nubos-pilot 0.5.2 → 0.5.4

package/workflows/add-todo.md

@@ -41,11 +41,16 @@ TODO_PATH="${PENDING_DIR}/${DATE}-${SLUG}.md"
 ```
 
 Extract from init JSON: `commit_docs`, `date`, `timestamp`, `slug`,
-`todo_count`, `todos_dir_exists`, `pending_dir`, `state_path`. The
-init handler sanitises the slug through `lib/layout.cjs.slugify`
-(strips to `[a-z0-9-]` only; filename-injection mitigation) and
-validates the description length (<= 500 chars) before any filesystem
-write occurs.
+`todo_count`, `todos_dir_exists`, `pending_dir`, `state_path`,
+`text_mode`, `text_mode_source`. The init handler sanitises the slug
+through `lib/layout.cjs.slugify` (strips to `[a-z0-9-]` only;
+filename-injection mitigation) and validates the description length
+(<= 500 chars) before any filesystem write occurs.
+
+**Text-mode routing.** If `text_mode == true`, skip every `np-tools.cjs askuser`
+call below and render questions as plain-text numbered lists in the main
+chat. Auto-enabled in Claude Code (CLAUDECODE=1); opt-in via
+`.nubos-pilot/config.json` → `workflow.text_mode`.
 
 ## Create Pending Dir
 

package/workflows/discuss-phase.md

@@ -19,13 +19,37 @@ workflow delivers PLAN-01 and nothing beyond it.
 ## Initialize
 
 ```bash
+LANG_DIRECTIVE=$(node .nubos-pilot/bin/np-tools.cjs lang-directive)
 INIT=$(node .nubos-pilot/bin/np-tools.cjs init discuss-phase "$PHASE")
 if [[ "$INIT" == @file:* ]]; then INIT=$(cat "${INIT#@file:}"); fi
 ```
 
+**Language (SSOT = `.nubos-pilot/config.json` → `response_language`).**
+`$LANG_DIRECTIVE` is authoritative for this workflow. Obey it for ALL
+subsequent output, askuser prompt texts, status updates, and the CONTEXT.md
+rendering. This supersedes any directive in CLAUDE.md managed block if they
+conflict — the config is the single source of truth.
+
 Parse JSON for: `milestone`, `milestone_id`, `milestone_dir`, `milestone_name`,
 `milestone_context_path`, `has_context`, `has_milestone_dir`, `goal`,
-`requirements`, `agent_skills`, `mode`.
+`requirements`, `agent_skills`, `mode`, `text_mode`, `text_mode_source`.
+
+**Text-mode routing (SSOT = INIT payload).** If `text_mode` is `true`, do
+**not** shell out to `np-tools.cjs askuser` for any prompt in this workflow.
+Present every question inline as a plain-text numbered list and wait for the
+user's reply in the main chat. This is the correct path whenever:
+
+- `text_mode_source == "runtime"` → Claude Code Bash has no TTY and cannot
+  forward interactive menu selections; the askuser marker-block protocol
+  never completes.
+- `text_mode_source == "config"` → the user explicitly opted into text mode
+  via `.nubos-pilot/config.json` → `workflow.text_mode`.
+
+When text mode is active, skip every `node .nubos-pilot/bin/np-tools.cjs
+askuser …` block below and substitute the plain-text equivalent. Collect the
+answers from the user's reply, then proceed to the next step as normal. The
+rest of the workflow (validation, canonical-ref accumulation, template
+render, commit) is unchanged.
 
 If the user passed `--assumptions`, route to
 `workflows/discuss-phase-assumptions.md` and exit this workflow.
@@ -118,25 +142,33 @@ Capture the idea in a "Deferred Ideas" section. Don't lose it, don't act on it.
 ## Answer Validation
 
 <answer_validation>
-**IMPORTANT: Answer validation** — After every interactive prompt, check if the
-response is empty or whitespace-only. If so:
-1. Retry the question once with the same parameters
-2. If still empty, present the options as a plain-text numbered list and ask
-   the user to type their choice number
+**Routing decision (made once at Initialize):** If INIT payload
+`text_mode == true`, skip every `np-tools.cjs askuser` call in this workflow
+and use plain-text numbered lists in the main chat instead. The
+`text_mode_source` field (`runtime` / `config` / `default`) tells you why
+text mode is active — it is informational only and does not change behavior.
+
+**When text_mode is false and askuser is used — per-prompt validation:**
+1. If `askuser` exits with structured error `askuser-no-tty` (exit code 1,
+   stderr JSON with `"code":"askuser-no-tty"`), that means the runtime
+   detection missed something; **skip retry** and treat the remainder of the
+   workflow as text-mode (plain-text numbered lists).
+2. If the response is empty or whitespace-only (exit 0 but no value), retry
+   the question once with the same parameters.
+3. If still empty, present the options as a plain-text numbered list and ask
+   the user to type their choice number.
 Never proceed with an empty answer.
 
-**Text mode (`workflow.text_mode: true` in config or `--text` flag):**
-When text mode is active, **do not use `np-tools.cjs askuser` at all**.
-Instead, present every question as a plain-text numbered list and ask the
-user to type their choice number. This is required for Claude Code remote
-sessions (`/rc` mode) where the Claude App cannot forward TUI menu selections
-back to the host.
-
-Enable text mode:
-- Per-session: pass `--text` flag
-- Per-project: `np-tools.cjs config-set workflow.text_mode true`
+**Enable text mode:**
+- Auto-detected: any Claude Code session (`CLAUDECODE=1` /
+  `CLAUDE_CODE_ENTRYPOINT` set) — default behavior, no user action needed.
+- Opt-in per project: set `workflow.text_mode: true` in
+  `.nubos-pilot/config.json`.
+- Opt-out per project: set `workflow.text_mode: false` in
+  `.nubos-pilot/config.json` (overrides runtime detection).
 
-Text mode applies to ALL workflows in the session, not just discuss-phase.
+Text mode applies to ALL workflows that emit `text_mode` in their INIT
+payload, not just discuss-phase.
 </answer_validation>
 
 ## Process

package/workflows/discuss-project.md

@@ -51,10 +51,16 @@ Never:
 ## Single-Call Init
 
