pi-crew 0.2.3 → 0.2.5

package/src/runtime/model-resolver.ts

@@ -0,0 +1,118 @@
+export interface ModelEntry {
+	id: string;
+	name: string;
+	provider: string;
+}
+
+export interface ModelRegistry {
+	find(provider: string, modelId: string): any;
+	getAll(): any[];
+	getAvailable?(): any[];
+}
+
+/**
+ * Resolve a model string to a Model instance.
+ * Exact match first ("provider/modelId"), then fuzzy match.
+ * Returns Model on success, error message string on failure.
+ */
+export function resolveModel(input: string, registry: ModelRegistry): any | string {
+	const all = (registry.getAvailable?.() ?? registry.getAll()) as ModelEntry[];
+	const availableSet = new Set(all.map((m) => `${m.provider}/${m.id}`.toLowerCase()));
+
+	// Exact match
+	const slashIdx = input.indexOf("/");
+	if (slashIdx !== -1) {
+		const provider = input.slice(0, slashIdx);
+		const modelId = input.slice(slashIdx + 1);
+		if (availableSet.has(input.toLowerCase())) {
+			const found = registry.find(provider, modelId);
+			if (found) return found;
+		}
+	}
+
+	// Fuzzy match
+	const query = input.toLowerCase();
+	let bestMatch: ModelEntry | undefined;
+	let bestScore = 0;
+
+	for (const m of all) {
+		const id = m.id.toLowerCase();
+		const name = m.name.toLowerCase();
+		const full = `${m.provider}/${m.id}`.toLowerCase();
+
+		let score = 0;
+		if (id === query || full === query) {
+			score = 100;
+		} else if (id.includes(query) || full.includes(query)) {
+			score = 60 + (query.length / id.length) * 30;
+		} else if (name.includes(query)) {
+			score = 40 + (query.length / name.length) * 20;
+		} else if (
+			query
+				.split(/[\s\-/]+/)
+				.every((part) => id.includes(part) || name.includes(part) || m.provider.toLowerCase().includes(part))
+		) {
+			score = 20;
+		}
+
+		if (score > bestScore) {
+			bestScore = score;
+			bestMatch = m;
+		}
+	}
+
+	if (bestMatch && bestScore >= 20) {
+		const found = registry.find(bestMatch.provider, bestMatch.id);
+		if (found) return found;
+	}
+
+	const modelList = all
+		.map((m) => `  ${m.provider}/${m.id}`)
+		.sort()
+		.join("\n");
+	return `Model not found: "${input}".\n\nAvailable models:\n${modelList}`;
+}
+
+export interface SimpleModelEntry {
+	id: string;
+	name?: string;
+	provider: string;
+}
+
+/**
+ * Fuzzy-match a model query against a flat list of available models.
+ * Returns the best-match fullId (provider/id) or undefined.
+ */
+export function fuzzyResolveModelId(input: string, models: SimpleModelEntry[]): string | undefined {
+	const query = input.toLowerCase();
+	let bestMatch: SimpleModelEntry | undefined;
+	let bestScore = 0;
+
+	for (const m of models) {
+		const id = m.id.toLowerCase();
+		const name = (m.name ?? "").toLowerCase();
+		const full = `${m.provider}/${m.id}`.toLowerCase();
+
+		let score = 0;
+		if (id === query || full === query) {
+			score = 100;
+		} else if (id.includes(query) || full.includes(query)) {
+			score = 60 + (query.length / id.length) * 30;
+		} else if (name.includes(query)) {
+			score = 40 + (query.length / (name.length || 1)) * 20;
+		} else if (
+			query
+				.split(/[\s\-/]+/)
+				.every((part) => id.includes(part) || name.includes(part) || m.provider.toLowerCase().includes(part))
+		) {
+			score = 20;
+		}
+
+		if (score > bestScore) {
+			bestScore = score;
+			bestMatch = m;
+		}
+	}
+
+	return bestMatch && bestScore >= 20 ? `${bestMatch.provider}/${bestMatch.id}` : undefined;
+}

package/src/runtime/output-validator.ts

@@ -1,187 +1,187 @@
-/**
- * Phase 4 (caveman): Output format validation for live-session workers.
- *
- * Validates that worker output conforms to the structured output contract
- * for the given role. If validation fails, returns structured error info
- * that can be used for retry or fallback.
- *
- * Inspired by caveman's validate.py — check structural preservation
- * (headings, code blocks, URLs) after compression.
