits);
+		assert.equal(applied, 0);
+		assert.ok(skipped[0].includes("Ambiguous"));
+	});
+
+	it("remove: deletes matching text", () => {
+		const edits: AnalysisEdit[] = [{
+			type: "remove",
+			old_text: "- **Don't guess** — verify assumptions with data.",
+			new_text: "",
+		}];
+		const { result, applied } = applyEdits(baseContent, edits);
+		assert.equal(applied, 1);
+		assert.ok(!result.includes("Don't guess"));
+	});
+
+	it("remove: skips ambiguous matches", () => {
+		const content = "dup line\nstuff\ndup line\n";
+		const edits: AnalysisEdit[] = [{
+			type: "remove",
+			old_text: "dup line",
+			new_text: "",
+		}];
+		const { applied, skipped } = applyEdits(content, edits);
+		assert.equal(applied, 0);
+		assert.ok(skipped[0].includes("Ambiguous"));
+	});
+
+	it("merge: consolidates multiple sources into one", () => {
+		const edits: AnalysisEdit[] = [{
+			type: "merge",
+			merge_sources: [
+				"- **Always check logs first** before theorizing.",
+				"- **Don't guess** — verify assumptions with data.",
+			],
+			new_text: "- **Check logs and verify with data** — no guessing or theorizing.",
+		}];
+		const { result, applied } = applyEdits(baseContent, edits);
+		assert.equal(applied, 1);
+		assert.ok(result.includes("Check logs and verify with data"));
+		assert.ok(!result.includes("Always check logs first"));
+		assert.ok(!result.includes("Don't guess"));
+	});
+
+	it("merge: skips when a source is missing", () => {
+		const edits: AnalysisEdit[] = [{
+			type: "merge",
+			merge_sources: [
+				"- **Always check logs first** before theorizing.",
+				"- This source does not exist.",
+			],
+			new_text: "- Merged rule.",
+		}];
+		const { applied, skipped } = applyEdits(baseContent, edits);
+		assert.equal(applied, 0);
+		assert.ok(skipped[0].includes("Merge source not found"));
+	});
+
+	it("applies multiple edits in sequence", () => {
+		const edits: AnalysisEdit[] = [
+			{
+				type: "strengthen",
+				old_text: "- **Always check logs first** before theorizing.",
+				new_text: "- **Logs first** — no theorizing.",
+			},
+			{
+				type: "add",
+				new_text: "- **3-attempt rule**: Stop after 3 failures.",
+				after_text: "- **Use the oracle subagent early**: For complex problems, use the oracle BEFORE spending 10+ minutes.",
+			},
+		];
+		const { result, applied } = applyEdits(baseContent, edits);
+		assert.equal(applied, 2);
+		assert.ok(result.includes("Logs first"));
+		assert.ok(result.includes("3-attempt rule"));
+	});
+
+	it("handles invalid edit gracefully", () => {
+		const edits: AnalysisEdit[] = [{
+			type: "strengthen" as any,
+			new_text: "replacement",
+			// missing old_text
+		}];
+		const { applied, skipped } = applyEdits(baseContent, edits);
+		assert.equal(applied, 0);
+		assert.equal(skipped.length, 1);
+		assert.ok(skipped[0].includes("Invalid edit"));
+	});
+});
+
+// --- buildTranscriptBatches ---
+
+describe("buildTranscriptBatches", () => {
+	function makeSessions(sizes: number[]): SessionData[] {
+		return sizes.map((size, i) => ({
+			userCount: 1,
+			exchangeCount: 2,
+			transcript: "x".repeat(size),
+			size,
+			project: "test",
+			time: `session-${i}`,
+		}));
+	}
+
+	it("puts all sessions in one batch when they fit", () => {
+		const sessions = makeSessions([100, 200, 300]);
+		const batches = buildTranscriptBatches(sessions, 10000);
+		assert.equal(batches.length, 1);
+		assert.equal(batches[0].length, 3);
+	});
+
+	it("splits sessions into multiple batches by size", () => {
+		const sessions = makeSessions([500, 500, 500]);
+		// Each entry adds "\n---\n\n" (6 chars) to the transcript