 ```bash
+LANG_DIRECTIVE=$(node .nubos-pilot/bin/np-tools.cjs lang-directive)
 INIT=$(node .nubos-pilot/bin/np-tools.cjs init discuss-project ${BOOTSTRAP:+--bootstrap})
 if [[ "$INIT" == @file:* ]]; then INIT=$(cat "${INIT#@file:}"); fi
 ```
 
+**Language (SSOT = `.nubos-pilot/config.json` → `response_language`).**
+`$LANG_DIRECTIVE` is authoritative. Obey it for all askuser prompt texts,
+narrative status updates, and the prose written into PROJECT.md sections.
+Supersedes CLAUDE.md managed block.
+
 Parse: `mode`, `sub_mode` (`bootstrap` or `refresh`), `project_md_exists`,
 `scan_context`, `questions[]`, `required_fields[]`.
 

package/workflows/execute-phase.md

@@ -16,13 +16,27 @@ Execute every slice of a milestone in wave order: slice S001 first (all its task
 
 ```bash
 PHASE="$1"
-INIT=$(node .nubos-pilot/bin/np-tools.cjs init execute-milestone "$PHASE")
+LANG_DIRECTIVE=$(node .nubos-pilot/bin/np-tools.cjs lang-directive)
+INIT=$(node .nubos-pilot/bin/np-tools.cjs init execute-milestone init "$PHASE")
 if [[ "$INIT" == @file:* ]]; then INIT=$(cat "${INIT#@file:}"); fi
 AGENT_SKILLS_EXECUTOR=$(node .nubos-pilot/bin/np-tools.cjs agent-skills executor 2>/dev/null)
 RUNTIME=$(node -e "console.log(require('./lib/runtime/index.cjs').detect().runtime)")
 ```
 
-Parse JSON for: `milestone`, `milestone_id`, `milestone_dir`, `waves[]` (each with `wave` (= slice number), `slice_id`, `slice_full_id`, `slice_dir`, `tasks[]`), `total_tasks`, `slice_count`, `executor_tier`, `agent_skills`.
+**Language (SSOT = `.nubos-pilot/config.json` → `response_language`).**
+`$LANG_DIRECTIVE` is authoritative for this workflow. Obey it for all user-
+facing output, askuser prompts, and status updates. Pass `$LANG_DIRECTIVE`
+into every np-executor spawn prompt as a system-level rule so task summaries
+and checkpoint notes follow the project language. This supersedes any
+directive in CLAUDE.md managed block.
+
+Parse JSON for: `milestone`, `milestone_id`, `milestone_dir`, `waves[]` (each with `wave` (= slice number), `slice_id`, `slice_full_id`, `slice_dir`, `tasks[]`), `total_tasks`, `slice_count`, `executor_tier`, `text_mode`, `text_mode_source`, `agent_skills`.
+
+**Text-mode routing.** If `text_mode == true`, skip every `np-tools.cjs askuser`
+call below (including the orphan-checkpoint and empty-milestone prompts)
+and render the options as a plain-text numbered list in the main chat.
+Auto-enabled in Claude Code (CLAUDECODE=1); opt-in via
+`.nubos-pilot/config.json` → `workflow.text_mode`.
 
 `PLAN_ID` is iterated per slice as `${milestone_id}-${slice_id}` (e.g. `M001-S001`). `TASK_ID` is iterated from each slice's `tasks[]` (e.g. `M001-S001-T0001`).
 

package/workflows/new-milestone.md

@@ -52,10 +52,22 @@ The subcommand raises `project-not-initialized` anyway, but the shell check give
 ## Single-Call Init
 
