@nforma.ai/nforma 0.2.1 → 0.28.0

package/commands/{qgsd → nf}/quorum.md

@@ -1,6 +1,6 @@
 ---
-name: qgsd:quorum
-description: Answer a question using full quorum consensus (Claude + native CLI agents + all configured claude-mcp-server instances) following QGSD quorum protocol. Use when no arguments provided to answer the current conversation's open question.
+name: nf:quorum
+description: Answer a question using full quorum consensus (Claude + native CLI agents + all configured claude-mcp-server instances) following nForma quorum protocol. Use when no arguments provided to answer the current conversation's open question.
 argument-hint: "[question or prompt]"
 allowed-tools:
   - Read
@@ -12,7 +12,7 @@ allowed-tools:
 ---
 
 <objective>
-Run a question or prompt through the full QGSD quorum (R3 protocol): Claude + native CLI agents (Codex, Gemini, OpenCode, Copilot) + all claude-mcp-server instances configured in `~/.claude.json`.
+Run a question or prompt through the full nForma quorum (R3 protocol): Claude + native CLI agents (Codex, Gemini, OpenCode, Copilot) + all claude-mcp-server instances configured in `~/.claude.json`.
 
 **Two modes** based on context:
 - **Mode A — Pure Question**: No commands required. Claude forms its own position first, then dispatches all slot-workers as parallel Tasks, deliberates to consensus.
@@ -31,7 +31,7 @@ Resolve the question to pass:
    - **Priority 1** — Most recent message containing `?` without a substantive answer yet.
    - **Priority 2** — Most recent message describing a choice/trade-off (keywords: "should we", "which approach", "option A vs", "do we", "whether to").
    - **Priority 3** — Most recent open concern or blocker ("not sure", "concern", "blocker", "unclear", "wondering").
-   - If none found: stop with `"No open question found. Provide one explicitly: /qgsd:quorum <question>"`
+   - If none found: stop with `"No open question found. Provide one explicitly: /nf:quorum <question>"`
 
 When question is inferred, display before dispatching:
 ```
@@ -62,7 +62,7 @@ If $ARGUMENTS is empty: use the most recent open question or decision from the c
 Before any model calls, run a fast HTTP probe of the underlying LLM providers:
 
 ```bash
-node "$HOME/.claude/qgsd-bin/check-provider-health.cjs" --json
+node "$HOME/.claude/nf-bin/check-provider-health.cjs" --json
 ```
 
 Parse the JSON output. Build two structures:
@@ -73,12 +73,12 @@ Parse the JSON output. Build two structures:
 
 Any server with `available: false` must be marked UNAVAIL immediately — skip its health_check and inference calls entirely. This prevents hangs from unresponsive provider endpoints.
 
-3. **`$QUORUM_ACTIVE`**: read from `~/.claude/qgsd.json` (project config takes precedence over global):
+3. **`$QUORUM_ACTIVE`**: read from `~/.claude/nf.json` (project config takes precedence over global):
 ```bash
 node -e "
 const fs = require('fs'), os = require('os'), path = require('path');
-const globalCfg = path.join(os.homedir(), '.claude', 'qgsd.json');
-const projCfg   = path.join(process.cwd(), '.claude', 'qgsd.json');
+const globalCfg = path.join(os.homedir(), '.claude', 'nf.json');
+const projCfg   = path.join(process.cwd(), '.claude', 'nf.json');
 let cfg = {};
 for (const f of [globalCfg, projCfg]) {
   try { Object.assign(cfg, JSON.parse(fs.readFileSync(f, 'utf8'))); } catch(_){}
@@ -96,12 +96,12 @@ A server in `$QUORUM_ACTIVE` but absent from `$CLAUDE_MCP_SERVERS` = skip silent
 - Reorder the remaining working list: healthy servers first (preserving discovery order within each group).
 - Log the final working list as: `Active slots: <slot1>, <slot2>, ...`
 
