edsger 0.77.0 → 0.78.0

package/dist/phases/product-techniques/prompts.js

@@ -1,66 +0,0 @@
-/**
- * Prompts for the product-level techniques phase.
- *
- * Agent's job: explore the cloned product repo and write a markdown catalogue
- * of the techniques it actually uses — languages, frameworks, patterns, state
- * idioms, build/deploy choices, plus the clever non-obvious bits. Focused on
- * what a new engineer needs to recognize and recreate, not feature lists.
- *
- * The final result is submitted via the `submit_techniques` MCP tool, NOT a
- * fenced JSON block. Tool calls let the SDK enforce the schema with Zod and
- * let the agent self-correct on validation errors. See mcp-server.ts.
- */
-export function createProductTechniquesSystemPrompt() {
-    return `You are a senior staff engineer cataloguing the techniques a product's codebase uses.
-
-The current working directory is a fresh clone of the product's repository. Use Glob/Grep/Read (and Bash for git/log when helpful) to explore it before writing.
-
-Your audience: a new engineer joining the team. They need to recognize WHICH techniques this repo employs — what frameworks, what patterns, what idioms — so they can read code without surprise and contribute in the codebase's existing style. This is not a feature list and not an architecture proposal. Anchor every technique in the actual code; reference real file paths.
-
-## Output protocol
-
-When you are ready to submit, call the \`submit_techniques\` tool EXACTLY ONCE with:
-- \`summary\` — 1-2 sentence summary suitable for a tab header (plain text, no markdown).
-- \`content\` — the full markdown body, structured as described below.
-
-Do NOT also paste the same content as a fenced code block — the tool call is the only channel for the result. If validation fails, the tool response tells you what to fix; call the tool again with corrected data.
-
-## Required sections in \`content\`
-
-The markdown body MUST contain these H2 sections, in this order:
-
-1. **## Languages & Runtime** — languages used (with versions if pinned), runtime targets (Node / Bun / Deno / browser / native / etc), package manager.
-2. **## Frameworks & Libraries** — the major frameworks and libraries this repo depends on, and what each one is doing here. Don't just list package names; explain the role. Group by area (UI, server, data, infra) if it helps.
-3. **## Architecture Patterns** — the architectural choices the code embodies: layering / module boundaries / dependency direction / how features are organised. Include one Mermaid \`flowchart TD\` showing the dominant pattern (≤ 10 nodes). Name the pattern when you can ("repository pattern", "feature-sliced design", "hexagonal", "BFF", etc.).
-4. **## State & Data Techniques** — state management, data fetching, caching, optimistic updates, validation, type-safety across boundaries. How is server state vs client state handled? How are forms managed? How is data flowing between server and client?
-5. **## Build & Deploy Techniques** — bundler / compiler choices, code-splitting, environment handling, monorepo tooling, CI/CD specifics, deployment target (edge / serverless / containers / static).
-6. **## Notable Techniques** — the genuinely interesting bits that aren't obvious from the stack: clever abstractions, performance optimizations, security hardening, custom hooks/utilities worth knowing, intentional deviations from defaults, surprising workarounds. Each one: what it is, where it lives (file path), why it matters. This section is the whole point — be specific and useful here.
-
-The validator will reject content missing the "Languages & Runtime" or "Notable Techniques" headings.
-
-## Rules
-
-- Use relative file paths from the repo root (e.g. \`src/services/auth.ts\`). Link to specific files where the technique is most visible.
-- Mermaid blocks must be valid (no unclosed quotes, no unsupported node shapes). Prefer simple \`flowchart TD\` syntax.
-- Don't invent or guess. If a section has nothing distinctive ("just a stock CRA app"), say so briefly — don't pad.
-- Focus on TECHNIQUES, not features. Bad: "the app has a login screen". Good: "uses next-auth's credentials provider with a custom JWT callback at src/lib/auth.ts:42".
-- The whole document should fit in one screen-readable scroll — aim for ~1500-3000 words of content. Less is fine for small repos.`;
-}
-export function createProductTechniquesUserPrompt(context) {
-    const lines = [];
-    lines.push(`# Product: ${context.productName}`);
-    if (context.productDescription) {
-        lines.push('');
-        lines.push('## Description');
-        lines.push(context.productDescription);
-    }
-    if (context.guidance && context.guidance.trim()) {
-        lines.push('');
-        lines.push('## Reviewer guidance (focus or exclusions)');
-        lines.push(context.guidance.trim());
-    }
-    lines.push('');
-    lines.push('## Task');
-    lines.push('Explore the cloned repository and produce the techniques catalogue, then submit it via the `submit_techniques` MCP tool as specified in your system prompt. Pay particular attention to the "Notable Techniques" section — that is the real value to the reader.');
-    return lines.join('\n');
-}

