@mjasnikovs/pi-task 0.14.6 → 0.14.8

package/dist/task/auto-orchestrator.js

@@ -29,6 +29,19 @@ import { getParentContextWindow, resolveContextUsage } from './context-usage.js'
 const MAX_CLARIFY_QUESTIONS = 8;
 // Matches pi's @-file completion token (a path after @, until whitespace).
 const MENTION_RE = /(?:^|\s)@([^\s]+)/g;
+// Trailing punctuation a user naturally types AFTER an @-mention when it sits in
+// prose — "Implement @design.md, reuse…" or "see @spec.md." — which the greedy
+// [^\s]+ above would otherwise swallow into the path. Left unstripped, the
+// resulting "design.md," resolves to no file, expansion is silently skipped, and
+// the planner reasons over a one-line "Implement @design.md" with NO spec inline
+// → it fabricates generic questions/tasks the spec never called for (validated:
+// a stray comma turned a 32KB design into a contentless prompt). None of these
+// chars are legitimate trailing characters of a referenced doc path.
+const MENTION_TRAILING_PUNCT = /[.,;:!?)\]}>"']+$/;
+/** The cleaned path token of an @-mention: greedy match minus trailing prose punctuation. */
+function mentionPath(token) {
+    return token.replace(MENTION_TRAILING_PUNCT, '');
+}
 /**
  * Expand any @file references in the feature text by appending each referenced
  * file's contents, so the planning children (clarify, decompose) always see the
@@ -42,8 +55,8 @@ export async function expandFeatureMentions(cwd, feature) {
     const seen = new Set();
     const blocks = [];
     for (const m of feature.matchAll(MENTION_RE)) {
-        const rel = m[1];
-        if (seen.has(rel))
+        const rel = mentionPath(m[1]);
+        if (rel === '' || seen.has(rel))
             continue;
         seen.add(rel);
         try {
@@ -68,8 +81,8 @@ export async function readableMentions(cwd, feature) {
     const out = [];
     const seen = new Set();
     for (const m of feature.matchAll(MENTION_RE)) {
-        const rel = m[1];
-        if (seen.has(rel))
+        const rel = mentionPath(m[1]);
+        if (rel === '' || seen.has(rel))
             continue;
         seen.add(rel);
         try {

package/dist/task/phases.js

@@ -6,6 +6,7 @@ import { docsFocused } from '../workers/docs-core.js';
 import { fetchFocused } from '../workers/fetch-core.js';
 import { formatNpmVersionSection } from '../workers/npm-version.js';
 import { runWorker } from '../workers/pi-worker-core.js';
+import { findPhantomImports, formatApiCorrections } from '../workers/phantom-imports.js';
 import { search as defaultSearch } from '../workers/search-core.js';
 import { extractEnrichTargets } from './enrichment.js';
 import { getFileInventory } from './file-inventory.js';
@@ -663,6 +664,17 @@ export const PHASES = [
             const tVerify = Date.now();
             const out = await phaseVerifyTooling(d, rawResearch);
             d.recordSubStep?.('verify-tooling', Date.now() - tVerify);
+            // Deterministically verify every runtime builtin specifier the refined
+            // task names (`bun:sql`, `node:…`) against the installed types. A doc can
+            // confidently name a module that does not exist; left unchecked it rides
+            // through every phase and the implementer fabricates a `declare module`
+            // shim to compile it. Append the corrections so compose folds them into
+            // CONSTRAINTS. No LLM cost and silent when nothing is wrong.
+            const corrections = formatApiCorrections(findPhantomImports(p.refined, d.cwd));
+            if (corrections) {
+                d.logDebug?.(`phantom imports flagged:\n${corrections}`);
+                return `${out}\n\n${corrections}`;
+            }
             return out;
         }
     },

package/dist/task/prompts.js

@@ -309,6 +309,8 @@ VERIFY must exercise the surface area the task actually touches. Draw VERIFY com
 
 If TOOLING is empty for a category the change clearly touches, still include the best-effort standard command for that ecosystem (e.g. \`npx tsc --noEmit\` for a TS repo with no script) and note that the receiving agent may need to install it.
 
+If the research contains an "API CORRECTIONS" section, it is AUTHORITATIVE — each line was verified against the installed types. For every correction: use the import it prescribes, never the specifier it marks non-existent, and add a CONSTRAINT recording it verbatim (e.g. Use \`import { sql } from "bun"\`; \`bun:sql\` is not a module — do not import it or declare a module for it). This overrides any conflicting identifier carried up from the refined task or the spec doc.
+
 Refined task:
 ${refined}
 