 ```bash
+LANG_DIRECTIVE=$(node .nubos-pilot/bin/np-tools.cjs lang-directive)
 INIT=$(node .nubos-pilot/bin/np-tools.cjs init new-milestone)
 if [[ "$INIT" == @file:* ]]; then INIT=$(cat "${INIT#@file:}"); fi
 ```
 
+**Language (SSOT = `.nubos-pilot/config.json` → `response_language`).**
+`$LANG_DIRECTIVE` is authoritative. Obey it for askuser prompt texts,
+user-facing output, and any prose written into milestone artefacts (YAML
+keys, IDs, and identifiers stay canonical English). Supersedes CLAUDE.md.
+
+**Text-mode routing.** If INIT payload `text_mode == true`, skip every
+`np-tools.cjs askuser` call below and render each question as a plain-text
+prompt in the main chat; collect the answer inline. Auto-enabled in Claude
+Code (CLAUDECODE=1); opt-in via `.nubos-pilot/config.json` →
+`workflow.text_mode`.
+
 Payload: three questions — `milestone_name`, `milestone_goal`, `create_req_prefix` (confirm).
 
 ## Interview

package/workflows/new-project.md

@@ -106,10 +106,23 @@ Use the scan to propose:
 The 5 structural questions. All prompts go through the askuser gateway.
 
 ```bash
+LANG_DIRECTIVE=$(node .nubos-pilot/bin/np-tools.cjs lang-directive)
 INIT=$(node .nubos-pilot/bin/np-tools.cjs init new-project)
 if [[ "$INIT" == @file:* ]]; then INIT=$(cat "${INIT#@file:}"); fi
 ```
 
+**Language (SSOT = `.nubos-pilot/config.json` → `response_language`).**
+`$LANG_DIRECTIVE` is authoritative. Obey it for all askuser prompt texts,
+user-facing output, and any narrative prose written into PROJECT.md /
+REQUIREMENTS.md (field names and YAML keys stay canonical English).
+Supersedes CLAUDE.md.
+
+**Text-mode routing.** If INIT payload `text_mode == true`, skip every
+`np-tools.cjs askuser` call below and render each question as a plain-text
+prompt in the main chat; collect the user's answer inline and move on.
+Auto-enabled in Claude Code (CLAUDECODE=1); opt-in via
+`.nubos-pilot/config.json` → `workflow.text_mode`.
+
 ```bash
 ANS_PROJECT_NAME=$(node .nubos-pilot/bin/np-tools.cjs askuser --json '{"type":"input","prompt":"Project name?"}')
 ANS_CORE_VALUE=$(node .nubos-pilot/bin/np-tools.cjs askuser --json '{"type":"input","prompt":"Core value — one sentence that must stay true if everything else fails?"}')

package/workflows/note.md

@@ -52,6 +52,17 @@ into `$TEXT`. Empty text after stripping is an error — there is no
 `list` or `promote` subcommand here (deferred to a future
 capture-management plan).
 
+**Text-mode routing.** Resolve once at the start:
+
+```bash
+TEXT_MODE=$(node .nubos-pilot/bin/np-tools.cjs text-mode 2>/dev/null || echo false)
+```
+
+If `$TEXT_MODE == "true"`, skip every `np-tools.cjs askuser` call below and
+render questions as plain-text numbered lists in the main chat. Auto-enabled
+in Claude Code (CLAUDECODE=1); opt-in via `.nubos-pilot/config.json` →
+`workflow.text_mode`.
+
 ## Compute Paths
 
