nazgul-config

name	nazgul:config
description	View and change Nazgul settings — model assignments, formatter, notifications. Use when user says "configure nazgul", "change models", "nazgul settings", or wants to adjust config after init.
argument-hint	[models]
allowed-tools	Read, Write, Edit, Bash, Glob, Grep, ToolSearch
metadata	{"author":"Jose Mejia","version":"2.7.1"}

Nazgul Config

Examples

• /nazgul:config — View current settings and change any of them
• /nazgul:config models — Jump directly to model assignment configuration

Arguments

$ARGUMENTS

Pre-flight

0. Load the AskUserQuestion tool (deferred by default): run ToolSearch with query select:AskUserQuestion. Do this BEFORE any step that uses AskUserQuestion.
1. Check if nazgul/config.json exists. If not: "Nazgul not initialized. Run /nazgul:init first." and STOP.

Step 1: Display Current Settings

Read nazgul/config.json and display:

─── ◈ NAZGUL ▸ CONFIGURATION ───────────────────────────

Models:
  Planning:        [models.planning]
  Discovery:       [models.discovery]
  Docs:            [models.docs]
  Review:          [models.review]
  Implementation:  [models.implementation]
  Specialists:     [models.specialists]
  Post-loop:       [models.post_loop]
  Default:         [models.default]

Formatter:         [formatter.enabled ? "enabled" : "disabled"]
Notifications:     [notifications.on_complete || "disabled"]
Default mode:      [default_mode || "ask each run"]

Review:
  Granularity:     [review_gate.granularity || "task"]
  Require all approve: [review_gate.require_all_approve]
  Confidence threshold: [review_gate.confidence_threshold]
  Max retries/unit:  [review_gate.max_retries_per_task]
  Block on security reject: [review_gate.block_on_security_reject]

Guards:
  Lean comments:     [guards.lean_comments != false ? "on" : "off"] (max run: [guards.max_consecutive_comment_lines || 2])

Step 2: Ask What to Change

If the ## Arguments block above contains the token models, skip directly to the Model Assignment sub-flow.

Otherwise, use AskUserQuestion (multiSelect: true):

• header: "Settings"
• question: "What would you like to change?"
• options:
  • "Model assignments" — "Configure which AI model runs each pipeline stage"
  • "Formatter" — "Enable or disable auto-formatting after edits"
  • "Notifications" — "Configure completion notifications"
  • "Default run mode" — "Set the mode /nazgul:start uses when no flag is passed (or clear it to be asked each time)"
  • "Review granularity" — "Choose whether the review board fires per task, per parallel group, or once per feature"
  • "Lean comments guard" — "Toggle the deterministic comment-bloat guard, or tune the max consecutive comment-line run"

Step 3: Run Selected Sub-flows

Lean Comments Guard Sub-flow

1. AskUserQuestion:
   • header: "Lean comments"
   • question: "Comment-bloat guard (blocks banners, restating comment runs, and <remarks>/multi-paragraph docs on non-public members at write time)?"
   • options:
     • "On (default)" — "Block comment bloat in source writes and reject it in review"
     • "Off" — "Opt out — the guard becomes a no-op for this project"
2. Write the choice: jq '.guards.lean_comments = (<true|false>)' nazgul/config.json > nazgul/config.json.tmp && mv nazgul/config.json.tmp nazgul/config.json (substitute true/false; do NOT run the placeholder verbatim).
3. If the user wants to tune the threshold, write the integer to guards.max_consecutive_comment_lines the same way (default 2 = runs of 3+ are blocked).

Model Assignment Sub-flow

1. AskUserQuestion:

   • header: "Models"
   • question: "How would you like to configure models?"
   • options:
     • "Use a preset" — "Quick setup with predefined model assignments"
     • "Customize per stage" — "Pick a model for each pipeline stage individually"

2. If preset, use AskUserQuestion:

   • header: "Preset"
   • question: "Which model preset?"
   • options:
     • "Balanced (Recommended)" — "Opus for planning, Sonnet for implementation/review, Haiku for post-loop"
     • "Quality" — "Opus for everything — best results, highest cost"
     • "Fast/cheap" — "Haiku for docs/review/post-loop, Sonnet for planning/implementation"

   Preset values:

   • Balanced: { planning: "opus", discovery: "sonnet", docs: "sonnet", review: "sonnet", implementation: "sonnet", specialists: "sonnet", post_loop: "haiku", default: "sonnet" }
   • Quality: { planning: "opus", discovery: "opus", docs: "opus", review: "opus", implementation: "opus", specialists: "opus", post_loop: "opus", default: "opus" }
   • Fast/cheap: { planning: "sonnet", discovery: "haiku", docs: "haiku", review: "haiku", implementation: "sonnet", specialists: "sonnet", post_loop: "haiku", default: "haiku" }