package/dist/workers/phantom-imports.d.ts

@@ -0,0 +1,35 @@
+export interface PhantomImport {
+    /** The bad specifier exactly as written, e.g. `bun:sql`. */
+    spec: string;
+    /** Real `<runtime>:` modules the types DO declare (evidence for the correction). */
+    realModules: string[];
+    /** Corrective guidance to inject into the spec. */
+    suggestion: string;
+}
+/** Every distinct runtime-namespace specifier mentioned in the text, in first-seen order. */
+export declare function extractRuntimeSpecifiers(text: string): string[];
+export interface RuntimeImportVerdict {
+    spec: string;
+    real: boolean;
+    /** When phantom: a base-module symbol matching the submodule leaf, if the types
+     *  declare one (so the correction can name the real import). */
+    baseSymbol: string | null;
+    /** Real `<runtime>:` modules declared in the type text. */
+    realModules: string[];
+}
+/**
+ * Pure verdict over a runtime's concatenated type text — no I/O, so the rule is
+ * unit-testable with synthetic declarations.
+ */
+export declare function classifyRuntimeImport(spec: string, runtime: string, sub: string, typeText: string): RuntimeImportVerdict;
+/** Default loader: resolve the runtime's installed types and read their .d.ts. */
+export declare function loadRuntimeTypeText(runtime: string, cwd: string): string | null;
+/**
+ * Scan `text` for runtime-namespace specifiers and return the ones the installed
+ * types do not declare. `loadText` is injectable for tests; in production it reads
+ * the runtime's type package. A runtime whose types can't be loaded is skipped
+ * (we never flag what we can't verify), so this is silent when types are absent.
+ */
+export declare function findPhantomImports(text: string, cwd: string, loadText?: (runtime: string, cwd: string) => string | null): PhantomImport[];
+/** Render flagged phantoms as an authoritative research section, or '' if none. */
+export declare function formatApiCorrections(phantoms: PhantomImport[]): string;