-**max_quorum_size check:** Read `max_quorum_size` from `~/.claude/qgsd.json` (project config takes precedence; default: 3 if absent):
+**max_quorum_size check:** Read `max_quorum_size` from `~/.claude/nf.json` (project config takes precedence; default: 3 if absent):
 ```bash
 node -e "
 const fs = require('fs'), os = require('os'), path = require('path');
-const globalCfg = path.join(os.homedir(), '.claude', 'qgsd.json');
-const projCfg   = path.join(process.cwd(), '.claude', 'qgsd.json');
+const globalCfg = path.join(os.homedir(), '.claude', 'nf.json');
+const projCfg   = path.join(process.cwd(), '.claude', 'nf.json');
 let cfg = {};
 for (const f of [globalCfg, projCfg]) {
   try { Object.assign(cfg, JSON.parse(fs.readFileSync(f, 'utf8'))); } catch(_){}
@@ -136,7 +136,7 @@ node -e "
 const fs = require('fs'), path = require('path'), os = require('os');
 
 const searchPaths = [
-  path.join(os.homedir(), '.claude', 'qgsd-bin', 'providers.json'),
+  path.join(os.homedir(), '.claude', 'nf-bin', 'providers.json'),
 ];
 try {
   const cj = JSON.parse(fs.readFileSync(path.join(os.homedir(), '.claude.json'), 'utf8'));
@@ -150,8 +150,8 @@ for (const p of searchPaths) {
   try { providers = JSON.parse(fs.readFileSync(p, 'utf8')).providers; break; } catch(_) {}
 }
 
-const globalCfg = path.join(os.homedir(), '.claude', 'qgsd.json');
-const projCfg   = path.join(process.cwd(), '.claude', 'qgsd.json');
+const globalCfg = path.join(os.homedir(), '.claude', 'nf.json');
+const projCfg   = path.join(process.cwd(), '.claude', 'nf.json');
 let cfg = {};
 for (const f of [globalCfg, projCfg]) {
   try { Object.assign(cfg, JSON.parse(fs.readFileSync(f, 'utf8'))); } catch(_) {}
@@ -178,7 +178,7 @@ Detect Claude's model ID from: `CLAUDE_MODEL` env var → `ANTHROPIC_MODEL` env
 
 Run:
 ```bash
-node "$HOME/.claude/qgsd-bin/update-scoreboard.cjs" init-team \
+node "$HOME/.claude/nf-bin/update-scoreboard.cjs" init-team \
   --claude-model "<claude_model_id>" \
   --team '<TEAM_JSON>'
 ```
@@ -236,7 +236,7 @@ echo "Adaptive fan-out: risk_level=${RISK_LEVEL} → fan_out_count=${FAN_OUT_COU
 
 **Apply cap — build DISPATCH_LIST:** Take the first `FAN_OUT_COUNT - 1` slots from the active working list (healthy-first order from pre-flight). This is the definitive slot cap. Call this `$DISPATCH_LIST`. All subsequent round dispatches (Round 1 and deliberation) use `$DISPATCH_LIST` — never the full working list.
 
-> **Why here and not only in the hook:** `qgsd-prompt.js` enforces this cap via `--n` for main-session prompts. But quorum is also dispatched inline from subagents (e.g., `qgsd-planner` in step 8.5 of `plan-phase.md`) where no UserPromptSubmit hook fires. `quorum.md` must self-enforce the cap for all dispatch contexts.
+> **Why here and not only in the hook:** `nf-prompt.js` enforces this cap via `--n` for main-session prompts. But quorum is also dispatched inline from subagents (e.g., `nf-planner` in step 8.5 of `plan-phase.md`) where no UserPromptSubmit hook fires. `quorum.md` must self-enforce the cap for all dispatch contexts.
 
 ### R6.4 Reduced-Quorum Note (FAN-05)
 
@@ -250,7 +250,7 @@ fi
 # When FAN_OUT_COUNT = MAX_QUORUM_SIZE (high/absent): no note emitted.
 ```
 
-The RISK_LEVEL variable is available downstream for use by Phase v0.18-04 Adaptive Fan-Out. The fan-out logic is now implemented in both quorum.md (above) and qgsd-prompt.js (which emits `--n N` for downstream ceiling verification).
+The RISK_LEVEL variable is available downstream for use by Phase v0.18-04 Adaptive Fan-Out. The fan-out logic is now implemented in both quorum.md (above) and nf-prompt.js (which emits `--n N` for downstream ceiling verification).
 
