mixdog 0.9.63 → 0.9.65

package/scripts/internal-comms-smoke.mjs

@@ -1,277 +0,0 @@
-#!/usr/bin/env node
-// Static smoke for internal-comms token-optimization rules (min chars / max
-// info). Asserts the Lead brief contract and the agent handoff contract are
-// present and injected, without any model call. Live token A/B is a separate
-// bench (scripts/internal-comms-bench.mjs).
-import { createRequire } from 'node:module';
-import { mkdirSync, mkdtempSync, readFileSync, rmSync, writeFileSync } from 'node:fs';
-import { tmpdir } from 'node:os';
-import { dirname, join, resolve } from 'node:path';
-import { fileURLToPath } from 'node:url';
-
-const root = resolve(dirname(fileURLToPath(import.meta.url)), '..');
-const require = createRequire(import.meta.url);
-const rulesBuilder = require('../src/lib/rules-builder.cjs');
-
-function assert(condition, message) {
-  if (!condition) throw new Error(message);
-}
-function readSrc(...parts) {
-  return readFileSync(join(root, 'src', ...parts), 'utf8');
-}
-function normalize(text) {
-  return String(text || '').replace(/\s+/g, ' ').trim();
-}
-function rawBlock(text, start, end, label) {
-  const source = String(text);
-  const from = source.indexOf(start);
-  if (from < 0) throw new Error(`${label}: missing start marker ${start}`);
-  const finish = source.indexOf(end, from + start.length);
-  if (finish < 0) throw new Error(`${label}: missing end marker ${end}`);
-  return source.slice(from, finish);
-}
-function block(text, start, end, label) {
-  return normalize(rawBlock(text, start, end, label));
-}
-function bodyAfterFrontmatter(text, label) {
-  const source = String(text);
-  const opening = source.match(/^---[ \t]*\r?\n/);
-  if (!opening) throw new Error(`${label}: missing frontmatter opening fence`);
-  const closing = source.slice(opening[0].length).match(/^---[ \t]*(?:\r?\n|$)/m);
-  if (!closing) throw new Error(`${label}: missing frontmatter closing fence`);
-  return source.slice(opening[0].length + closing.index + closing[0].length);
-}
-function roleBody(text, label) {
-  return normalize(bodyAfterFrontmatter(text, label));
-}
-function frontmatter(text, label) {
-  const source = String(text);
-  const opening = source.match(/^---[ \t]*\r?\n/);
-  if (!opening) throw new Error(`${label}: missing frontmatter opening fence`);
-  const closing = source.slice(opening[0].length).match(/^---[ \t]*(?:\r?\n|$)/m);
-  if (!closing) throw new Error(`${label}: missing frontmatter closing fence`);
-  return source.slice(opening[0].length, opening[0].length + closing.index);
-}
-
-// --- Compact rule contracts -------------------------------------------------
-const workflow = readSrc('workflows', 'default', 'WORKFLOW.md');
-const leadBrief = readSrc('rules', 'lead', 'lead-brief.md');
-const solo = readSrc('workflows', 'solo', 'WORKFLOW.md');
-const general = readSrc('rules', 'lead', '01-general.md');
-const leadTool = readSrc('rules', 'lead', 'lead-tool.md');
-const core = readSrc('rules', 'agent', '00-core.md');
-const common = readSrc('rules', 'agent', '00-common.md');
-const skip = readSrc('rules', 'agent', '20-skip-protocol.md');
-const OPTIONAL_BRIEF_FIELDS = ['Anchors:', 'Allow/Forbid:', 'Deliver:'];
-const TOKEN_PRINCIPLE = /minimum (?:characters|chars), maximum (?:information|info)/i;
-function requireAll(text, label, patterns) {
-  for (const pattern of patterns) assert(pattern.test(normalize(text).toLowerCase()), `${label}: missing ${pattern}`);
-}
-assert(TOKEN_PRINCIPLE.test(normalize(leadBrief)), 'lead-brief.md: brief must state min-char/max-info principle');
-assert(leadBrief.includes('Task:'), 'lead-brief.md: brief must require labeled field Task:');
