nubos-pilot 0.9.2 → 0.9.3

package/agents/np-build-fixer.md

@@ -90,7 +90,7 @@ node .nubos-pilot/bin/np-tools.cjs handoff-write \
 - **Success:** verify command exits 0; no extra files written; control returned to executor.
 - **Stuck after 3 attempts:** write `T<NNNN>-FIX-NOTES.md` next to the task plan; emit `## FIX FAILED` block listing attempts + suspected cause.
 - **Out-of-scope failure:** emit `## SCOPE EXPANSION REQUEST` block listing the out-of-scope path + the symbol involved; do NOT edit.
-- **Infra failure:** emit `## INFRA BLOCKER` block listing the missing dependency; do NOT edit.
+- **Infrastructure mismatch (container down, wrong runtime version, missing service):** this is NOT a fix-target. Emit a finding tagged `information-missing` with the specific mismatch (e.g., `composer requires php ^8.5, container runs 8.4`) so `loop-evaluate` routes to the researcher swarm or plan-checker, not back to you. Do NOT edit Dockerfiles, compose configs, or other infra paths to "make verify green" — that's outside any task's `files_modified`.
 
 <scope_guardrail>
 **Do:**
@@ -98,6 +98,7 @@ node .nubos-pilot/bin/np-tools.cjs handoff-write \
 - Run the task's verify command via Bash.
 - Use `knowledge-search` for unfamiliar symbols.
 - Stop after 3 failed attempts and document.
+- Distinguish code failures (your job) from infrastructure failures (route via finding).
 
 **Don't:**
 - Expand `files_modified` — that's the planner's job; emit a SCOPE EXPANSION REQUEST instead.
@@ -106,4 +107,6 @@ node .nubos-pilot/bin/np-tools.cjs handoff-write \
 - Silence failures with empty catches, skipped tests, or commented-out assertions.
 - Re-litigate locked decisions in `M<NNN>-CONTEXT.md` or `RULES.md`.
 - Spawn other agents.
+- Edit infrastructure (Dockerfile, docker-compose, k8s, CI configs) to fix verify-red — those paths are out of scope for any task; surface the mismatch as an `information-missing` finding instead.
+- Treat container-down / runtime-version-skew as a code bug. It's an environment routing signal, not a code-fixable failure.
 </scope_guardrail>

package/agents/np-critic-acceptance.md

@@ -48,6 +48,7 @@ The orchestrator provides these paths in your prompt context. Read every path it
 2. **Locked-decision conformance** — the diff does not violate any locked decision in `M<NNN>-CONTEXT.md`. Violations are findings of category `locked-decision-violation`.
 3. **Scope creep** — the diff does not edit files outside `files_modified`. Out-of-scope edits are findings of category `scope-creep`.
 4. **Stuck-marker check** — if the task is on round 3 with no progress between rounds, you flag `stuck-detected` so the orchestrator escalates.
+5. **Infrastructure-mismatch detection** — if the verify output indicates an infrastructure failure (container exited, runtime version skew, missing service: `php -v` mismatch, `docker exec` errors, port-not-bound, DB-unreachable), do NOT downgrade affected criteria to `Unsatisfied` or `Satisfied`. Mark them `Information-Missing` with a finding of category `information-missing` whose `remediation` names the specific environment delta (e.g., `composer requires php ^8.5, container runs 8.4 — Dockerfile bump required outside this milestone`). The orchestrator routes that to researcher / plan-checker, not back to executor — the code is not at fault.
 
 ## Output Schema
 

package/agents/np-executor.md

@@ -117,6 +117,7 @@ into the `task(…)` commit. If `workflow.commit_docs=true`, the
 - Commit via `node np-tools.cjs commit-task <task-id>`.
 - Write checkpoint state transitions via the wrapper.
 - Stay within the task's declared scope even if you spot tangential issues — log them, do not fix them.
+- Run the task's `<verify>` command and capture its exit code + output. If it fails because the runtime environment is wrong (container exited, wrong PHP/Node version, missing service), surface that in the verify output verbatim — the Nubosloop's `loop-run-round --phase post-executor` reads the exit code and routes accordingly. The infra issue is a routing signal, not your decision.
 