 ---
 
@@ -267,7 +267,7 @@ Priority 2 - Pending decision: Find the most recent message that describes a cho
 Priority 3 - Open concern or blocker: Find the most recent message that raises a concern, flags a risk, or states something is unclear (keywords: "not sure", "concern", "blocker", "question:", "unclear", "wondering"). Restate it as a question.
 
 If none of the above applies: stop with:
-"No open question found. Looked for: explicit '?' question, pending decision, or open concern in recent conversation. Provide a question explicitly: /qgsd:quorum <question>"
+"No open question found. Looked for: explicit '?' question, pending decision, or open concern in recent conversation. Provide a question explicitly: /nf:quorum <question>"
 
 When a question is inferred via any priority, Claude MUST display before proceeding:
 "Using conversation context as question (Priority N - [type]):
@@ -276,7 +276,7 @@ When a question is inferred via any priority, Claude MUST display before proceed
 Display:
 ```
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
- QGSD ► QUORUM: Round 1 — N workers dispatched
+ nForma ► QUORUM: Round 1 — N workers dispatched
  Active: gemini-1, opencode-1, copilot-1, codex-1
  Fallback pool: T1 = unused slots with auth_type=sub; T2 = slots with auth_type≠sub (on UNAVAIL)
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
@@ -296,7 +296,7 @@ Store as `$CLAUDE_POSITION`.
 
 ### Query models (parallel — one Task per slot)
 
-Dispatch one `Task(subagent_type="qgsd-quorum-slot-worker", model="haiku", max_turns=100, ...)` per slot in **`$DISPATCH_LIST`** (capped to `FAN_OUT_COUNT - 1` external slots) as **parallel sibling calls** in one message turn. Do NOT dispatch slots outside `$DISPATCH_LIST`. Build a YAML prompt block per the slot-worker argument spec:
+Dispatch one `Task(subagent_type="nf-quorum-slot-worker", model="haiku", max_turns=100, ...)` per slot in **`$DISPATCH_LIST`** (capped to `FAN_OUT_COUNT - 1` external slots) as **parallel sibling calls** in one message turn. Do NOT dispatch slots outside `$DISPATCH_LIST`. Build a YAML prompt block per the slot-worker argument spec:
 
 ```
 slot: <slotName>
@@ -318,11 +318,11 @@ Set it to the evaluation criteria appropriate for the artifact type, e.g.:
 - Audit:    "This is a milestone audit. Evaluate whether the work achieves the stated milestone goals."
 