 Scope branching is explicit (T-10-05-02 mitigation + Pitfall 10

package/workflows/pause-work.md

@@ -15,9 +15,14 @@ progress.
 ## Execution
 
 ```bash
+LANG_DIRECTIVE=$(node .nubos-pilot/bin/np-tools.cjs lang-directive)
 node .nubos-pilot/bin/np-tools.cjs init pause-work
 ```
 
+**Language (SSOT = `.nubos-pilot/config.json` → `response_language`).**
+Obey `$LANG_DIRECTIVE` for the resume-hint narration and any status lines
+printed around the JSON payload. Supersedes CLAUDE.md.
+
 Output is a small JSON payload `{ ok, stopped_at, resume_file }`. The
 workflow simply displays it.
 

package/workflows/plan-phase.md

@@ -72,6 +72,7 @@ fi
 ### Read milestone state
 
 ```bash
+LANG_DIRECTIVE=$(node .nubos-pilot/bin/np-tools.cjs lang-directive)
 INIT=$(node .nubos-pilot/bin/np-tools.cjs init plan-milestone init "$PHASE")
 if [[ "$INIT" == @file:* ]]; then INIT=$(cat "${INIT#@file:}"); fi
 AGENT_SKILLS_PLANNER=$(node .nubos-pilot/bin/np-tools.cjs agent-skills planner 2>/dev/null)
@@ -79,7 +80,18 @@ AGENT_SKILLS_CHECKER=$(node .nubos-pilot/bin/np-tools.cjs agent-skills plan-chec
 RUNTIME=$(node -e "console.log(require('./lib/runtime/index.cjs').detect().runtime)")
 ```
 
-Parse JSON for: `milestone`, `milestone_id`, `milestone_dir`, `milestone_context_path`, `milestone_roadmap_path`, `milestone_meta_path`, `name`, `goal`, `requirements`, `success_criteria`, `has_context`, `has_roadmap`, `has_meta`, `existing_slices[]`, `planner_tier`, `checker_tier`, `agent_skills`.
+**Language (SSOT = `.nubos-pilot/config.json` → `response_language`).**
+`$LANG_DIRECTIVE` is authoritative. Obey it for all user-facing output,
+askuser prompts, status updates, and any narrative text the spawned planner
+or plan-checker subagents emit. Pass `$LANG_DIRECTIVE` into their spawn
+prompts as a system-level rule. This supersedes any directive in CLAUDE.md.
+
+Parse JSON for: `milestone`, `milestone_id`, `milestone_dir`, `milestone_context_path`, `milestone_roadmap_path`, `milestone_meta_path`, `name`, `goal`, `requirements`, `success_criteria`, `has_context`, `has_roadmap`, `has_meta`, `existing_slices[]`, `planner_tier`, `checker_tier`, `text_mode`, `text_mode_source`, `agent_skills`.
+
+**Text-mode routing.** If `text_mode == true`, skip every `np-tools.cjs askuser`
+call below and present questions as plain-text numbered lists in the main
+chat. Auto-enabled in Claude Code (CLAUDECODE=1); opt-in per-project via
+`.nubos-pilot/config.json` → `workflow.text_mode`.
 
 `PLAN_ID` and `TASK_ID` default to `${milestone_id}-plan` / `${milestone_id}-planner-run` for the metrics records.
 

package/workflows/research-phase.md

@@ -92,14 +92,26 @@ payload; larger payloads are written to a tmp file and referenced via
 `@file:<path>`.
 
 ```bash
+LANG_DIRECTIVE=$(node .nubos-pilot/bin/np-tools.cjs lang-directive)
 INIT=$(node .nubos-pilot/bin/np-tools.cjs init research-phase "$PHASE")
 if [[ "$INIT" == @file:* ]]; then INIT=$(cat "${INIT#@file:}"); fi
 RUNTIME=$(node -e "console.log(require('./lib/runtime/index.cjs').detect().runtime)")
 ```
 
+**Language (SSOT = `.nubos-pilot/config.json` → `response_language`).**
+`$LANG_DIRECTIVE` is authoritative. Obey it for user-facing output and
+askuser prompts, and pass it into the np-researcher spawn prompt so
+RESEARCH.md prose (not URLs, citations, or code snippets) follows the
+project language. This supersedes CLAUDE.md.
+
 `RUNTIME` is resolved once here and reused by the metrics-record call at the
 researcher spawn site (Step 4) per D-06 workflow-writer pattern.
 