package/dist/phases/product-techniques/types.d.ts

@@ -1,13 +0,0 @@
-/**
- * Shape returned by the agent via the `submit_techniques` MCP tool. Mirrors
- * the Zod schema in `mcp-server.ts` — kept duplicated as a plain TS type so
- * consumers (the phase orchestrator, persistence helpers, tests) don't have
- * to inflate Zod into their dependency graph.
- */
-export interface TechniquesExtraction {
-    summary: string;
-    content: string;
-}
-export declare function isTechniquesExtraction(v: unknown): v is TechniquesExtraction;
-export declare const TECHNIQUES_SUMMARY_MAX = 500;
-export declare const TECHNIQUES_CONTENT_MAX = 200000;

package/dist/phases/product-techniques/types.js

@@ -1,13 +0,0 @@
-export function isTechniquesExtraction(v) {
-    if (!v || typeof v !== 'object') {
-        return false;
-    }
-    const obj = v;
-    return typeof obj.summary === 'string' && typeof obj.content === 'string';
-}
-// Defensive caps mirroring the DB CHECK constraints from
-// 20260521000000_create_product_techniques.sql. Keeping the limits here
-// too lets the MCP tool reject oversized output with an actionable error
-// message instead of getting a Postgres constraint violation.
-export const TECHNIQUES_SUMMARY_MAX = 500;
-export const TECHNIQUES_CONTENT_MAX = 200_000;

package/dist/phases/product-test-cases/index.d.ts

@@ -1,25 +0,0 @@
-/**
- * Product-test-cases phase: clone the product's repo, ask Claude to draft a
- * product-level regression suite (deduping against existing cases), and save
- * the new ones as drafts via MCP. Approved test cases are NEVER touched —
- * only draft/pending_approval entries may be replaced.
- *
- * Mirrors the find-bugs pattern: clone to ~/edsger/product-test-cases-<id>,
- * run a bounded Claude session, persist via MCP, cleanup on success.
- */
-export interface GenerateProductTestCasesOptions {
-    productId: string;
-    githubToken: string;
-    owner: string;
-    repo: string;
-    branch?: string;
-    verbose?: boolean;
-}
-export interface GenerateProductTestCasesResult {
-    status: 'success' | 'error';
-    message: string;
-    createdCount?: number;
-    deletedCount?: number;
-    summary?: string;
-}
-export declare function generateProductTestCases(options: GenerateProductTestCasesOptions): Promise<GenerateProductTestCasesResult>;