 Example dispatch (all Tasks in one message turn):
-- `Task(subagent_type="qgsd-quorum-slot-worker", model="haiku", max_turns=100, description="gemini-1 [gemini-cli · gemini-3-pro-preview] quorum R1", prompt=<YAML block>)`
-- `Task(subagent_type="qgsd-quorum-slot-worker", model="haiku", max_turns=100, description="codex-1 [codex-cli · gpt-5.3-codex] quorum R1", prompt=<YAML block>)`
-- `Task(subagent_type="qgsd-quorum-slot-worker", model="haiku", max_turns=100, description="opencode-1 [opencode-cli · grok-code-fast-1] quorum R1", prompt=<YAML block>)`
-- `Task(subagent_type="qgsd-quorum-slot-worker", model="haiku", max_turns=100, description="copilot-1 [copilot-cli · gpt-4.1] quorum R1", prompt=<YAML block>)`
-- `Task(subagent_type="qgsd-quorum-slot-worker", model="haiku", max_turns=100, description="claude-1 [claude-code-router · deepseek-ai/DeepSeek-V3.2] quorum R1", prompt=<YAML block>)` ← one per claude-mcp server with `available: true`
+- `Task(subagent_type="nf-quorum-slot-worker", model="haiku", max_turns=100, description="gemini-1 [gemini-cli · gemini-3-pro-preview] quorum R1", prompt=<YAML block>)`
+- `Task(subagent_type="nf-quorum-slot-worker", model="haiku", max_turns=100, description="codex-1 [codex-cli · gpt-5.3-codex] quorum R1", prompt=<YAML block>)`
+- `Task(subagent_type="nf-quorum-slot-worker", model="haiku", max_turns=100, description="opencode-1 [opencode-cli · grok-code-fast-1] quorum R1", prompt=<YAML block>)`
+- `Task(subagent_type="nf-quorum-slot-worker", model="haiku", max_turns=100, description="copilot-1 [copilot-cli · gpt-4.1] quorum R1", prompt=<YAML block>)`
+- `Task(subagent_type="nf-quorum-slot-worker", model="haiku", max_turns=100, description="claude-1 [claude-code-router · deepseek-ai/DeepSeek-V3.2] quorum R1", prompt=<YAML block>)` ← one per claude-mcp server with `available: true`
 (model="haiku" — slot-workers are orchestrators (read files, build prompt, run Bash subprocess), NOT reasoners. The actual reasoning is done by the external CLI. Haiku is faster with zero quality loss.)
 
 The slot-worker reads repo context, builds its own prompt from the YAML arguments, calls the slot via `call-quorum-slot.cjs`, and returns a structured result block.
@@ -369,7 +369,7 @@ If all available models agree → skip to **Consensus output**.
 
 Run up to 9 deliberation rounds (max 10 total rounds including Round 1).
 
-For each round, dispatch one `Task(subagent_type="qgsd-quorum-slot-worker", model="haiku", max_turns=100, ...)` per slot in **`$DISPATCH_LIST`** as **parallel sibling calls**. Append `prior_positions` to the YAML block for Round 2+ dispatch:
+For each round, dispatch one `Task(subagent_type="nf-quorum-slot-worker", model="haiku", max_turns=100, ...)` per slot in **`$DISPATCH_LIST`** as **parallel sibling calls**. Append `prior_positions` to the YAML block for Round 2+ dispatch:
 
 ```
 slot: <slotName>
@@ -405,7 +405,7 @@ After 10 total rounds with no consensus → **Escalate**.
 
 ```
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
- QGSD ► QUORUM CONSENSUS REACHED
+ nForma ► QUORUM CONSENSUS REACHED
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 
 Question: [question]
@@ -427,7 +427,7 @@ Update the scoreboard: for each model that voted this round, run:
 
 ```bash
 # For native agents:
-node "$HOME/.claude/qgsd-bin/update-scoreboard.cjs" \
+node "$HOME/.claude/nf-bin/update-scoreboard.cjs" \
   --model <model_name> \
   --result <vote_code> \
   --task "<task_label>" \
@@ -436,7 +436,7 @@ node "$HOME/.claude/qgsd-bin/update-scoreboard.cjs" \
   --task-description "<question or topic being debated>"
 
 # For each claude-mcp server (use slot + full model-id, NOT --model):
-node "$HOME/.claude/qgsd-bin/update-scoreboard.cjs" \
+node "$HOME/.claude/nf-bin/update-scoreboard.cjs" \
   --slot <slotName> \
   --model-id <fullModelId> \
   --result <vote_code> \
@@ -519,7 +519,7 @@ Only write the `## Improvements` section when `request_improvements: true` AND i
 