+**Text-mode routing.** If `text_mode == true` in the payload below, skip every
+`np-tools.cjs askuser` call in this workflow and render questions as
+plain-text numbered lists in the main chat. Auto-enabled in Claude Code
+(CLAUDECODE=1); opt-in via `.nubos-pilot/config.json` → `workflow.text_mode`.
+
 The payload shape:
 
 ```json

package/workflows/resume-work.md

@@ -13,10 +13,22 @@ on each accordingly.
 ## Initialize
 
 ```bash
+LANG_DIRECTIVE=$(node .nubos-pilot/bin/np-tools.cjs lang-directive)
 INIT=$(node .nubos-pilot/bin/np-tools.cjs init resume-work)
 STATUS=$(echo "$INIT" | node -e "process.stdin.on('data', d => console.log(JSON.parse(d).status))")
 ```
 
+**Language (SSOT = `.nubos-pilot/config.json` → `response_language`).**
+`$LANG_DIRECTIVE` is authoritative. Obey it for user-facing output and
+askuser prompts. When spawning the np-executor to continue a checkpoint,
+pass `$LANG_DIRECTIVE` into the spawn prompt so resumed task summaries
+follow the project language. Supersedes CLAUDE.md.
+
+**Text-mode routing.** If INIT payload `text_mode == true`, skip every
+`np-tools.cjs askuser` call below and render prompts as plain-text numbered
+lists in the main chat. Auto-enabled in Claude Code (CLAUDECODE=1); opt-in
+via `.nubos-pilot/config.json` → `workflow.text_mode`.
+
 ## Execution
 
 ### status: resume

package/workflows/scan-codebase.md

@@ -60,10 +60,18 @@ end.
 Scan, group, write manifest + stubs, emit module-facts in one call:
 
 ```bash
+LANG_DIRECTIVE=$(node .nubos-pilot/bin/np-tools.cjs lang-directive)
 PLAN=$(node .nubos-pilot/bin/np-tools.cjs scan-codebase --project-name "$PROJECT_NAME")
 if [[ "$PLAN" == @file:* ]]; then PLAN=$(cat "${PLAN#@file:}"); fi
 ```
 
+**Language (SSOT = `.nubos-pilot/config.json` → `response_language`).**
+`$LANG_DIRECTIVE` is authoritative. Obey it for user-facing output and
+askuser prompts. Pass it into any writer-subagent spawn so module-doc
+prose (overview, responsibilities, notes) follows the project language.
+Module IDs, file paths, symbol names, and language labels stay canonical
+English. Supersedes CLAUDE.md.
+
 `--project-name` is optional; when provided it goes into `INDEX.md`. Other
 flags: `--batch-size N` (default 500), `--max-files N`.
 

package/workflows/session-report.md

@@ -32,6 +32,7 @@ record. Pitfall 9 / `workflow-missing-metrics` is exempt.
 ## Initialize
 
 ```bash
+LANG_DIRECTIVE=$(node .nubos-pilot/bin/np-tools.cjs lang-directive)
 SINCE_OVERRIDE=""
 for arg in "$@"; do
   case "$arg" in
@@ -49,6 +50,23 @@ LOCAL_FILENAME_TS=$(date +"%Y-%m-%dT%H%M")
 REPORT_PATH="${REPORTS_DIR}/${LOCAL_FILENAME_TS}-session-report.md"
 ```
 
+**Language (SSOT = `.nubos-pilot/config.json` → `response_language`).**
+`$LANG_DIRECTIVE` is authoritative. Obey it for the report's narrative
+sections (summary, highlights, notable events) and any askuser prompts.
+Task IDs, milestone IDs, commit SHAs, metrics keys, and file paths stay
+canonical English. Supersedes CLAUDE.md.
+
+**Text-mode routing.** Resolve once:
+
+```bash
+TEXT_MODE=$(node .nubos-pilot/bin/np-tools.cjs text-mode 2>/dev/null || echo false)
+```
+
+If `$TEXT_MODE == "true"`, skip every `np-tools.cjs askuser` call below and
+render questions as plain-text numbered lists in the main chat. Auto-enabled
+in Claude Code (CLAUDECODE=1); opt-in via `.nubos-pilot/config.json` →
+`workflow.text_mode`.
+
 The filename format is `YYYY-MM-DDTHHMM-session-report.md` (D-17 —
 4-char HHMM, no seconds, local time) so reports sort
 lexicographically by invocation order.

package/workflows/update-docs.md

@@ -51,10 +51,17 @@ If `.doc-index.json` is missing, fall back to a full rescan via
 ## Single-Call Init
 