- */
-
-/** Role-specific output format patterns — constructed fresh per call to avoid /g lastIndex leak */
-const ROLE_PATTERN_DEFS: Record<string, () => RegExp> = {
-	explorer: () => /^(\S+:\d+|Defs:|Refs:|Callers:|Tests:|Sites:|No match\.|totals:)/m,
-	executor: () => /^(\S+:\d+(-\d+)? — .{1,80}\.|verified:|too-big\.|needs-confirm\.|ambiguous\.|regressed\.)/m,
-	reviewer: () => /^([^:\s]+:\d+:\s+\p{Emoji_Presentation}|No issues\.|totals:)/mu,
-	"security-reviewer": () => /^([^:\s]+:\d+:\s+\p{Emoji_Presentation}|No issues\.|totals:)/mu,
-	verifier: () => /^(PASS:|FAIL:)/m,
-};
-
-/** Fresh RegExp factories for structural preservation checks (avoids /g lastIndex leak) */
-const makeUrlRe = () => /\bhttps?:\/\/[^\s<>)\]"',;]+/gi;
-const makeFencedCodeRe = () => /```[\s\S]*?```/g;
-const makeInlineCodeRe = () => /`[^`\n]+`/g;
-const makeHeadingRe = () => /^#{1,6}\s+.+/gm;
-
-export interface OutputValidationResult {
-	/** Whether the output passes validation */
-	valid: boolean;
-	/** Whether the output follows the role's contract format */
-	formatMatch: boolean;
-	/** Whether structural elements (code, URLs, headings) are preserved */
-	structurePreserved: boolean;
-	/** Specific issues found */
-	issues: string[];
-}
-
-/**
- * Validate worker output against role-specific contract + structural preservation.
- */
-export function validateWorkerOutput(role: string, output: string): OutputValidationResult {
-	const issues: string[] = [];
-
-	// Empty output always fails
-	if (!output || !output.trim()) {
-		return { valid: false, formatMatch: false, structurePreserved: false, issues: ["Empty output"] };
-	}
-
-	// Check role-specific format
-	const patternFactory = ROLE_PATTERN_DEFS[role];
-	const pattern = patternFactory ? patternFactory() : undefined;
-	const formatMatch = !pattern || pattern.test(output);
-	if (!formatMatch) {
-		issues.push(`Output does not match expected ${role} contract format`);
-	}
-
-	// Check structural preservation (code blocks, URLs, headings)
-	let structurePreserved = true;
-	const trimmedOutput = output.trim();
-
-	// Detect if output was truncated mid-code-block
-	const opens = (trimmedOutput.match(/```/g) ?? []).length;
-	if (opens % 2 !== 0) {
-		structurePreserved = false;
-		issues.push("Unclosed code block — output may be truncated");
-	}
-
-	// Check for malformed URLs
-	const urls = trimmedOutput.match(makeUrlRe()) ?? [];
-	for (const url of urls) {
-		if (url.endsWith(".") || url.endsWith(",")) {
-			structurePreserved = false;
-			issues.push(`URL with trailing punctuation: ${url.slice(-20)}`);
-		}
-	}
-
-	return {
-		valid: formatMatch && structurePreserved,
-		formatMatch,
-		structurePreserved,
-		issues,
-	};
-}
-
-/**
- * Extract structured findings from reviewer output.
- * Returns array of { file, line, severity, message } objects.
- */
-export function parseReviewerFindings(output: string): Array<{ file: string; line: number; severity: string; message: string }> {
-	const findings: Array<{ file: string; line: number; severity: string; message: string }> = [];
-	const lines = output.split("\n");
-
-	const SEVERITY_MAP: Record<string, string> = {
-		"🔴": "bug",
-		"🟡": "risk",
-		"🔵": "nit",
-		"❓": "question",
-	};
-
-	for (const line of lines) {
-		// Match: path/to/file.ts:42: 🔴 bug: problem. fix.
-		const match = line.match(/^([^:\s]+):(\d+):\s+(\p{Emoji_Presentation}) (\w+):\s+(.+)/u);
-		if (match) {
-			findings.push({
-				file: match[1],
-				line: Number(match[2]),
-				severity: SEVERITY_MAP[match[3]] ?? match[3],
-				message: match[5].trim(),
-			});
-		}
-	}
-
-	return findings;
-}
-
-/**
- * Extract explorer results from structured output.
- * Returns array of { file, line, symbol, note } objects.
- */
-export function parseExplorerResults(output: string): Array<{ file: string; line: number; symbol: string; note: string }> {
-	const results: Array<{ file: string; line: number; symbol: string; note: string }> = [];
-	const lines = output.split("\n");
-
-	for (const line of lines) {
-		// Match: path/to/file.ts:42 — `symbol` — note
-		const match = line.match(/^[- ]*(\S+):(\d+)\s*[—–-]\s*`([^`]+)`\s*[—–-]\s*(.+)/);
-		if (match) {
-			results.push({
-				file: match[1],
-				line: Number(match[2]),
-				symbol: match[3],
-				note: match[4].trim(),
-			});
-		}
-	}
-
-	return results;
-}
-
-/**
- * Validate that compressed prose preserves structural elements from original.