 ```
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
- QGSD ► QUORUM ESCALATING — NO CONSENSUS AFTER 10 ROUNDS
+ nForma ► QUORUM ESCALATING — NO CONSENSUS AFTER 10 ROUNDS
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 
 Question: [question]
@@ -541,7 +541,7 @@ Update the scoreboard: for each model that voted this round, run:
 
 ```bash
 # For native agents:
-node "$HOME/.claude/qgsd-bin/update-scoreboard.cjs" \
+node "$HOME/.claude/nf-bin/update-scoreboard.cjs" \
   --model <model_name> \
   --result <vote_code> \
   --task "<task_label>" \
@@ -550,7 +550,7 @@ node "$HOME/.claude/qgsd-bin/update-scoreboard.cjs" \
   --task-description "<question or topic being debated>"
 
 # For each claude-mcp server (use slot + full model-id, NOT --model):
-node "$HOME/.claude/qgsd-bin/update-scoreboard.cjs" \
+node "$HOME/.claude/nf-bin/update-scoreboard.cjs" \
   --slot <slotName> \
   --model-id <fullModelId> \
   --result <vote_code> \
@@ -581,7 +581,7 @@ Extract command(s) to run from $ARGUMENTS. If unclear, ask the user to specify.
 Display:
 ```
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
- QGSD ► QUORUM: Mode B — Execution + Trace Review
+ nForma ► QUORUM: Mode B — Execution + Trace Review
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 
 Question: [original question]
@@ -615,7 +615,7 @@ $TRACES
 
 ### Dispatch quorum workers via Task (parallel per round)
 
-Dispatch one `Task(subagent_type="qgsd-quorum-slot-worker", model="haiku", max_turns=100, ...)` per slot in **`$DISPATCH_LIST`** (capped to `FAN_OUT_COUNT - 1` external slots) as **parallel sibling calls** in one message turn. Do NOT dispatch slots outside `$DISPATCH_LIST`. Build a YAML prompt block per the slot-worker argument spec:
+Dispatch one `Task(subagent_type="nf-quorum-slot-worker", model="haiku", max_turns=100, ...)` per slot in **`$DISPATCH_LIST`** (capped to `FAN_OUT_COUNT - 1` external slots) as **parallel sibling calls** in one message turn. Do NOT dispatch slots outside `$DISPATCH_LIST`. Build a YAML prompt block per the slot-worker argument spec:
 
 ```
 slot: <slotName>
@@ -643,11 +643,11 @@ prior_positions: |
 Populate `citations:` from the `citations:` field in each model's slot-worker result block. If the result block had no `citations:` field or it was empty, write `(none)`. For Claude's own position, include any file paths or line numbers Claude cited in its reasoning.
 
 Example dispatch (all Tasks in one message turn):
-- `Task(subagent_type="qgsd-quorum-slot-worker", model="haiku", max_turns=100, description="gemini-1 [gemini-cli · gemini-3-pro-preview] quorum R1", prompt=<YAML block>)`
-- `Task(subagent_type="qgsd-quorum-slot-worker", model="haiku", max_turns=100, description="codex-1 [codex-cli · gpt-5.3-codex] quorum R1", prompt=<YAML block>)`
-- `Task(subagent_type="qgsd-quorum-slot-worker", model="haiku", max_turns=100, description="opencode-1 [opencode-cli · grok-code-fast-1] quorum R1", prompt=<YAML block>)`
-- `Task(subagent_type="qgsd-quorum-slot-worker", model="haiku", max_turns=100, description="copilot-1 [copilot-cli · gpt-4.1] quorum R1", prompt=<YAML block>)`
-- `Task(subagent_type="qgsd-quorum-slot-worker", model="haiku", max_turns=100, description="claude-1 [claude-code-router · deepseek-ai/DeepSeek-V3.2] quorum R1", prompt=<YAML block>)` ← one per claude-mcp server with `available: true`
+- `Task(subagent_type="nf-quorum-slot-worker", model="haiku", max_turns=100, description="gemini-1 [gemini-cli · gemini-3-pro-preview] quorum R1", prompt=<YAML block>)`
+- `Task(subagent_type="nf-quorum-slot-worker", model="haiku", max_turns=100, description="codex-1 [codex-cli · gpt-5.3-codex] quorum R1", prompt=<YAML block>)`
+- `Task(subagent_type="nf-quorum-slot-worker", model="haiku", max_turns=100, description="opencode-1 [opencode-cli · grok-code-fast-1] quorum R1", prompt=<YAML block>)`
+- `Task(subagent_type="nf-quorum-slot-worker", model="haiku", max_turns=100, description="copilot-1 [copilot-cli · gpt-4.1] quorum R1", prompt=<YAML block>)`
+- `Task(subagent_type="nf-quorum-slot-worker", model="haiku", max_turns=100, description="claude-1 [claude-code-router · deepseek-ai/DeepSeek-V3.2] quorum R1", prompt=<YAML block>)` ← one per claude-mcp server with `available: true`
 (model="haiku" — slot-workers are orchestrators (read files, build prompt, run Bash subprocess), NOT reasoners. The actual reasoning is done by the external CLI. Haiku is faster with zero quality loss.)
 
 The slot-worker reads repo context, builds the Mode B prompt (with execution traces) from the YAML arguments, calls the slot via `call-quorum-slot.cjs`, and returns a structured result block with a `verdict: APPROVE | REJECT | FLAG` field.
@@ -669,7 +669,7 @@ If split: run deliberation (up to 9 deliberation rounds, max 10 total rounds inc
 