+		const batches = buildTranscriptBatches(sessions, 600);
+		assert.ok(batches.length >= 2);
+	});
+
+	it("handles a single oversized session", () => {
+		const sessions = makeSessions([10000]);
+		const batches = buildTranscriptBatches(sessions, 100);
+		// Even oversized, it should be in its own batch (not dropped)
+		assert.equal(batches.length, 1);
+		assert.equal(batches[0].length, 1);
+	});
+
+	it("returns empty for no sessions", () => {
+		const batches = buildTranscriptBatches([], 10000);
+		assert.equal(batches.length, 0);
+	});
+});
+
+// --- buildReflectionPrompt ---
+
+describe("buildReflectionPrompt", () => {
+	it("includes file name, content, and transcripts", () => {
+		const prompt = buildReflectionPrompt(
+			"/path/to/AGENTS.md",
+			"# Rules\n- Rule 1\n",
+			"## Session 1\nUser: fix this\nAssistant: done",
+		);
+		assert.ok(prompt.includes("AGENTS.md"));
+		assert.ok(prompt.includes("# Rules"));
+		assert.ok(prompt.includes("Rule 1"));
+		assert.ok(prompt.includes("Session 1"));
+		assert.ok(prompt.includes("fix this"));
+	});
+
+	it("includes conciseness instructions", () => {
+		const prompt = buildReflectionPrompt("/AGENTS.md", "content", "transcripts");
+		assert.ok(prompt.includes("Conciseness"));
+		assert.ok(prompt.includes("SHORTER"));
+	});
+
+	it("includes all edit types in instructions", () => {
+		const prompt = buildReflectionPrompt("/AGENTS.md", "content", "transcripts");
+		assert.ok(prompt.includes("strengthen"));
+		assert.ok(prompt.includes("add"));
+		assert.ok(prompt.includes("remove"));
+		assert.ok(prompt.includes("merge"));
+	});
+});
+
+// --- buildPromptForTarget ---
+
+describe("buildPromptForTarget", () => {
+	it("uses default prompt when no custom prompt", () => {
+		const target: ReflectTarget = { ...DEFAULT_TARGET, path: "/AGENTS.md" };
+		const prompt = buildPromptForTarget(target, "/AGENTS.md", "content", "transcripts");
+		assert.ok(prompt.includes("reviewing recent agent session"));
+	});
+
+	it("interpolates custom prompt template", () => {
+		const target: ReflectTarget = {
+			...DEFAULT_TARGET,
+			path: "/AGENTS.md",
+			prompt: "File: {fileName}\nContent: {targetContent}\nTranscripts: {transcripts}\nContext: {context}",
+		};
+		const prompt = buildPromptForTarget(target, "/path/to/AGENTS.md", "my content", "my transcripts", "my context");
+		assert.equal(prompt, "File: AGENTS.md\nContent: my content\nTranscripts: my transcripts\nContext: my context");
+	});
+});
+
+// --- Config and History I/O ---
+
+describe("config and history I/O", () => {
+	let origConfigFile: string;
+	let origHistoryFile: string;
+	let tmp: string;
+
+	before(() => {
+		tmp = makeTmpDir();
+		// Monkey-patch the module-level file paths
+		// These are exported as const, so we need to use Object.defineProperty
+		origConfigFile = CONFIG_FILE;
+		origHistoryFile = HISTORY_FILE;
+	});
+
+	after(() => {
+		rmrf(tmp);
+	});
+
+	it("loadHistory returns empty array for missing file", () => {
+		// loadHistory reads from HISTORY_FILE which we can't easily redirect
+		// So just test the function exists and handles errors
+		const history = loadHistory();
+		assert.ok(Array.isArray(history));
+	});
+
+	it("loadConfig returns empty targets for missing file", () => {
+		const config = loadConfig();
+		assert.ok(Array.isArray(config.targets));
+	});
+});
+
+// --- Dedup in runReflection ---
+
+describe("dedup logic", () => {
+	it("sourceDate is computed from lookbackDays", () => {
+		// Verify the date computation matches what runReflection uses
+		const lookbackDays = 1;
