@delegance/claude-autopilot 1.2.7 → 1.3.0

package/CHANGELOG.md

@@ -1,5 +1,50 @@
 # Changelog
 
+## [1.2.8] — 2026-04-21
+
+### Added
+- 8 new tests covering npm placeholder detection, pyproject.toml FastAPI detection, `resolveGitTouchedFiles` ignore list, deduplication, and status fallback — **136 total**
+
+## [1.2.7] — 2026-04-21
+
+### Fixed
+- `autopilot run` now loads `.env.local` / `.env` at startup so `OPENAI_API_KEY` (and other env vars) are available without exporting them in the shell first
+
+## [1.2.6] — 2026-04-21
+
+### Added
+- `skills/autopilot.md` included in npm package — install once, then `cp node_modules/@delegance/claude-autopilot/skills/autopilot.md .claude/skills/` to give Claude Code full context on when and how to invoke the CLI
+
+## [1.2.5] — 2026-04-21
+
+### Added
+- `--version` / `-v` flag — prints package version and exits
+- Built-in ignore list for git diff output: `node_modules/`, `dist/`, `build/`, `.next/`, `.nuxt/`, `out/`, `coverage/`, `.turbo/`, `.cache/`, `vendor/`, `__pycache__/`, `.venv/`, `venv/`, `target/`, `.gradle/` — prevents build artifact floods from polluting the changed-files list
+
+## [1.2.4] — 2026-04-21
+
+### Changed
+- `autopilot init` is now deprecated — prints a notice and delegates to `autopilot setup`
+
+### Fixed
+- Removed superpowers plugin check from `doctor` — it was warning all external developers about a Delegance-internal tool they cannot install
+
+## [1.2.3] — 2026-04-21
+
+### Fixed
+- README rewrite: `setup` and `doctor` commands now prominent; config schema accurate; public API section added
+
+## [1.2.2] — 2026-04-21
+
+### Fixed
+- Hook install called from `setup` no longer double-prints stderr; added `silent` option to `runHook()` to suppress output when invoked programmatically
+
+## [1.2.1] — 2026-04-21
+
+### Fixed
+- `bin/autopilot.js` tsx resolution now checks the consumer's `node_modules/.bin/tsx` before falling back to PATH — fixes "tsx not found" on fresh installs
+- npm default test placeholder (`echo "Error: no test specified" && exit 1`) is now detected and replaced with `npm test` instead of being used as the test command
+
 ## [1.2.0] — 2026-04-21
 
 ### Added

package/package.json

@@ -1,8 +1,8 @@
 {
   "name": "@delegance/claude-autopilot",
-  "version": "1.2.7",
+  "version": "1.3.0",
   "type": "module",
-  "description": "Claude Code automation pipeline: spec \u2192 plan \u2192 implement \u2192 validate \u2192 PR",
+  "description": "Claude Code automation pipeline: spec → plan → implement → validate → PR",
   "keywords": [
     "claude",
     "autopilot",
@@ -45,6 +45,7 @@
     "autoregress": "tsx scripts/autoregress.ts"
   },
   "dependencies": {
+    "@anthropic-ai/sdk": "^0.90.0",
     "ajv": "^8",
     "dotenv": ">=16",
     "js-yaml": "^4",

package/presets/go/autopilot.config.yaml

@@ -1,5 +1,5 @@
 configVersion: 1
-reviewEngine: { adapter: codex }
+reviewEngine: { adapter: auto }
 vcsHost: { adapter: github }
 reviewBot: { adapter: cursor }
 protectedPaths:

package/presets/nextjs-supabase/autopilot.config.yaml

@@ -1,5 +1,5 @@
 configVersion: 1
-reviewEngine: { adapter: codex }
+reviewEngine: { adapter: auto }
 vcsHost: { adapter: github }
 migrationRunner: { adapter: supabase }
 reviewBot: { adapter: cursor }

package/presets/python-fastapi/autopilot.config.yaml

@@ -1,5 +1,5 @@
 configVersion: 1
-reviewEngine: { adapter: codex }
+reviewEngine: { adapter: auto }
 vcsHost: { adapter: github }
 reviewBot: { adapter: cursor }
 protectedPaths:

package/presets/rails-postgres/autopilot.config.yaml

@@ -1,5 +1,5 @@
 configVersion: 1
-reviewEngine: { adapter: codex }
+reviewEngine: { adapter: auto }
 vcsHost: { adapter: github }
 reviewBot: { adapter: cursor }
 protectedPaths:

package/presets/t3/autopilot.config.yaml

@@ -1,5 +1,5 @@
 configVersion: 1
-reviewEngine: { adapter: codex }
+reviewEngine: { adapter: auto }
 vcsHost: { adapter: github }
 reviewBot: { adapter: cursor }
 protectedPaths:

package/src/adapters/loader.ts

@@ -13,7 +13,11 @@ export interface LoadAdapterOptions {
 }
 
 const BUILTIN_PATHS: Record<IntegrationPoint, Record<string, string>> = {
-  'review-engine': { codex: './review-engine/codex.ts' },
+  'review-engine': {
+    codex: './review-engine/codex.ts',
+    claude: './review-engine/claude.ts',
+    auto: './review-engine/auto.ts',
+  },
   'vcs-host': { github: './vcs-host/github.ts' },
   'migration-runner': { supabase: './migration-runner/supabase.ts' },
   'review-bot-parser': { cursor: './review-bot-parser/cursor.ts' },

package/src/adapters/review-engine/auto.ts

@@ -0,0 +1,39 @@
+import type { Capabilities } from '../base.ts';
+import type { ReviewEngine, ReviewInput, ReviewOutput } from './types.ts';
+import { AutopilotError } from '../../core/errors.ts';
+
+// Priority order: ANTHROPIC_API_KEY → claude, OPENAI_API_KEY → codex
+async function resolveAdapter(): Promise<ReviewEngine> {
+  if (process.env.ANTHROPIC_API_KEY) {
+    const { claudeAdapter } = await import('./claude.ts');
+    return claudeAdapter;
+  }
+  if (process.env.OPENAI_API_KEY) {
+    const { codexAdapter } = await import('./codex.ts');
+    return codexAdapter;
+  }
+  throw new AutopilotError(
+    'No LLM API key found — set ANTHROPIC_API_KEY (recommended) or OPENAI_API_KEY to enable review',
+    { code: 'auth', provider: 'auto' }
+  );
+}
+
+export const autoAdapter: ReviewEngine = {
+  name: 'auto',
+  apiVersion: '1.0.0',
+
+  getCapabilities(): Capabilities {
+    return { structuredOutput: false, streaming: false, maxContextTokens: 200000, inlineComments: false };
+  },
+
+  estimateTokens(content: string): number {
+    return Math.ceil(content.length / 3.5);
+  },
+
+  async review(input: ReviewInput): Promise<ReviewOutput> {
+    const adapter = await resolveAdapter();
+    return adapter.review(input);
+  },
+};
+
+export default autoAdapter;