 ```
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
- QGSD ► QUORUM VERDICT
+ nForma ► QUORUM VERDICT
 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 
 ┌────────────────────────────────┬──────────────┬──────────────────────────────────────────┐

package/commands/{qgsd → nf}/reapply-patches.md

@@ -28,7 +28,7 @@ Read `backup-meta.json` from the patches directory.
 ```
 No local patches found. Nothing to reapply.
 
-Local patches are automatically saved when you run /qgsd:update
+Local patches are automatically saved when you run /nf:update
 after modifying any GSD workflow, command, or agent files.
 ```
 Exit.
@@ -76,7 +76,7 @@ For each file in `backup-meta.json`:
 After reapplying, regenerate the file manifest so future updates correctly detect these as user modifications:
 
 ```bash
-# The manifest will be regenerated on next /qgsd:update
+# The manifest will be regenerated on next /nf:update
 # For now, just note which files were modified
 ```
 

package/commands/{qgsd → nf}/remove-phase.md

@@ -1,5 +1,5 @@
 ---
-name: qgsd:remove-phase
+name: nf:remove-phase
 description: Remove a future phase from roadmap and renumber subsequent phases
 argument-hint: <phase-number>
 allowed-tools:
@@ -16,7 +16,7 @@ Output: Phase deleted, all subsequent phases renumbered, git commit as historica
 </objective>
 
 <execution_context>
-@~/.claude/qgsd/workflows/remove-phase.md
+@~/.claude/nf/workflows/remove-phase.md
 </execution_context>
 
 <context>
@@ -26,6 +26,6 @@ Roadmap and state are resolved in-workflow via `init phase-op` and targeted read
 </context>
 
 <process>
-Execute the remove-phase workflow from @~/.claude/qgsd/workflows/remove-phase.md end-to-end.
+Execute the remove-phase workflow from @~/.claude/nf/workflows/remove-phase.md end-to-end.
 Preserve all validation gates (future phase check, work check), renumbering logic, and commit.
 </process>

package/commands/{qgsd → nf}/research-phase.md

@@ -1,6 +1,6 @@
 ---
-name: qgsd:research-phase
-description: Research how to implement a phase (standalone - usually use /qgsd:plan-phase instead)
+name: nf:research-phase
+description: Research how to implement a phase (standalone - usually use /nf:plan-phase instead)
 argument-hint: "[phase]"
 allowed-tools:
   - Read
@@ -9,9 +9,9 @@ allowed-tools:
 ---
 
 <objective>
-Research how to implement a phase. Spawns qgsd-phase-researcher agent with phase context.
+Research how to implement a phase. Spawns nf-phase-researcher agent with phase context.
 
-**Note:** This is a standalone research command. For most workflows, use `/qgsd:plan-phase` which integrates research automatically.
+**Note:** This is a standalone research command. For most workflows, use `/nf:plan-phase` which integrates research automatically.
 