-assert(!leadBrief.includes('Goal:'), 'lead-brief.md: Goal: must be replaced by Task:');
-for (const field of OPTIONAL_BRIEF_FIELDS) assert(leadBrief.includes(field), `lead-brief.md: brief missing optional labeled field ${field}`);
-assert(!leadBrief.includes('Verify:'), 'lead-brief.md: must not prescribe Verify:');
-assert(
-  !/copy a prior role(?:'s)? brief|worker\s*(?:->|→)\s*debugger\s*(?:->|→)\s*reviewer|\blead\s*\/\s*worker\b/i.test(leadBrief),
-  'lead-brief.md: must not restore cross-role Task copying or lineage',
-);
-requireAll(leadBrief, 'Lead brief task', [
-  /`task:` is mandatory and lossless/, /intent/, /required outcomes/, /negative outcomes/,
-  /completion\/stop boundary/, /all other fields are optional task-specific deltas/,
-  /preserve user-supplied exact targets and exact replacements\/outputs in `task:`/,
-  /never infer exactness from a task name, file count, or perceived difficulty/,
-  /each role independently constructs a role-appropriate, lossless `task:` from/,
-  /the original request and official spec\/test acceptance criteria/,
-]);
-assert(/role-known|already (?:owns|knows)|wasted cost|wasted/i.test(normalize(leadBrief)), 'lead-brief.md: brief must ban restating known rules/background as cost');
-assert(/original request and official spec\/test acceptance criteria/i.test(normalize(leadBrief)), 'lead-brief.md: original request and official acceptance criteria must beat summary');
-requireAll(leadBrief, 'Lead brief lifecycle', [
-  /full brief only for fresh spawn\/`respawned: true`/, /live follow-ups are delta/,
-  /dead-tag send is cold: re-supply anchors/, /never `send` mid-run/,
-  /batch one follow-up after completion/, /interrupt only to cancel/,
-  /agent communication is english/,
-]);
-assert(/lead brief contract/i.test(normalize(workflow)), 'WORKFLOW.md: must defer to the lead brief contract');
-assert(!OPTIONAL_BRIEF_FIELDS.every((field) => workflow.includes(field)), 'WORKFLOW.md: must not duplicate the optional brief field list');
-
-assert(/fragments/i.test(core), '00-core: handoff must require fragments');
-assert(/file:line/i.test(core), '00-core: handoff must anchor evidence to file:line');
-for (const banned of ['headings', 'tables', 'narration', 'raw logs', 'next-checks']) {
-  assert(core.toLowerCase().includes(banned), `00-core: banned list missing ${banned}`);
-}
-assert(/Public Agent Constraints/i.test(common), '00-common: must be titled public-only constraints');
-assert(/git operations deferred to Lead/i.test(normalize(common)), '00-common: must refuse git/Ship');
-assert(/Overflow goes to a file/i.test(common), '00-common: must keep overflow-to-file rule');
-requireAll(common, 'Public-agent shell', [
-  /shell only verifies own edits/, /no exploration, install, or state change beyond brief/,
-]);
-
-// --- Per-role output contracts --------------------------------------------
-const roles = {
-  'worker/AGENT.md': readSrc('agents', 'worker', 'AGENT.md'),
-  'heavy-worker/AGENT.md': readSrc('agents', 'heavy-worker', 'AGENT.md'),
-  'reviewer/AGENT.md': readSrc('agents', 'reviewer', 'AGENT.md'),
-  'debugger/AGENT.md': readSrc('agents', 'debugger', 'AGENT.md'),
-};
-const reviewerRole = roles['reviewer/AGENT.md'];
-assert(
-  !/construct the review\s*`task:` independently/i.test(reviewerRole),
-  'reviewer/AGENT.md: must not restore Reviewer-specific Task construction',
-);
-assert(
-  !/treat prior reports as unverified claims rather than evidence/i.test(reviewerRole),
-  'reviewer/AGENT.md: must not restore prior-report framing',
-);
-
-// Semantic contracts deliberately avoid prose snapshots.