package/src/adapters/review-engine/claude.ts

@@ -0,0 +1,126 @@
+import Anthropic from '@anthropic-ai/sdk';
+import type { Finding } from '../../core/findings/types.ts';
+import { AutopilotError } from '../../core/errors.ts';
+import type { Capabilities } from '../base.ts';
+import type { ReviewEngine, ReviewInput, ReviewOutput } from './types.ts';
+
+const DEFAULT_MODEL = 'claude-sonnet-4-6';
+const MAX_OUTPUT_TOKENS = 4096;
+
+// Cost per million tokens (USD) — sonnet-4-6 pricing
+const COST_PER_M_INPUT = 3.0;
+const COST_PER_M_OUTPUT = 15.0;
+
+const SYSTEM_PROMPT_TEMPLATE = `You are a senior software architect reviewing code changes for quality, security, and correctness.
+
+The codebase context:
+{STACK}
+
+Provide structured feedback in exactly this format:
+
+## Review Summary
+One paragraph overall assessment.
+
+## Findings
+
+For each finding, use this format:
+### [CRITICAL|WARNING|NOTE] <short title>
+<explanation>
+**Suggestion:** <actionable fix>
+
+Rules:
+- CRITICAL: Blocks merge (security issues, data loss risks, broken contracts)
+- WARNING: Should address before merging (logic errors, missing error handling, test gaps)
+- NOTE: Improvement suggestion (style, performance, clarity)
+- Maximum 10 findings, ranked by severity
+- Be specific and constructive
+- Reference the file and line when possible`;
+
+export const claudeAdapter: ReviewEngine = {
+  name: 'claude',
+  apiVersion: '1.0.0',
+
+  getCapabilities(): Capabilities {
+    return { structuredOutput: false, streaming: false, maxContextTokens: 200000, inlineComments: false };
+  },
+
+  estimateTokens(content: string): number {
+    return Math.ceil(content.length / 3.5);
+  },
+
+  async review(input: ReviewInput): Promise<ReviewOutput> {
+    const apiKey = process.env.ANTHROPIC_API_KEY;
+    if (!apiKey) {
+      throw new AutopilotError('ANTHROPIC_API_KEY not set', { code: 'auth', provider: 'claude' });
+    }
+
+    const model = (input.context as Record<string, unknown> | undefined)?.['model'] as string | undefined ?? DEFAULT_MODEL;
+    const stack = input.context?.stack ?? 'A web application — stack details unspecified.';
+    const systemPrompt = SYSTEM_PROMPT_TEMPLATE.replace('{STACK}', stack);
+
+    const client = new Anthropic({ apiKey });
+    let response: Anthropic.Message;
+    try {
+      response = await client.messages.create({
+        model,
+        max_tokens: MAX_OUTPUT_TOKENS,
+        system: systemPrompt,
+        messages: [{ role: 'user', content: `Please review the following:\n\n---\n\n${input.content}` }],
+      });
+    } catch (err) {
+      const message = err instanceof Error ? err.message : String(err);
+      const isRateLimit = /rate.limit|429|overloaded/i.test(message);
+      const isAuth = /unauthorized|401|invalid.api.key|authentication/i.test(message);
+      throw new AutopilotError(`Claude review call failed: ${message}`, {
+        code: isAuth ? 'auth' : isRateLimit ? 'rate_limit' : 'transient_network',
+        provider: 'claude',
+        retryable: isRateLimit,
+      });
+    }
+
+    const rawOutput = response.content
+      .filter(b => b.type === 'text')
+      .map(b => (b as Anthropic.TextBlock).text)
+      .join('');
+
+    const costUSD = response.usage
+      ? (response.usage.input_tokens / 1_000_000) * COST_PER_M_INPUT +
+        (response.usage.output_tokens / 1_000_000) * COST_PER_M_OUTPUT
+      : undefined;
+
+    return {
+      findings: parseClaudeOutput(rawOutput),
+      rawOutput,
+      usage: response.usage
+        ? { input: response.usage.input_tokens, output: response.usage.output_tokens, costUSD }
+        : undefined,
+    };
+  },
+};
+
+export default claudeAdapter;
+
+function parseClaudeOutput(output: string): Finding[] {
+  const findings: Finding[] = [];
+  const regex = /### \[(CRITICAL|WARNING|NOTE)\]\s*(.+?)(?=\n### \[|## Review Summary|$)/gs;
+  let match: RegExpExecArray | null;
+  while ((match = regex.exec(output)) !== null) {
+    const severity = match[1]!.toLowerCase() as Finding['severity'];
+    const body = match[2]!.trim();
+    const titleEnd = body.indexOf('\n');
+    const title = (titleEnd > 0 ? body.slice(0, titleEnd) : body).trim();
+    const suggestion = body.match(/\*\*Suggestion:\*\*\s*(.+)/s)?.[1]?.trim();
+    findings.push({
+      id: `claude-${findings.length}`,
+      source: 'review-engine',
+      severity,
+      category: 'claude-review',
+      file: '<unspecified>',
+      message: title,
+      suggestion,
+      protectedPath: false,
+      createdAt: new Date().toISOString(),
+    });
+  }
+  return findings;
+}

package/src/cli/run.ts

@@ -107,13 +107,13 @@ export async function runCommand(options: RunCommandOptions = {}): Promise<numbe
     return 0;
   }
 
-  // Load review engine (optional — skip if no OPENAI_API_KEY or not configured)
+  // Load review engine (optional — skip gracefully if no API key configured)
   let reviewEngine: ReviewEngine | undefined;
   if (config.reviewEngine) {
     const ref = typeof config.reviewEngine === 'string' ? config.reviewEngine : config.reviewEngine.adapter;
-    const hasKey = !!(process.env.OPENAI_API_KEY);
-    if (!hasKey && ref === 'codex') {
-      console.log(fmt('yellow', '\n  [run] OPENAI_API_KEY not set — Codex review step will be skipped'));
+    const hasAnyKey = !!(process.env.ANTHROPIC_API_KEY || process.env.OPENAI_API_KEY);
+    if (!hasAnyKey && (ref === 'auto' || ref === 'claude' || ref === 'codex')) {
+      console.log(fmt('yellow', '\n  [run] No LLM API key found — set ANTHROPIC_API_KEY or OPENAI_API_KEY to enable review'));
     } else {
       try {
         reviewEngine = await loadAdapter<ReviewEngine>({