 **Use this command when:**
 - You want to research without planning yet
@@ -34,20 +34,20 @@ Normalize phase input in step 1 before any directory lookups.
 ## 0. Initialize Context
 
 ```bash
-INIT=$(node ~/.claude/qgsd/bin/gsd-tools.cjs init phase-op "$ARGUMENTS")
+INIT=$(node ~/.claude/nf/bin/gsd-tools.cjs init phase-op "$ARGUMENTS")
 ```
 
 Extract from init JSON: `phase_dir`, `phase_number`, `phase_name`, `phase_found`, `commit_docs`, `has_research`, `state_path`, `requirements_path`, `context_path`, `research_path`.
 
 Resolve researcher model:
 ```bash
-RESEARCHER_MODEL=$(node ~/.claude/qgsd/bin/gsd-tools.cjs resolve-model qgsd-phase-researcher --raw)
+RESEARCHER_MODEL=$(node ~/.claude/nf/bin/gsd-tools.cjs resolve-model nf-phase-researcher --raw)
 ```
 
 ## 1. Validate Phase
 
 ```bash
-PHASE_INFO=$(node ~/.claude/qgsd/bin/gsd-tools.cjs roadmap get-phase "${phase_number}")
+PHASE_INFO=$(node ~/.claude/nf/bin/gsd-tools.cjs roadmap get-phase "${phase_number}")
 ```
 
 **If `found` is false:** Error and exit. **If `found` is true:** Extract `phase_number`, `phase_name`, `goal` from JSON.
@@ -71,7 +71,7 @@ Use paths from INIT (do not inline file contents in orchestrator context):
 
 Present summary with phase description and what files the researcher will load.
 
-## 4. Spawn qgsd-phase-researcher Agent
+## 4. Spawn nf-phase-researcher Agent
 
 Research modes: ecosystem (default), feasibility, implementation, comparison.
 
@@ -109,7 +109,7 @@ Mode: ecosystem
 </additional_context>
 
 <downstream_consumer>
-Your RESEARCH.md will be loaded by `/qgsd:plan-phase` which uses specific sections:
+Your RESEARCH.md will be loaded by `/nf:plan-phase` which uses specific sections:
 - `## Standard Stack` → Plans use these libraries
 - `## Architecture Patterns` → Task structure follows these
 - `## Don't Hand-Roll` → Tasks NEVER build custom solutions for listed problems
@@ -135,7 +135,7 @@ Write to: .planning/phases/${PHASE}-{slug}/${PHASE}-RESEARCH.md
 
 ```
 Task(
-  prompt="First, read ~/.claude/agents/qgsd-phase-researcher.md for your role and instructions.\n\n" + filled_prompt,
+  prompt="First, read ~/.claude/agents/nf-phase-researcher.md for your role and instructions.\n\n" + filled_prompt,
   subagent_type="general-purpose",
   model="{researcher_model}",
   description="Research Phase {phase}"
@@ -171,7 +171,7 @@ Continue research for Phase {phase_number}: {phase_name}
 
 ```
 Task(
-  prompt="First, read ~/.claude/agents/qgsd-phase-researcher.md for your role and instructions.\n\n" + continuation_prompt,
+  prompt="First, read ~/.claude/agents/nf-phase-researcher.md for your role and instructions.\n\n" + continuation_prompt,
   subagent_type="general-purpose",
   model="{researcher_model}",
   description="Continue research Phase {phase}"
@@ -183,7 +183,7 @@ Task(
 <success_criteria>
 - [ ] Phase validated against roadmap
 - [ ] Existing research checked
-- [ ] qgsd-phase-researcher spawned with context
+- [ ] nf-phase-researcher spawned with context
 - [ ] Checkpoints handled correctly
 - [ ] User knows next steps
 </success_criteria>

package/commands/{qgsd → nf}/resume-work.md

@@ -1,5 +1,5 @@
 ---
-name: qgsd:resume-work
+name: nf:resume-work
 description: Resume work from previous session with full context restoration
 allowed-tools:
   - Read
@@ -22,11 +22,11 @@ Routes to the resume-project workflow which handles:
   </objective>
 
 <execution_context>
-@~/.claude/qgsd/workflows/resume-project.md
+@~/.claude/nf/workflows/resume-project.md
 </execution_context>
 
 <process>
-**Follow the resume-project workflow** from `@~/.claude/qgsd/workflows/resume-project.md`.
+**Follow the resume-project workflow** from `@~/.claude/nf/workflows/resume-project.md`.
 