3. If per-stage, use AskUserQuestion in two batches:

   Batch 1:

   • "Which model for Planning?" (header: "Planning") → Opus (Recommended) / Sonnet / Haiku
   • "Which model for Discovery?" (header: "Discovery") → Opus / Sonnet (Recommended) / Haiku
   • "Which model for Docs?" (header: "Docs") → Opus / Sonnet (Recommended) / Haiku
   • "Which model for Review?" (header: "Review") → Opus / Sonnet (Recommended) / Haiku

   Batch 2:

   • "Which model for Implementation?" (header: "Implement") → Opus / Sonnet (Recommended) / Haiku
   • "Which model for Specialists?" (header: "Specialist") → Opus / Sonnet (Recommended) / Haiku
   • "Which model for Post-loop?" (header: "Post-loop") → Opus / Sonnet / Haiku (Recommended)

4. Update nazgul/config.json → models with the selected values using jq:

   jq '.models.planning = "opus" | .models.discovery = "sonnet" | ...' nazgul/config.json > nazgul/config.json.tmp && mv nazgul/config.json.tmp nazgul/config.json
   

5. Update model: field in generated agents by role:

   • Read nazgul/config.json → agents.reviewers and agents.specialists arrays
   • For each file in .claude/agents/generated/*.md, extract the agent name from frontmatter
   • If name appears in agents.reviewers → set model: [models.review]
   • If name appears in agents.specialists → set model: [models.specialists]

6. Display confirmation:

   ─── ◈ NAZGUL ▸ MODELS UPDATED ──────────────────────────
   
   Planning:        opus
   Discovery:       sonnet
   ...
   
   Changes take effect on the next pipeline run.
   

Formatter Sub-flow

Same as init Step 7 Question 1. Use AskUserQuestion:

• header: "Formatter"
• question: "Auto-format files after edits?"
• options:
  • "Yes" — "Run prettier/ruff/gofmt/etc. automatically based on file type"
  • "No" — "Skip auto-formatting, handle it manually"
• Update nazgul/config.json → formatter.enabled

Notifications Sub-flow

Use AskUserQuestion:

• header: "Notify"
• question: "Notify when the loop completes?"
• options:
  • "Voice alert (Recommended)" — platform default: say 'Nazgul loop complete' (macOS) or notify-send 'Nazgul' 'Loop complete' (Linux)
  • "Silent" — "No notification when the loop finishes"
• If Voice alert: set nazgul/config.json → notifications.on_complete to the platform-appropriate command
• If Silent: remove or empty nazgul/config.json → notifications.on_complete
• If Other (user selects the built-in "Other" free-text option): set nazgul/config.json → notifications.on_complete to the user's custom command

Default Run Mode Sub-flow

1. AskUserQuestion:

   • header: "Default mode"
   • question: "What mode should /nazgul:start use when you don't pass a flag?"
   • options:
     • "Ask each run (Recommended)" — "No default; /nazgul:start prompts for mode each time"
     • "HITL" — "Review each step"
     • "AFK" — "Autonomous; pauses on risky decisions"
     • "YOLO" — "Fully autonomous; no permission prompts (start still confirms YOLO each run)"

2. Write the choice to config: "Ask each run" → null, otherwise the lowercase mode string:

   jq '.default_mode=<value>' nazgul/config.json > nazgul/config.json.tmp && mv nazgul/config.json.tmp nazgul/config.json
   

   (use null unquoted for "Ask each run", e.g. jq '.default_mode=null' …).

Review Granularity Sub-flow

1. AskUserQuestion:

   • header: "Review unit"
   • question: "When should the review board run?"
   • options:
     • "Per task (Recommended)" — "Review each task the moment it reaches IMPLEMENTED (current default)"
     • "Per group" — "Review once per planner-defined parallel wave/group, over that group's combined diff"
     • "Per feature" — "Implement ALL tasks to IMPLEMENTED, then ONE review over the whole feature diff (base..HEAD)"

2. Write the selected lowercase value to review_gate.granularity (task | group | feature) — substitute the user's choice; do NOT run the placeholder verbatim:

   jq --arg g "<task|group|feature>" '.review_gate.granularity = $g' nazgul/config.json > nazgul/config.json.tmp && mv nazgul/config.json.tmp nazgul/config.json
   

3. Note for the user: group/feature defer review until the unit is fully implemented. All other review settings — require_all_approve, confidence_threshold, block_on_security_reject — still apply in every mode; max_retries_per_task is counted per review unit (task/group/feature), and in group/feature mode a CHANGES_REQUESTED re-opens only the tasks whose files own the findings, not the whole unit. Takes effect on the next pipeline run.