+		const sourceDate = new Date();
+		sourceDate.setDate(sourceDate.getDate() - lookbackDays);
+		const sourceDateStr = sourceDate.toISOString().slice(0, 10);
+
+		// Should be yesterday's date
+		const yesterday = new Date();
+		yesterday.setDate(yesterday.getDate() - 1);
+		assert.equal(sourceDateStr, yesterday.toISOString().slice(0, 10));
+	});
+
+	it("history entry sourceDate matching works", () => {
+		const targetPath = "/path/to/AGENTS.md";
+		const sourceDateStr = "2026-05-02";
+
+		const history: ReflectRun[] = [
+			{
+				timestamp: "2026-05-03T05:00:00Z",
+				targetPath,
+				sessionsAnalyzed: 10,
+				correctionsFound: 3,
+				editsApplied: 2,
+				summary: "test",
+				diffLines: 5,
+				correctionRate: 0.3,
+				sourceDate: sourceDateStr,
+			},
+		];
+
+		const alreadyRan = history.some(
+			(r) => r.targetPath === targetPath && r.sourceDate === sourceDateStr,
+		);
+		assert.ok(alreadyRan, "should detect already-processed date");
+
+		const notRan = history.some(
+			(r) => r.targetPath === targetPath && r.sourceDate === "2026-05-03",
+		);
+		assert.ok(!notRan, "should not match different date");
+
+		const diffTarget = history.some(
+			(r) => r.targetPath === "/other/file.md" && r.sourceDate === sourceDateStr,
+		);
+		assert.ok(!diffTarget, "should not match different target");
+	});
+});
+
+// --- extractTranscript ---
+
+describe("extractTranscript", () => {
+	let tmp: string;
+	before(() => { tmp = makeTmpDir(); });
+	after(() => { rmrf(tmp); });
+
+	it("extracts exchanges from a session file", async () => {
+		// Create a minimal session file (JSONL format)
+		// extractTranscript expects {type: "message", message: {role, content}} wrapper
+		const sessionFile = path.join(tmp, "test-session.jsonl");
+		const entries = [
+			{ type: "message", message: { role: "user", content: [{ type: "text", text: "fix the bug" }] } },
+			{ type: "message", message: { role: "assistant", content: [{ type: "text", text: "I'll look into it." }] } },
+			{ type: "message", message: { role: "user", content: [{ type: "text", text: "thanks" }] } },
+		];
+		fs.writeFileSync(sessionFile, entries.map(e => JSON.stringify(e)).join("\n"), "utf-8");
+
+		const exchanges = await extractTranscript(sessionFile);
+		assert.ok(exchanges.length >= 2, `expected >= 2 exchanges, got ${exchanges.length}`);
+		assert.equal(exchanges[0].role, "user");
+		assert.equal(exchanges[0].text, "fix the bug");
+		assert.equal(exchanges[1].role, "assistant");
+		assert.equal(exchanges[1].text, "I'll look into it.");
+	});
+
+	it("handles missing file gracefully", async () => {
+		const exchanges = await extractTranscript("/nonexistent/file.jsonl");
+		assert.deepEqual(exchanges, []);
+	});
+});
+
+// --- formatSessionTranscript ---
+
+describe("formatSessionTranscript", () => {
+	it("formats exchanges into readable text", () => {
+		const exchanges: SessionExchange[] = [
+			{ role: "user", text: "fix it", thinking: null },
+			{ role: "assistant", text: "done", thinking: "I should check the logs" },
+		];
+		const result = formatSessionTranscript(exchanges, "session-1", "my-project");
+		assert.ok(result.includes("fix it"));
+		assert.ok(result.includes("done"));
+		assert.ok(result.includes("session-1") || result.includes("my-project"));
+	});
+
+	it("handles empty exchanges", () => {
+		const result = formatSessionTranscript([], "empty-session", "proj");
+		assert.equal(typeof result, "string");
+	});
+});
+
+// --- resolvePath ---
+
+describe("resolvePath", () => {
+	it("resolves tilde paths", () => {
+		const home = process.env.HOME ?? os.homedir();