package/dist/phases/product-test-cases/index.js

@@ -1,174 +0,0 @@
-/**
- * Product-test-cases phase: clone the product's repo, ask Claude to draft a
- * product-level regression suite (deduping against existing cases), and save
- * the new ones as drafts via MCP. Approved test cases are NEVER touched —
- * only draft/pending_approval entries may be replaced.
- *
- * Mirrors the find-bugs pattern: clone to ~/edsger/product-test-cases-<id>,
- * run a bounded Claude session, persist via MCP, cleanup on success.
- */
-import { query } from '@anthropic-ai/claude-agent-sdk';
-import { batchDeleteTestCases } from '../../api/issues/batch-operations.js';
-import { createProductTestCases, getProductTestCases, } from '../../api/products/test-cases.js';
-import { DEFAULT_MODEL } from '../../constants.js';
-import { logError, logInfo, logSuccess, logWarning, } from '../../utils/logger.js';
-import { cleanupIssueRepo, cloneIssueRepo, ensureWorkspaceDir, syncRepoToRef, } from '../../workspace/workspace-manager.js';
-import { detectDefaultBranch } from '../find-shared/git.js';
-import { fetchProductBasics } from '../find-shared/mcp.js';
-import { createScanStateModule } from '../find-shared/scan-state.js';
-import { createPromptGenerator, extractTextFromContent, tryExtractResult, } from '../pr-shared/agent-utils.js';
-import { createProductTestCasesSystemPrompt, createProductTestCasesUserPrompt, } from './prompts.js';
-import { isProductTestCasesAgentResult, } from './types.js';
-const WORKSPACE_KEY = 'product-test-cases';
-// Generation is read-heavy and open-scope. 200 turns matches find-bugs and is
-// enough for a mid-sized repo while still bounding cost.
-const MAX_TURNS = 200;
-// Per-product lock so two concurrent invocations (e.g. user clicks twice in
-// the UI, or CLI + desktop fire at once) don't race on the shared clone dir.
-// We don't persist any state besides the lock — generation is stateless.
-const lockModule = createScanStateModule({
-    dirName: 'product-test-cases-state',
-});
-// UUID regex matching MCP-issued ids — defensive filter before we trust
-// strings the agent puts in deleted_test_case_ids.
-const UUID_RE = /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i;
-// eslint-disable-next-line complexity
-export async function generateProductTestCases(options) {
-    const { productId, githubToken, owner, repo, verbose } = options;
-    logInfo(`Starting product test-cases generation for product ${productId} (${owner}/${repo})`);
-    const lock = lockModule.acquireLock(productId);
-    if (!lock) {
-        logWarning(`Another product test-cases generation is already running for product ${productId}; skipping.`);
-        return {
-            status: 'error',
-            message: 'Another product test-cases generation is already running for this product',
-        };
-    }
-    let repoPath;
-    let succeeded = false;
-    try {
-        const workspaceRoot = ensureWorkspaceDir();
-        const repoKey = `${WORKSPACE_KEY}-${productId}`;
-        ({ repoPath } = cloneIssueRepo(workspaceRoot, repoKey, owner, repo, githubToken));
-        const branch = options.branch ?? detectDefaultBranch(repoPath);
-        logInfo(`Syncing ${owner}/${repo} to branch ${branch}`);
-        syncRepoToRef(repoPath, { branch }, githubToken);
-        const [product, existing] = await Promise.all([
-            fetchProductBasics(productId),
-            getProductTestCases(productId, verbose),
-        ]);
-        const approved = [];
-        const replaceable = [];
-        for (const tc of existing) {
-            const slot = {
-                id: tc.id,
-                name: tc.name,
-                description: tc.description,
-                is_critical: tc.is_critical,
-                status: tc.status ?? 'draft',
-            };
-            if (slot.status === 'approved') {
-                approved.push(slot);
-            }
-            else {
-                replaceable.push(slot);
-            }
-        }
-        const replaceableIds = new Set(replaceable.map((tc) => tc.id));
-        logInfo(`Existing test cases: ${approved.length} approved (locked), ${replaceable.length} replaceable`);
-        const systemPrompt = createProductTestCasesSystemPrompt();
-        const userPrompt = createProductTestCasesUserPrompt({
-            productName: product.name,
-            productDescription: product.description,
-            approvedTestCases: approved,
-            replaceableTestCases: replaceable,
-        });
-        let lastAssistantResponse = '';
-        let agentResult = null;
-        logInfo('Running Claude agent to draft test cases...');
-        for await (const message of query({
-            prompt: createPromptGenerator(userPrompt),
-            options: {
-                systemPrompt: {
-                    type: 'preset',
-                    preset: 'claude_code',
-                    append: systemPrompt,
-                },
-                model: DEFAULT_MODEL,
-                maxTurns: MAX_TURNS,
-                permissionMode: 'bypassPermissions',
-                cwd: repoPath,
-            },
-        })) {
-            if (message.type === 'assistant') {
-                lastAssistantResponse += extractTextFromContent(message.message?.content ?? [], verbose);
-                continue;
-            }
-            if (message.type !== 'result') {
-                continue;
-            }
-            const responseText = message.subtype === 'success'
-                ? message.result || lastAssistantResponse
-                : lastAssistantResponse;
-            const parsed = tryExtractResult(responseText, 'test_cases_result');
-            if (isProductTestCasesAgentResult(parsed)) {
-                agentResult = parsed;
-            }
-            else if (message.subtype !== 'success') {
-                logError(`Agent run incomplete: ${message.subtype}`);
-            }
-        }
-        if (!agentResult) {
-            return {
-                status: 'error',
-                message: 'Test cases generation failed: could not parse a test_cases_result from the agent',
-            };
-        }
-        // Apply deletions. Filter to only ids that are valid UUIDs AND belong
-        // to the replaceable set — never trust the agent's word that an id
-        // is deletable. Approved cases must never be removed.
-        let deletedCount = 0;
-        const requestedDeletes = (agentResult.deleted_test_case_ids ?? []).filter((id) => typeof id === 'string' && UUID_RE.test(id) && replaceableIds.has(id));
-        if (requestedDeletes.length > 0) {
-            logInfo(`Deleting ${requestedDeletes.length} obsolete replaceable test cases...`);
-            const ok = await batchDeleteTestCases(requestedDeletes, verbose);
-            if (ok) {
-                deletedCount = requestedDeletes.length;
-            }
-            else {
-                logWarning('Some deletions failed; leaving any remaining cases in place.');
-            }
-        }
-        const { createdIds } = await createProductTestCases(productId, agentResult.created_test_cases.map((tc) => ({
-            name: tc.name,
-            description: tc.description,
-            is_critical: tc.is_critical ?? false,
-        })), verbose);
-        logSuccess(`Created ${createdIds.length} test cases, deleted ${deletedCount} obsolete entries. ${agentResult.summary}`);
-        succeeded = true;
-        return {
-            status: 'success',
-            message: `Generated ${createdIds.length} test cases (deleted ${deletedCount})`,
-            createdCount: createdIds.length,
-            deletedCount,
-            summary: agentResult.summary,
-        };
-    }
-    catch (error) {
-        const message = error instanceof Error ? error.message : String(error);
-        logError(`Product test cases generation failed: ${message}`);
-        return {
-            status: 'error',
-            message: `Product test cases generation failed: ${message}`,
-        };
-    }
-    finally {
-        if (succeeded) {
-            cleanupIssueRepo(repoPath);
-        }
-        else if (repoPath) {
-            logInfo(`Workspace preserved for inspection: ${repoPath}`);
-        }
-        lock.release();
-    }
-}