- * Returns list of specific issues (empty = valid).
- */
-export function validateCompressionPreservation(original: string, compressed: string): string[] {
-	const issues: string[] = [];
-
-	// Check code blocks preserved
-	const origBlocks = original.match(makeFencedCodeRe()) ?? [];
-	const compBlocks = compressed.match(makeFencedCodeRe()) ?? [];
-	if (origBlocks.length !== compBlocks.length) {
-		issues.push(`Code block count: ${origBlocks.length} → ${compBlocks.length}`);
-	}
-	for (let i = 0; i < Math.min(origBlocks.length, compBlocks.length); i++) {
-		if (origBlocks[i] !== compBlocks[i]) {
-			issues.push(`Code block ${i + 1} content changed`);
-		}
-	}
-
-	// Check URLs preserved
-	const origUrls = new Set(original.match(makeUrlRe()) ?? []);
-	const compUrls = new Set(compressed.match(makeUrlRe()) ?? []);
-	for (const url of origUrls) {
-		if (!compUrls.has(url)) {
-			issues.push(`URL lost: ${url.slice(0, 60)}...`);
-		}
-	}
-
-	// Check inline code preserved
-	const origInline = original.match(makeInlineCodeRe()) ?? [];
-	const compInline = compressed.match(makeInlineCodeRe()) ?? [];
-	const origInlineSet = new Set(origInline);
-	const compInlineSet = new Set(compInline);
-	for (const code of origInlineSet) {
-		if (!compInlineSet.has(code)) {
-			issues.push(`Inline code lost: ${code}`);
-		}
-	}
-
-	// Check headings preserved
-	const origHeadings = original.match(makeHeadingRe()) ?? [];
-	const compHeadings = compressed.match(makeHeadingRe()) ?? [];
-	if (origHeadings.length !== compHeadings.length) {
-		issues.push(`Heading count: ${origHeadings.length} → ${compHeadings.length}`);
-	}
-
-	return issues;
-}
+/**
+ * Phase 4 (caveman): Output format validation for live-session workers.
+ *
+ * Validates that worker output conforms to the structured output contract
+ * for the given role. If validation fails, returns structured error info
+ * that can be used for retry or fallback.
+ *
+ * Inspired by caveman's validate.py — check structural preservation
+ * (headings, code blocks, URLs) after compression.
+ */
+
+/** Role-specific output format patterns — constructed fresh per call to avoid /g lastIndex leak */
+const ROLE_PATTERN_DEFS: Record<string, () => RegExp> = {
+	explorer: () => /^(\S+:\d+|Defs:|Refs:|Callers:|Tests:|Sites:|No match\.|totals:)/m,
+	executor: () => /^(\S+:\d+(-\d+)? — .{1,80}\.|verified:|too-big\.|needs-confirm\.|ambiguous\.|regressed\.)/m,
+	reviewer: () => /^([^:\s]+:\d+:\s+\p{Emoji_Presentation}|No issues\.|totals:)/mu,
+	"security-reviewer": () => /^([^:\s]+:\d+:\s+\p{Emoji_Presentation}|No issues\.|totals:)/mu,
+	verifier: () => /^(PASS:|FAIL:)/m,
+};
+
+/** Fresh RegExp factories for structural preservation checks (avoids /g lastIndex leak) */
+const makeUrlRe = () => /\bhttps?:\/\/[^\s<>)\]"',;]+/gi;
+const makeFencedCodeRe = () => /```[\s\S]*?```/g;
+const makeInlineCodeRe = () => /`[^`\n]+`/g;
+const makeHeadingRe = () => /^#{1,6}\s+.+/gm;
+
+export interface OutputValidationResult {
+	/** Whether the output passes validation */
+	valid: boolean;
+	/** Whether the output follows the role's contract format */
+	formatMatch: boolean;
+	/** Whether structural elements (code, URLs, headings) are preserved */
+	structurePreserved: boolean;
+	/** Specific issues found */
+	issues: string[];
+}
+
+/**
+ * Validate worker output against role-specific contract + structural preservation.