 ```bash
+LANG_DIRECTIVE=$(node .nubos-pilot/bin/np-tools.cjs lang-directive)
 PLAN=$(node .nubos-pilot/bin/np-tools.cjs update-docs)
 if [[ "$PLAN" == @file:* ]]; then PLAN=$(cat "${PLAN#@file:}"); fi
 ```
 
+**Language (SSOT = `.nubos-pilot/config.json` → `response_language`).**
+`$LANG_DIRECTIVE` is authoritative. Obey it for user-facing output and
+askuser prompts. Pass it into any writer-subagent spawn so refreshed
+module-doc prose follows the project language. Module IDs, file paths,
+and symbol names stay canonical English. Supersedes CLAUDE.md.
+
 Parse: `mode`, `diff_summary` (added/removed/changed/unchanged counts),
 `stale_modules[]`, `added_modules[]`, `removed_modules[]`.
 

package/workflows/validate-phase.md

@@ -19,12 +19,24 @@ if [[ -z "$PHASE" ]]; then
   exit 2
 fi
 
+LANG_DIRECTIVE=$(node .nubos-pilot/bin/np-tools.cjs lang-directive)
 INIT=$(node .nubos-pilot/bin/np-tools.cjs init verify-work "$PHASE")
 if [[ "$INIT" == @file:* ]]; then INIT=$(cat "${INIT#@file:}"); fi
 RUNTIME=$(node -e "console.log(require('./lib/runtime/index.cjs').detect().runtime)")
 ```
 
-Parse JSON for: `milestone`, `milestone_id`, `milestone_dir`, `milestone_name`, `slice_uat`.
+**Language (SSOT = `.nubos-pilot/config.json` → `response_language`).**
+`$LANG_DIRECTIVE` is authoritative. Obey it for user-facing output, askuser
+prompts, and pass it into the np-nyquist-auditor spawn prompt so gap-fill
+narrative follows the project language. Test IDs, file paths, and canonical
+field names stay English. Supersedes CLAUDE.md.
+
+Parse JSON for: `milestone`, `milestone_id`, `milestone_dir`, `milestone_name`, `slice_uat`, `text_mode`, `text_mode_source`.
+
+**Text-mode routing.** If `text_mode == true`, skip every `np-tools.cjs askuser`
+call in this workflow and render options as plain-text numbered lists in
+the main chat. Auto-enabled in Claude Code (CLAUDECODE=1); opt-in via
+`.nubos-pilot/config.json` → `workflow.text_mode`.
 
 ```bash
 MILESTONE_ID=$(echo "$INIT" | jq -r '.milestone_id')

package/workflows/verify-work.md

@@ -16,12 +16,26 @@ Slice-level acceptance (UAT) is validated separately by `/np:validate-phase <N>`
 
 ```bash
 PHASE="$1"
+LANG_DIRECTIVE=$(node .nubos-pilot/bin/np-tools.cjs lang-directive)
 INIT=$(node .nubos-pilot/bin/np-tools.cjs init verify-work "$PHASE")
 if [[ "$INIT" == @file:* ]]; then INIT=$(cat "${INIT#@file:}"); fi
 AGENT_SKILLS_VERIFIER=$(node .nubos-pilot/bin/np-tools.cjs agent-skills verifier 2>/dev/null)
 ```
 
-Parse: `milestone`, `milestone_id`, `milestone_dir`, `milestone_name`, `success_criteria`, `draft_results`, `verification_path`, `slice_uat`, `verifier_tier`, `agent_skills`.
+**Language (SSOT = `.nubos-pilot/config.json` → `response_language`).**
+`$LANG_DIRECTIVE` is authoritative. Obey it for user-facing output, askuser
+prompts, and pass it into the np-verifier spawn prompt so VERIFICATION.md
+prose (Pass/Fail findings, root-cause notes) follows the project language.
+Test-case IDs, file paths, and stack traces stay canonical. Supersedes
+CLAUDE.md.
+
+Parse: `milestone`, `milestone_id`, `milestone_dir`, `milestone_name`, `success_criteria`, `draft_results`, `verification_path`, `slice_uat`, `verifier_tier`, `text_mode`, `text_mode_source`, `agent_skills`.
+
+**Text-mode routing.** If `text_mode == true`, skip every `np-tools.cjs askuser`
+call below (including the Pass-2 `needs_user_confirm` gate) and render the
+options as a plain-text numbered list in the main chat. Auto-enabled in
+Claude Code (CLAUDECODE=1); opt-in via `.nubos-pilot/config.json` →
+`workflow.text_mode`.
 
 ## Pass 1 — verifier agent