+		const resolved = resolvePath("~/test.md");
+		assert.equal(resolved, path.join(home, "test.md"));
+	});
+
+	it("resolves relative paths", () => {
+		const resolved = resolvePath("./test.md");
+		assert.ok(path.isAbsolute(resolved));
+	});
+
+	it("keeps absolute paths unchanged", () => {
+		const resolved = resolvePath("/absolute/path.md");
+		assert.equal(resolved, "/absolute/path.md");
+	});
+});

package/tests/run-reflection.test.ts

@@ -54,7 +54,7 @@ function makeDeps(llmResponseJson: string, transcripts = "### Session: test\n\n*
 function makeModelRegistry() {
 	return {
 		find: () => ({ provider: "test", id: "test-model" }),
-		getApiKey: async () => "test-key",
+		getApiKeyAndHeaders: async () => ({ ok: true, apiKey: "test-key", headers: undefined }),
 	};
 }
 
@@ -96,7 +96,7 @@ describe("runReflection", () => {
 			...makeDeps("{}"),
 			getModel: () => null,
 		};
-		const registry = { find: () => null, getApiKey: async () => null };
+		const registry = { find: () => null, getApiKeyAndHeaders: async () => ({ ok: false, error: "no key" }) };
 		const result = await runReflection(target, registry, notify, deps);
 		assert.equal(result, null);
 		assert.ok(notifications.some(n => n.level === "error" && n.msg.includes("Model not found")));
@@ -108,13 +108,42 @@ describe("runReflection", () => {
 		const target = makeTarget({ path: fp });
 		const registry = {
 			find: () => ({ provider: "test", id: "test" }),
-			getApiKey: async () => null,
+			getApiKeyAndHeaders: async () => ({ ok: false, error: "No API key" }),
 		};
 		const result = await runReflection(target, registry, notify, makeDeps("{}"));
 		assert.equal(result, null);
 		assert.ok(notifications.some(n => n.level === "error" && n.msg.includes("No API key")));
 	});
 
+	it("supports header-only auth from the current model registry", async () => {
+		const fp = path.join(tmpDir, "AGENTS.md");
+		fs.writeFileSync(fp, SAMPLE_AGENTS_MD);
+		let receivedOptions: any;
+		const deps = makeDeps(makeLlmResponse([], 0));
+		deps.completeSimple = async (_model, _request, options) => {
+			receivedOptions = options;
+			return { content: [{ type: "text", text: makeLlmResponse([], 0) }] };
+		};
+		const registry = {
+			find: () => ({ provider: "test", id: "test" }),
+			getApiKeyAndHeaders: async () => ({ ok: true, headers: { "X-Custom": "value" } }),
+		};
+		const result = await runReflection(makeTarget({ path: fp }), registry, notify, deps);
+		assert.notEqual(result, null);
+		assert.deepEqual(receivedOptions.headers, { "X-Custom": "value" });
+	});
+
+	it("supports the legacy modelRegistry.getApiKey API", async () => {
+		const fp = path.join(tmpDir, "AGENTS.md");
+		fs.writeFileSync(fp, SAMPLE_AGENTS_MD);
+		const registry = {
+			find: () => ({ provider: "test", id: "test" }),
+			getApiKey: async () => "legacy-key",
+		};
+		const result = await runReflection(makeTarget({ path: fp }), registry, notify, makeDeps(makeLlmResponse([], 0)));
+		assert.notEqual(result, null);
+	});
+
 	it("returns run with 0 edits when LLM says no edits needed", async () => {
 		const fp = path.join(tmpDir, "AGENTS.md");
 		fs.writeFileSync(fp, SAMPLE_AGENTS_MD);
@@ -312,7 +341,7 @@ describe("runReflection", () => {
 		let registryFindCalled = false;
 		const registry = {
 			find: () => { registryFindCalled = true; return { provider: "test", id: "test" }; },
-			getApiKey: async () => "test-key",
+			getApiKeyAndHeaders: async () => ({ ok: true, apiKey: "test-key", headers: undefined }),
 		};
 		const deps: RunReflectionDeps = {
 			...makeDeps(llmResponse),