 **Don't:**
 - Add files to the commit beyond `files_modified` (D-04 authoritative).
@@ -124,6 +125,8 @@ into the `task(…)` commit. If `workflow.commit_docs=true`, the
 - Bypass the checkpoint wrapper.
 - Use `--no-verify`, `--force`, `git reset --hard`, `git clean`, `git restore .`, or any destructive git flag.
 - Auto-discover files via `git status` — the plan declares scope, not the filesystem.
+- **Pre-validate the runtime environment** (`docker ps`, `php -v`, `node -v`, container-status checks, DB connectivity probes). The orchestrator's pre-flight phase covers what needs to be checked; you do code edits and run verify. If the container is down or the runtime is wrong, the verify command will fail and the loop routes that — never declare a "hard blocker" or abort the spawn over environment state.
+- **Refuse to spawn / halt before editing because of infra mismatch** (PHP version skew, missing image, etc.). Tasks edit code, not infrastructure. Run your edits, run verify, let the result speak.
 </scope_guardrail>
 
 ## Handoff Protocol

package/bin/np-tools/config.cjs

@@ -1,6 +1,7 @@
 const fs = require('node:fs');
 const path = require('node:path');
 const { findProjectRoot, NubosPilotError } = require('../../lib/core.cjs');
+const { DEFAULT_CONFIG_TREE } = require('../../lib/config-defaults.cjs');
 
 const SEGMENT_RE = /^[a-zA-Z0-9_-]+$/;
 const BLOCKED_SEGMENTS = new Set(['__proto__', 'constructor', 'prototype']);
@@ -75,11 +76,10 @@ function run(argv, ctx) {
   try {
     _validateSegments(segments);
     const config = _readConfig(cwd);
-    if (config == null) {
-      if (!raw) stdout.write('\n');
-      return 0;
+    let value = config == null ? undefined : _walkPath(config, segments);
+    if (value === undefined) {
+      value = _walkPath(DEFAULT_CONFIG_TREE, segments);
     }
-    const value = _walkPath(config, segments);
     if (value === undefined) {
       if (!raw) stdout.write('\n');
       return 0;

package/bin/np-tools/config.test.cjs

@@ -69,3 +69,64 @@ test('CONFIG-4: object value serialized as JSON', () => {
   assert.equal(code, 0);
   assert.equal(stdout.toString(), '{"k":"v"}');
 });
+
+test('CONFIG-5: returns DEFAULT_CONFIG_TREE value when key absent from user config', () => {
+  const sb = makeSandbox({ runtime: 'claude' });
+  const stdout = makeSink();
+  const code = configCli.run(['loop.maxRounds'], { cwd: sb, stdout, stderr: makeSink() });
+  assert.equal(code, 0);
+  assert.equal(stdout.toString(), '3\n');
+});
+
+test('CONFIG-6: defaults walk into nested swarm.research.* keys', () => {
+  const sb = makeSandbox({});
+  const out1 = makeSink(); configCli.run(['swarm.research.k'], { cwd: sb, stdout: out1, stderr: makeSink() });
+  const out2 = makeSink(); configCli.run(['swarm.research.threshold'], { cwd: sb, stdout: out2, stderr: makeSink() });
+  const out3 = makeSink(); configCli.run(['swarm.research.minOccurrence'], { cwd: sb, stdout: out3, stderr: makeSink() });
+  assert.equal(out1.toString(), '3\n');
+  assert.equal(out2.toString(), '0.9\n');
+  assert.equal(out3.toString(), '3\n');
+});
+
+test('CONFIG-7: user-set value wins over default', () => {
+  const sb = makeSandbox({ loop: { maxRounds: 5 } });
+  const stdout = makeSink();
+  configCli.run(['loop.maxRounds'], { cwd: sb, stdout, stderr: makeSink() });
+  assert.equal(stdout.toString(), '5\n');
+});
+
+test('CONFIG-8: partial user override falls through to defaults for sibling keys', () => {
+  const sb = makeSandbox({ swarm: { research: { k: 7 } } });
+  const k = makeSink(); configCli.run(['swarm.research.k'], { cwd: sb, stdout: k, stderr: makeSink() });
+  const t = makeSink(); configCli.run(['swarm.research.threshold'], { cwd: sb, stdout: t, stderr: makeSink() });
+  assert.equal(k.toString(), '7\n');
+  assert.equal(t.toString(), '0.9\n');
+});
+
+test('CONFIG-9: unknown key without a default still returns empty', () => {
+  const sb = makeSandbox({});
+  const stdout = makeSink();
+  configCli.run(['really.not.a.thing'], { cwd: sb, stdout, stderr: makeSink() });
+  assert.equal(stdout.toString(), '\n');
+});
+
+test('CONFIG-10: defaults resolve even without config.json present', () => {
+  const sb = makeSandbox(); // no config.json
+  const stdout = makeSink();
+  configCli.run(['loop.maxRounds'], { cwd: sb, stdout, stderr: makeSink() });
+  assert.equal(stdout.toString(), '3\n');
+});
+
+test('CONFIG-11: explicit user false wins over default true (boolean handling)', () => {
+  const sb = makeSandbox({ auto_log_learning: false });
+  const stdout = makeSink();
+  configCli.run(['auto_log_learning'], { cwd: sb, stdout, stderr: makeSink() });
+  assert.equal(stdout.toString(), 'false\n');
+});
+
+test('CONFIG-12: --raw mode resolves defaults without trailing newline', () => {
+  const sb = makeSandbox({});
+  const stdout = makeSink();
+  configCli.run(['loop.maxRounds', '--raw'], { cwd: sb, stdout, stderr: makeSink() });
+  assert.equal(stdout.toString(), '3');
+});