-function snapshot(actual, expected, label) {
-  assert(normalize(actual) === normalize(expected), `${label}: canonical snapshot changed`);
-}
-assert(/^agents:\s*$/m.test(frontmatter(solo, 'Solo workflow')), 'Solo: agents frontmatter must be empty');
-requireAll(workflow, 'Default approval', [
-  /read-only investigation\/planning while consulting/, /later explicit user message after the latest plan/,
-  /initial\/additional\/changed requests reset planning/, /scope change needs a revised plan and fresh approval/,
-  /no edits, state mutation, or delegation/,
-]);
-const WORKER_HEAVY_ROUTING_CONTRACT = [
-  /select worker or heavy worker by solution and validation complexity/,
-  /worker owns work whose approach is established and whose correctness can be shown with a focused check/,
-  /heavy worker owns work requiring investigation to establish the approach, resolution of material uncertainty or dependencies, iteration across hypotheses or checkpoints, or multi-step, empirical, integration, or real-use validation/,
-];
-requireAll(workflow, 'Default Worker/Heavy Worker routing', WORKER_HEAVY_ROUTING_CONTRACT);
-assert(
-  !/fully\s+specified\s+artifact/i.test(workflow),
-  'Default: removed Worker routing criterion must not return',
-);
-const ROUTING_REVIEW_POLICY = [
-  /lead-direct execution is limited to a conceptually one-step mechanical action with an immediate deterministic check/,
-  /routine configuration\/git work/,
-  /operational support for planning, delegation, or verification/,
-  /multiple steps must not be reframed as one command to fit this boundary/,
-  /work requiring implementation judgment, investigation, iteration, or artifact authoring delegates to worker or heavy worker/,
-  /debugger owns diagnosis and reverse engineering/,
-  /applying a debugger result is implementation, not diagnosis/,
-  /review is exempt only for pure read\/analysis with no edit, artifact, or state mutation; git\/configuration; or a request where the user explicitly supplies both the exact target and exact replacement\/output\. every non-exempt mutation or artifact/,
-  /every non-exempt mutation or artifact/,
-  /worker, heavy worker, debugger, or lead/,
-  /one reviewer .*lead integration\/cross-scope verification in parallel/,
-  /debugger analysis cannot substitute for implementation review/,
-  /applying a debugger result triggers the same reviewer \+ lead verification/,
-  /loop fix -> re-verify \(same reviewer \+ lead re-check\) until clean/,
-  /exempt mutations still require shell self-verification/,
-];
-requireAll(workflow, 'Default routing/review policy', ROUTING_REVIEW_POLICY);
-assert(
-  !/(high clarity|low structural complexity|immediate 1-step|genuinely simple)/i.test(workflow),
-  'Default: heuristic routing/review language must not return',
-);
-requireAll(workflow, 'Default lifecycle', [
-  /draft before any implementation/, /all ready scopes in one turn, with no count cap/,
-  /real dependency, overlapping write, or inseparable coupling/,
-  /after async spawn, end the turn/,
-  /high-risk scopes add distinct lenses/,
-  /original live session/,
-  /bug surviving 2\+ fix cycles/, /agent reports relay.*as in-progress, never conclusions/,
-  /final \(not interim\) report/, /never forward raw agent output/,
-  /explicit user request after issue-free feedback/, /outcome\/direction change, pause and re-consult/,
-]);
-requireAll(solo, 'Solo lifecycle', [
-  /no edits or state mutation/, /never spawn, send, delegate, or ask agents to work/,
-  /read-only investigation\/planning while consulting/,
-  /later explicit user message after the latest plan/,
-  /initial\/additional\/changed requests reset planning/,
-  /scope change needs a revised plan and fresh approval/,
-  /checks\/fixes directly until clean or reports a blocker/, /final \(not interim\) report/,
-  /interim updates are in-progress, never conclusions/,
-  /issue-free feedback/,
-]);
-requireAll(leadTool, 'Lead tools', [
-  /write-role agents self-verify/, /cross-scope verification.*benches.*all git/,
-]);
-requireAll(general, 'General safety', [
-  /identify as mixdog\/current coding agent/, /destructive\/hard-to-reverse action needs explicit confirmation/,
-]);
-requireAll(workflow, 'Default ship safety', [
-  /build\/deploy\/commit\/push require an explicit user request after issue-free feedback/,
-  /implementation approval alone is insufficient/,
-]);
-requireAll(skip, 'Silent skip', [
-  /webhook-handler/, /scheduler-task/, /label-only, duplicate\/dedup, no action needed\/report/,
-  /whole response.*\[meta:silent\]/,
-]);
-
-const roleSnapshots = {
-  'worker/AGENT.md': `# Worker
-Scoped implementation agent.
-
-Own only the bounded responsibility assigned in the brief. Trust its
-\`file:line\` anchors; do only minimal targeted discovery, then make the
-smallest coherent patch. No drive-by cleanup or scope expansion.
-
-EDIT-FIRST DISCIPLINE. Patch promptly rather than repeating read-only turns;
-stop and report blocked when the assigned scope cannot be completed.
-
-Self-verify with a targeted check (for example, \`node --check\` or a focused
-test), then report the changed \`file:line\` and stop.`,
-  'heavy-worker/AGENT.md': `# Heavy Worker
-Own the assigned implementation slice through staged delivery.
-
-Break work into bounded, dependency-aware slices and execute them in sequence.
-At each checkpoint, run the narrowest relevant test or build before expanding
-the slice. Keep the smallest coherent change; control blast radius rather than
-rewriting adjacent systems.
-
-EDIT-FIRST DISCIPLINE. Patch incrementally and stop at the first explicit
-boundary: unclear ownership, a missing dependency, or growing blast radius.
-Do not cross that boundary without a new bounded assignment; report blocked
-work with the relevant file:line.
-
-Self-verify each checkpoint and the final slice with shell (targeted test/build).`,
-  'reviewer/AGENT.md': `# Reviewer
-Independent regression/risk review agent.
-
-Review the diff and tests with independent judgment. Prioritize actionable
-correctness, regression, security, and verification risks; inspect affected
-boundaries. Do not reimplement the change or report non-risky nits. Independently
-evaluate the final deliverable with a critical lens, actively seeking errors,
-unsupported assumptions, and counterexamples before confirming.