package/dist/workers/phantom-imports.js

@@ -0,0 +1,187 @@
+/**
+ * Detect phantom runtime imports — `bun:sql`, `node:nope`, etc. — referenced in a
+ * task/spec but NOT actually declared by the runtime's installed types.
+ *
+ * A real `<runtime>:<sub>` builtin appears as `declare module "<runtime>:<sub>"`
+ * in the runtime's type files (bun-types declares bun:sqlite/bun:test/bun:ffi/…;
+ * @types/node declares node:fs/…). A specifier with no such declaration is a
+ * hallucination — the symbol, if it exists at all, lives on the base module. This
+ * is the exact `bun:sql` failure class: a design doc invented `bun:sql`, every
+ * phase echoed it, and the implementer fabricated a `declare module "bun:sql"`
+ * shim to make it compile. Verifying the specifier against the installed types
+ * catches it deterministically, with no false positives and no LLM call.
+ */
+import * as fs from 'node:fs';
+import * as path from 'node:path';
+import { resolvePackage, splitRuntimeNamespace, detectTypesRedirect, typesPackageName, hasTypeFiles, isDtsFile, ResolveError } from './docs-resolve.js';
+// Runtime builtin specifiers as they appear in prose/code: `bun:sql`,
+// `node:fs/promises`. Bounded to the runtimes splitRuntimeNamespace accepts.
+const SPEC_RE = /\b(?:bun|node|deno):[a-z0-9][a-z0-9/_-]*/gi;
+const DECLARE_MODULE_RE = /declare module "([^"]+)"/g;
+/** Every distinct runtime-namespace specifier mentioned in the text, in first-seen order. */
+export function extractRuntimeSpecifiers(text) {
+    const seen = new Set();
+    for (const m of text.matchAll(SPEC_RE)) {
+        // Normalise: drop a trailing punctuation the word boundary may include is
+        // already excluded by the class; lowercase the runtime only (subpaths are
+        // case-sensitive). splitRuntimeNamespace lowercases the runtime itself.
+        if (!seen.has(m[0]))
+            seen.add(m[0]);
+    }
+    return [...seen];
+}
+/**
+ * Pure verdict over a runtime's concatenated type text — no I/O, so the rule is
+ * unit-testable with synthetic declarations.
+ */
+export function classifyRuntimeImport(spec, runtime, sub, typeText) {
+    const declared = new Set();
+    for (const m of typeText.matchAll(DECLARE_MODULE_RE))
+        declared.add(m[1]);
+    const real = declared.has(spec);
+    const realModules = [...declared].filter(d => d.startsWith(`${runtime}:`)).sort();
+    let baseSymbol = null;
+    if (!real) {
+        const leaf = sub.split('/').pop() ?? sub;
+        // Does the base runtime module declare a symbol named like the submodule
+        // leaf (e.g. `bun:sql` → `const sql` / `class SQL` in `declare module "bun"`)?
+        // Case-insensitive so `sql` matches the `SQL` class too.
+        const re = new RegExp(`\\b(?:const|class|function|let|var|namespace|interface|type)\\s+(${escapeRe(leaf)})\\b`, 'i');
+        const m = re.exec(typeText);
+        baseSymbol = m ? m[1] : null;
+    }
+    return { spec, real, baseSymbol, realModules };
+}
+function escapeRe(s) {
+    return s.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
+}
+function suggestionFor(v, runtime) {
+    if (v.baseSymbol) {
+        return (`\`${v.spec}\` is NOT a module — the symbol is exported from the base runtime `
+            + `module: use \`import { ${v.baseSymbol} } from "${runtime}"\` (never import from `
+            + `\`${v.spec}\` and never declare a module for it).`);
+    }
+    const list = v.realModules.length > 0 ?
+        ` The real \`${runtime}:\` modules are: ${v.realModules.join(', ')}.`
+        : '';
+    return (`\`${v.spec}\` is NOT a real module.${list} Import the needed symbol from `
+        + `"${runtime}" or verify the correct specifier with pi-worker-docs; do not declare a module for it.`);
+}
+/** Sync resolution of a runtime to the package that actually holds its type
+ *  declarations (bun -> @types/bun -> bun-types), bounded to a few hops. No
+ *  auto-install: a runtime whose types aren't installed simply can't be verified
+ *  (returns null), so we never flag what we can't prove. */
+function resolveRuntimeTypesRoot(runtime, cwd) {
+    let cur;
+    try {
+        cur = resolvePackage(runtime, cwd);
+    }
+    catch (err) {
+        if (err instanceof ResolveError)
+            return null;
+        throw err;
+    }
+    const visited = new Set([cur.name, runtime]);
+    for (let hop = 0; hop < 3; hop++) {
+        let next = detectTypesRedirect(cur);
+        if (next && visited.has(next))
+            next = null;
+        if (!next && !hasTypeFiles(cur.root)) {
+            const types = typesPackageName(cur.name);
+            if (types && !visited.has(types))
+                next = types;
+        }
+        if (!next)
+            break;
+        visited.add(next);
+        try {
+            cur = resolvePackage(next, cwd);
+        }
+        catch {
+            break;
+        }
+    }
+    return hasTypeFiles(cur.root) ? cur.root : null;
+}
+const MAX_TYPE_BYTES = 4_000_000;
+function readRuntimeTypeText(root) {
+    const parts = [];
+    let total = 0;
+    const stack = [root];
+    while (stack.length) {
+        const dir = stack.pop();
+        let entries;
+        try {
+            entries = fs.readdirSync(dir, { withFileTypes: true });
+        }
+        catch {
+            continue;
+        }
+        for (const entry of entries) {
+            if (entry.name === 'node_modules')
+                continue;
+            const full = path.join(dir, entry.name);
+            if (entry.isDirectory()) {
+                stack.push(full);
+            }
+            else if (entry.isFile() && isDtsFile(entry.name)) {
+                try {
+                    const raw = fs.readFileSync(full, 'utf8');
+                    total += raw.length;
+                    parts.push(raw);
+                    if (total >= MAX_TYPE_BYTES)
+                        return parts.join('\n');
+                }
+                catch {
+                    // unreadable file — skip
+                }
+            }
+        }
+    }
+    return parts.join('\n');
+}
+/** Default loader: resolve the runtime's installed types and read their .d.ts. */
+export function loadRuntimeTypeText(runtime, cwd) {
+    const root = resolveRuntimeTypesRoot(runtime, cwd);
+    if (!root)
+        return null;
+    const text = readRuntimeTypeText(root);
+    return text.length > 0 ? text : null;
+}
+/**
+ * Scan `text` for runtime-namespace specifiers and return the ones the installed
+ * types do not declare. `loadText` is injectable for tests; in production it reads
+ * the runtime's type package. A runtime whose types can't be loaded is skipped
+ * (we never flag what we can't verify), so this is silent when types are absent.
+ */
+export function findPhantomImports(text, cwd, loadText = loadRuntimeTypeText) {
+    const out = [];
+    const typeTextByRuntime = new Map();
+    for (const spec of extractRuntimeSpecifiers(text)) {
+        const ns = splitRuntimeNamespace(spec);
+        if (!ns)
+            continue;
+        if (!typeTextByRuntime.has(ns.runtime)) {
+            typeTextByRuntime.set(ns.runtime, loadText(ns.runtime, cwd));
+        }
+        const typeText = typeTextByRuntime.get(ns.runtime);
+        if (!typeText)
+            continue;
+        const verdict = classifyRuntimeImport(spec, ns.runtime, ns.sub, typeText);
+        if (verdict.real)
+            continue;
+        out.push({
+            spec,
+            realModules: verdict.realModules,
+            suggestion: suggestionFor(verdict, ns.runtime)
+        });
+    }
+    return out;
+}
+/** Render flagged phantoms as an authoritative research section, or '' if none. */
+export function formatApiCorrections(phantoms) {
+    if (phantoms.length === 0)
+        return '';
+    const lines = phantoms.map(p => `  - ${p.suggestion}`);
+    return `API CORRECTIONS\n${lines.join('\n')}`;
+}

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@mjasnikovs/pi-task",
-    "version": "0.14.6",
+    "version": "0.14.8",
     "description": "Deterministic spec-orchestration for local models, with a bundled real-time remote web view and web/docs/fetch/worker subagent tools.",
     "type": "module",
     "main": "./dist/index.js",