package/dist/phases/product-test-cases/prompts.d.ts

@@ -1,24 +0,0 @@
-/**
- * Prompts for the product-level test cases phase.
- *
- * Agent's job: explore the cloned product repo and propose a regression-suite
- * worth of test cases for the whole product (not a single issue). Existing
- * test cases — both approved (locked) and unapproved (replaceable) — are
- * included in the context so the agent doesn't duplicate work and so the
- * unapproved set can be refreshed when the codebase has drifted.
- */
-export interface ExistingProductTestCase {
-    id: string;
-    name: string;
-    description: string;
-    is_critical: boolean;
-    status: string;
-}
-export interface ProductTestCasesPromptContext {
-    productName: string;
-    productDescription?: string;
-    approvedTestCases: ExistingProductTestCase[];
-    replaceableTestCases: ExistingProductTestCase[];
-}
-export declare function createProductTestCasesSystemPrompt(): string;
-export declare function createProductTestCasesUserPrompt(context: ProductTestCasesPromptContext): string;

package/dist/phases/product-test-cases/prompts.js

@@ -1,80 +0,0 @@
-/**
- * Prompts for the product-level test cases phase.
- *
- * Agent's job: explore the cloned product repo and propose a regression-suite
- * worth of test cases for the whole product (not a single issue). Existing
- * test cases — both approved (locked) and unapproved (replaceable) — are
- * included in the context so the agent doesn't duplicate work and so the
- * unapproved set can be refreshed when the codebase has drifted.
- */
-export function createProductTestCasesSystemPrompt() {
-    return `You are a senior QA engineer drafting a product-level regression test suite.
-
-The current working directory is a fresh clone of the product's repository. Use Glob/Grep/Read to explore it and understand the surface area: entry points, public APIs, user-facing flows, critical paths. Look at recent commits with Bash if it helps.
-
-Output a single JSON result of the form:
-\`\`\`json
-{
-  "test_cases_result": {
-    "summary": "Short 1-2 sentence summary of what you covered",
-    "created_test_cases": [
-      {
-        "name": "Concise name (max 200 chars)",
-        "description": "Markdown describing Given/When/Then steps + expected result",
-        "is_critical": true
-      }
-    ],
-    "deleted_test_case_ids": [
-      "uuid-of-replaceable-test-case-you-want-removed"
-    ],
-    "deletion_reasons": {
-      "uuid": "Why this test case is no longer relevant"
-    }
-  }
-}
-\`\`\`
-
-Rules:
-- DO NOT duplicate any test case in the "Approved test cases" or "Replaceable test cases" sections. If an existing case already covers a behaviour, skip it.
-- You MAY propose deletions ONLY for entries in the "Replaceable test cases" list. Approved test cases are locked by the human reviewer and cannot be touched.
-- Mark \`is_critical: true\` only for must-pass flows (auth, payments, data loss risks, primary user journeys).
-- Aim for 5-30 high-signal cases total. Quality over quantity.
-- Each test case must be executable by a human tester reading the description alone.
-- The "deleted_test_case_ids" and "deletion_reasons" fields are optional; omit or use [] / {} when you have nothing to delete.
-- Wrap the final JSON in a single \`\`\`json fenced block. No extra prose after it.`;
-}
-export function createProductTestCasesUserPrompt(context) {
-    const lines = [];
-    lines.push(`# Product: ${context.productName}`);
-    if (context.productDescription) {
-        lines.push('');
-        lines.push(`## Description`);
-        lines.push(context.productDescription);
-    }
-    lines.push('');
-    lines.push('## Approved test cases (locked — do not duplicate, cannot delete)');
-    if (context.approvedTestCases.length === 0) {
-        lines.push('_(none)_');
-    }
-    else {
-        for (const tc of context.approvedTestCases) {
-            lines.push(`- **${tc.name}**${tc.is_critical ? ' [critical]' : ''}`);
-            lines.push(`  ${tc.description.replace(/\n/g, '\n  ')}`);
-        }
-    }
-    lines.push('');
-    lines.push('## Replaceable test cases (draft / pending_approval — may be deleted if obsolete or duplicated)');
-    if (context.replaceableTestCases.length === 0) {
-        lines.push('_(none)_');
-    }
-    else {
-        for (const tc of context.replaceableTestCases) {
-            lines.push(`- \`${tc.id}\` (${tc.status})${tc.is_critical ? ' [critical]' : ''} — **${tc.name}**`);
-            lines.push(`  ${tc.description.replace(/\n/g, '\n  ')}`);
-        }
-    }
-    lines.push('');
-    lines.push('## Task');
-    lines.push('Explore the repository (Glob/Grep/Read/Bash) and propose product-level regression test cases. Follow the JSON output contract in your system prompt.');
-    return lines.join('\n');
-}