 The workflow handles all resumption logic including:
 

package/commands/nf/review-requirements.md

@@ -0,0 +1,31 @@
+---
+name: nf:review-requirements
+description: Review all requirements for quality — flags specificity issues, redundant overlaps, unmeasurable requirements, and merge candidates
+argument-hint: [--repair] [--dry-run] [--category="Category Name"] [--ids=REQ-01,REQ-02]
+allowed-tools:
+  - Read
+  - Write
+  - Edit
+  - Bash
+  - Agent
+  - AskUserQuestion
+---
+
+<objective>
+Scan all requirements in `.planning/formal/requirements.json` for quality issues:
+specificity (too narrow), redundancy (overlaps/duplicates), and measurability (vague/untestable).
+Proposes and optionally applies fixes.
+</objective>
+
+<execution_context>
+@~/.claude/nf/workflows/review-requirements.md
+</execution_context>
+
+<process>
+Execute the review-requirements workflow from @~/.claude/nf/workflows/review-requirements.md end-to-end.
+Pass through all --flags from arguments:
+  --repair        Fully autonomous mode — auto-apply safe fixes, skip prompts
+  --dry-run       Show proposed changes without writing
+  --category      Focus on a specific category
+  --ids           Focus on specific requirement IDs
+</process>

package/commands/{qgsd → nf}/set-profile.md

@@ -1,5 +1,5 @@
 ---
-name: qgsd:set-profile
+name: nf:set-profile
 description: Switch model profile for GSD agents (quality/balanced/budget)
 argument-hint: <profile>
 allowed-tools:
@@ -19,11 +19,11 @@ Routes to the set-profile workflow which handles:
 </objective>
 
 <execution_context>
-@~/.claude/qgsd/workflows/set-profile.md
+@~/.claude/nf/workflows/set-profile.md
 </execution_context>
 
 <process>
-**Follow the set-profile workflow** from `@~/.claude/qgsd/workflows/set-profile.md`.
+**Follow the set-profile workflow** from `@~/.claude/nf/workflows/set-profile.md`.
 
 The workflow handles all logic including:
 1. Profile argument validation

package/commands/{qgsd → nf}/settings.md

@@ -1,5 +1,5 @@
 ---
-name: qgsd:settings
+name: nf:settings
 description: Project manager hub — dashboard, smart routing, and configuration
 allowed-tools:
   - Read
@@ -14,20 +14,20 @@ allowed-tools:
 Guided project manager hub with state-aware dashboard and categorized action menu.
 
 Shows project status (milestone, phase, progress, config) then routes to:
-- Continue Working — smart routing based on project state (same as /qgsd:progress)
+- Continue Working — smart routing based on project state (same as /nf:progress)
 - Project Management — phases, milestones, todos, debug, roadmap
 - Configuration — workflow settings, project profile, baselines, quorum agents
-- Quick Task — ad-hoc tasks via /qgsd:quick
+- Quick Task — ad-hoc tasks via /nf:quick
 
-Backward compatible: /qgsd:settings --config goes directly to workflow settings (original 6-question config form).
+Backward compatible: /nf:settings --config goes directly to workflow settings (original 6-question config form).
 </objective>
 
 <execution_context>
-@~/.claude/qgsd/workflows/settings.md
+@~/.claude/nf/workflows/settings.md
 </execution_context>
 
 <process>
-**Follow the settings workflow** from `@~/.claude/qgsd/workflows/settings.md`.
+**Follow the settings workflow** from `@~/.claude/nf/workflows/settings.md`.
 
 The workflow handles all logic including:
 1. Flag check (--config for backward compat)