package/lib/config-defaults.cjs

@@ -47,6 +47,17 @@ const DEFAULT_MODEL_PROFILE = 'frontier';
 const DEFAULT_SCOPE = 'local';
 const DEFAULT_RESPONSE_LANGUAGE = 'en';
 
+const DEFAULT_CONFIG_TREE = Object.freeze({
+  scope: DEFAULT_SCOPE,
+  model_profile: DEFAULT_MODEL_PROFILE,
+  response_language: DEFAULT_RESPONSE_LANGUAGE,
+  workflow: DEFAULT_WORKFLOW,
+  agents: DEFAULT_AGENTS,
+  loop: DEFAULT_LOOP,
+  swarm: DEFAULT_SWARM,
+  auto_log_learning: DEFAULT_AUTO_LOG_LEARNING,
+});
+
 function buildInstallConfig(answers) {
   const a = answers || {};
   return {
@@ -80,5 +91,6 @@ module.exports = {
   DEFAULT_MODEL_PROFILE,
   DEFAULT_SCOPE,
   DEFAULT_RESPONSE_LANGUAGE,
+  DEFAULT_CONFIG_TREE,
   buildInstallConfig,
 };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "nubos-pilot",
-  "version": "0.9.2",
+  "version": "0.9.3",
   "description": "AI-driven planning and execution tool for code projects",
   "homepage": "https://github.com/Nubos-AI/nubos-pilot",
   "repository": {

package/workflows/execute-phase.md

@@ -363,6 +363,8 @@ After every slice completes, point the operator at `/np:validate-phase $PHASE` t
 - Bundle two tasks into one commit (ADR-0004 atomicity).
 - Skip the checkpoint start step — it's the crash-safety primitive `resume-work` depends on.
 - Pass `--no-verify` or `--force` anywhere in the pipeline.
+- **Introduce ad-hoc pre-flight checks beyond the two sanctioned guards** (orphan-checkpoint, empty-milestone). Container-status (`docker ps`), runtime-version probes (`php -v`, `node -v`), DB-connectivity, port-binding — none of these belong in the orchestrator's pre-flight. Tasks edit code; environment failures surface inside the Nubosloop as `verify-red` (→ `spawn-build-fixer`) or as `np-critic-acceptance` `information-missing` findings (→ researcher / plan-checker). They are **never** workflow-level halts.
+- **Declare a "hard blocker" because of infrastructure state.** Container down, PHP version skew, missing image, exited service — all of these are routing signals inside the loop, not reasons to abort the wave. The wave only halts on `commit-task` non-zero, `stuck` after `loop.maxRounds`, or `plan-checker` (locked-decision-violation). Infrastructure mismatch routes via critic findings to researcher/plan-checker; if it's truly out-of-scope for any task in the milestone, the operator handles it separately and re-runs the workflow.
 <!-- /scope_guardrail -->
 
 ## Output