package/dist/phases/product-test-cases/types.d.ts

@@ -1,17 +0,0 @@
-/**
- * Wire shape for the product-test-cases agent output. The agent is asked to
- * emit `{ test_cases_result: { ... } }`; we validate it loosely so a single
- * stray field doesn't kill an otherwise-good run.
- */
-export interface ProductTestCaseDraft {
-    name: string;
-    description: string;
-    is_critical?: boolean;
-}
-export interface ProductTestCasesAgentResult {
-    summary: string;
-    created_test_cases: ProductTestCaseDraft[];
-    deleted_test_case_ids?: string[];
-    deletion_reasons?: Record<string, string>;
-}
-export declare function isProductTestCasesAgentResult(value: unknown): value is ProductTestCasesAgentResult;

package/dist/phases/product-test-cases/types.js

@@ -1,27 +0,0 @@
-/**
- * Wire shape for the product-test-cases agent output. The agent is asked to
- * emit `{ test_cases_result: { ... } }`; we validate it loosely so a single
- * stray field doesn't kill an otherwise-good run.
- */
-export function isProductTestCasesAgentResult(value) {
-    if (!value || typeof value !== 'object') {
-        return false;
-    }
-    const v = value;
-    if (typeof v.summary !== 'string') {
-        return false;
-    }
-    if (!Array.isArray(v.created_test_cases)) {
-        return false;
-    }
-    for (const tc of v.created_test_cases) {
-        if (!tc || typeof tc !== 'object') {
-            return false;
-        }
-        const t = tc;
-        if (typeof t.name !== 'string' || typeof t.description !== 'string') {
-            return false;
-        }
-    }
-    return true;
-}

package/dist/phases/pull-request/creator.d.ts

@@ -1,30 +0,0 @@
-/**
- * Pull Request Creator for the development workflow
- * Creates pull requests from issue branches after successful testing
- */
-export interface PullRequestConfig {
-    readonly githubToken: string;
-    readonly owner: string;
-    readonly repo: string;
-    readonly baseBranch?: string;
-    readonly verbose?: boolean;
-}
-export interface IssueInfo {
-    readonly id: string;
-    readonly name: string;
-    readonly description: string;
-    readonly productId?: string;
-    readonly technicalDesign?: string;
-    readonly testResult?: string;
-    readonly testReportUrl?: string;
-}
-export interface PullRequestResult {
-    readonly success: boolean;
-    readonly pullRequestUrl?: string;
-    readonly pullRequestNumber?: number;
-    readonly error?: string;
-}
-/**
- * Create a pull request for the issue
- */
-export declare function createPullRequest(config: PullRequestConfig, issue: IssueInfo): Promise<PullRequestResult>;