+ */
+export function validateWorkerOutput(role: string, output: string): OutputValidationResult {
+	const issues: string[] = [];
+
+	// Empty output always fails
+	if (!output || !output.trim()) {
+		return { valid: false, formatMatch: false, structurePreserved: false, issues: ["Empty output"] };
+	}
+
+	// Check role-specific format
+	const patternFactory = ROLE_PATTERN_DEFS[role];
+	const pattern = patternFactory ? patternFactory() : undefined;
+	const formatMatch = !pattern || pattern.test(output);
+	if (!formatMatch) {
+		issues.push(`Output does not match expected ${role} contract format`);
+	}
+
+	// Check structural preservation (code blocks, URLs, headings)
+	let structurePreserved = true;
+	const trimmedOutput = output.trim();
+
+	// Detect if output was truncated mid-code-block
+	const opens = (trimmedOutput.match(/```/g) ?? []).length;
+	if (opens % 2 !== 0) {
+		structurePreserved = false;
+		issues.push("Unclosed code block — output may be truncated");
+	}
+
+	// Check for malformed URLs
+	const urls = trimmedOutput.match(makeUrlRe()) ?? [];
+	for (const url of urls) {
+		if (url.endsWith(".") || url.endsWith(",")) {
+			structurePreserved = false;
+			issues.push(`URL with trailing punctuation: ${url.slice(-20)}`);
+		}
+	}
+
+	return {
+		valid: formatMatch && structurePreserved,
+		formatMatch,
+		structurePreserved,
+		issues,
+	};
+}
+
+/**
+ * Extract structured findings from reviewer output.
+ * Returns array of { file, line, severity, message } objects.
+ */
+export function parseReviewerFindings(output: string): Array<{ file: string; line: number; severity: string; message: string }> {
+	const findings: Array<{ file: string; line: number; severity: string; message: string }> = [];
+	const lines = output.split("\n");
+
+	const SEVERITY_MAP: Record<string, string> = {
+		"🔴": "bug",
+		"🟡": "risk",
+		"🔵": "nit",
+		"❓": "question",
+	};
+
+	for (const line of lines) {
+		// Match: path/to/file.ts:42: 🔴 bug: problem. fix.
+		const match = line.match(/^([^:\s]+):(\d+):\s+(\p{Emoji_Presentation}) (\w+):\s+(.+)/u);
+		if (match) {
+			findings.push({
+				file: match[1],
+				line: Number(match[2]),
+				severity: SEVERITY_MAP[match[3]] ?? match[3],
+				message: match[5].trim(),
+			});
+		}
+	}
+
+	return findings;
+}
+
+/**
+ * Extract explorer results from structured output.
+ * Returns array of { file, line, symbol, note } objects.
+ */
+export function parseExplorerResults(output: string): Array<{ file: string; line: number; symbol: string; note: string }> {
+	const results: Array<{ file: string; line: number; symbol: string; note: string }> = [];
+	const lines = output.split("\n");
+
+	for (const line of lines) {
+		// Match: path/to/file.ts:42 — `symbol` — note
+		const match = line.match(/^[- ]*(\S+):(\d+)\s*[—–-]\s*`([^`]+)`\s*[—–-]\s*(.+)/);
+		if (match) {
+			results.push({
+				file: match[1],
+				line: Number(match[2]),
+				symbol: match[3],
+				note: match[4].trim(),
+			});
+		}
+	}
+
+	return results;
+}
+
+/**
+ * Validate that compressed prose preserves structural elements from original.
+ * Returns list of specific issues (empty = valid).
+ */
+export function validateCompressionPreservation(original: string, compressed: string): string[] {
+	const issues: string[] = [];
+
+	// Check code blocks preserved
+	const origBlocks = original.match(makeFencedCodeRe()) ?? [];
+	const compBlocks = compressed.match(makeFencedCodeRe()) ?? [];
+	if (origBlocks.length !== compBlocks.length) {
+		issues.push(`Code block count: ${origBlocks.length} → ${compBlocks.length}`);
+	}
+	for (let i = 0; i < Math.min(origBlocks.length, compBlocks.length); i++) {
+		if (origBlocks[i] !== compBlocks[i]) {
+			issues.push(`Code block ${i + 1} content changed`);
+		}
+	}
+
+	// Check URLs preserved
+	const origUrls = new Set(original.match(makeUrlRe()) ?? []);
+	const compUrls = new Set(compressed.match(makeUrlRe()) ?? []);
+	for (const url of origUrls) {
+		if (!compUrls.has(url)) {
+			issues.push(`URL lost: ${url.slice(0, 60)}...`);
+		}
+	}
+
+	// Check inline code preserved
+	const origInline = original.match(makeInlineCodeRe()) ?? [];
+	const compInline = compressed.match(makeInlineCodeRe()) ?? [];
+	const origInlineSet = new Set(origInline);
+	const compInlineSet = new Set(compInline);
+	for (const code of origInlineSet) {
+		if (!compInlineSet.has(code)) {
+			issues.push(`Inline code lost: ${code}`);
+		}
+	}
+
+	// Check headings preserved
+	const origHeadings = original.match(makeHeadingRe()) ?? [];
+	const compHeadings = compressed.match(makeHeadingRe()) ?? [];
+	if (origHeadings.length !== compHeadings.length) {
+		issues.push(`Heading count: ${origHeadings.length} → ${compHeadings.length}`);
+	}
+
+	return issues;
+}