-
-Report findings first, severity-ordered, with one line per \`file:line\`. If clean,
-say so in one line and include only material residual risk.
-
-When the work comes with stated criteria or reference material for judging
-it, verify against those as given — substituting your own interpretation or
-a self-built check is a verification risk to report.`,
-  'debugger/AGENT.md': `# Debugger
-Root-cause analysis agent.
-
-Smallest confirmed cause chain before fixes. Return likely cause, evidence
-(\`file:line\`), smallest next check/fix. Mark confirmed facts vs inferences;
-avoid broad speculation.
-
-Converge, don't sweep: when new evidence stops accruing, report the best
-cause chain so far.`,
-};
-for (const [name, expected] of Object.entries(roleSnapshots)) {
-  const text = roles[name];
-  const permission = text.match(/^permission:\s*(.+)$/m)?.[1];
-  assert(permission === (name === 'worker/AGENT.md' || name === 'heavy-worker/AGENT.md' ? 'read-write' : 'read'), `${name}: permission contract changed`);
-  snapshot(roleBody(text, name), expected, name);
-}
-assert(/confirmed facts vs inferences/i.test(normalize(roles['debugger/AGENT.md'])), 'debugger: must separate confirmed facts from inferences');
-assert(/file:line/i.test(normalize(roles['debugger/AGENT.md'])), 'debugger: must anchor evidence to file:line');
-
-// --- Injection: Lead rules actually carry the brief contract ---------------
-const dataDir = mkdtempSync(join(tmpdir(), 'mixdog-internal-comms-smoke-'));
-try {
-  const leadRules = rulesBuilder.buildInjectionContent({ PLUGIN_ROOT: join(root, 'src'), DATA_DIR: dataDir });
-  assert(TOKEN_PRINCIPLE.test(normalize(leadRules)), 'injected Lead rules must carry the brief token principle');
-  assert(leadRules.includes('Task:'), 'injected Lead rules missing mandatory brief field Task:');
-  assert(!leadRules.includes('Goal:'), 'injected Lead rules must not retain Goal:');
-  for (const field of OPTIONAL_BRIEF_FIELDS) assert(leadRules.includes(field), `injected Lead rules missing optional brief field ${field}`);
-} finally {
-  rmSync(dataDir, { recursive: true, force: true });
-}
-
-process.stdout.write('internal comms smoke passed\n');

package/scripts/internal-tools-normalization-test.mjs

@@ -1,62 +0,0 @@
-#!/usr/bin/env node
-import test from 'node:test';
-import assert from 'node:assert/strict';
-import {
-    executeInternalTool,
-    setInternalToolsProvider,
-} from '../src/runtime/agent/orchestrator/internal-tools.mjs';
-import { classifyResultKind } from '../src/runtime/agent/orchestrator/session/result-classification.mjs';
-
-async function normalizeToolResult(result) {
-    setInternalToolsProvider({
-        tools: [{ name: 'normalization_test' }],
-        executor: async () => result,
-    });
-    return executeInternalTool('normalization_test', {});
-}
-
-test('already-prefixed handler errors keep one canonical Error: prefix', async () => {
-    const result = await normalizeToolResult({
-        content: [{ type: 'text', text: 'Error: web search failed' }],
-        isError: true,
-    });
-    assert.equal(result, 'Error: web search failed');
-    assert.equal(classifyResultKind(result), 'error');
-});
-
-test('ordinary handler error text receives the canonical Error: prefix', async () => {
-    const result = await normalizeToolResult({
-        content: [{ type: 'text', text: 'core add: project_id required' }],
-        isError: true,
-    });
-    assert.equal(result, 'Error: core add: project_id required');
-    assert.equal(classifyResultKind(result), 'error');
-});
-
-test('successful empty output stays normal', async () => {
-    const result = await normalizeToolResult({
-        content: [{ type: 'text', text: '' }],
-        isError: false,
-    });
-    assert.equal(result, '');
-    assert.equal(classifyResultKind(result), 'normal');
-});
-
-test('isError:false output stays normal', async () => {
-    const result = await normalizeToolResult({
-        content: [{ type: 'text', text: 'search complete' }],
-        isError: false,
-    });
-    assert.equal(result, 'search complete');
-    assert.equal(classifyResultKind(result), 'normal');
-});
-
-test('structured image output survives internal tool normalization', async () => {
-    const input = {
-        content: [
-            { type: 'text', text: 'downloaded' },
-            { type: 'image', source: { type: 'base64', media_type: 'image/png', data: 'YWJj' } },
-        ],
-    };
-    assert.deepEqual(await normalizeToolResult(input), input);
-});