create-hq 5.0.0 → 5.2.0

package/template/.claude/commands/prd.md

@@ -1,238 +0,0 @@
----
-description: Plan a project and generate PRD for execution
-allowed-tools: Task, Read, Glob, Grep, Write, Bash, AskUserQuestion
-argument-hint: [project/feature description]
-visibility: public
----
-
-# /prd - Project Planning & PRD Generation
-
-Create execution-ready PRDs with full HQ context awareness.
-
-**User's input:** $ARGUMENTS
-
-**Important:** Do NOT implement. Just create the PRD.
-
-## Step 1: Get Project Description
-
-If $ARGUMENTS provided, use as starting point.
-If empty, ask: "Describe what you want to build or accomplish."
-
-## Step 2: Scan HQ Context
-
-Before asking questions, explore HQ:
-
-**Companies & Context:**
-- Read `agents.md` (roles, priorities)
-- Glob `companies/*/knowledge/` (which companies exist)
-
-**Workers:**
-- Read `workers/registry.yaml`
-- Glob `workers/*/worker.yaml`, `workers/public/dev-team/*/worker.yaml`
-
-**Existing Projects:**
-- `ls projects/`
-- Glob `projects/*/prd.json` (check overlap)
-
-**Knowledge (use qmd, not Grep):**
-- `qmd vsearch "<description keywords>" --json -n 10` — semantic search for related knowledge, prior work, workers
-
-Present:
-```
-Scanned HQ:
-- Workers: {relevant list}
-- Existing projects: {list or "none matching"}
-- Relevant knowledge: {if any}
-- Category: [company-specific | cross-company | personal | HQ infrastructure]
-```
-
-## Step 3: Get + Validate Project Name
-
-Ask for project slug (or infer from description). Then:
-1. Check if `projects/{name}/` exists
-   - If exists: "Project exists. Continue editing or choose different name?"
-2. Validate slug format (lowercase, hyphens only)
-
-## Step 4: Discovery Interview
-
-Ask questions in batches. Users respond: "1A, 2C"
-
-**Batch 1: Problem & Success**
-1. Core problem or goal?
-2. What does success look like? (measurable)
-3. Who benefits?
-
-**Batch 2: Scope & Constraints**
-4. What's in scope for MVP?
-5. Hard constraints (time, tech, budget)?
-6. Dependencies on other projects?
-
-**Batch 3: Integration & Quality**
-7. Quality gates? (detect repo from scan, suggest commands)
-   A. `pnpm typecheck && pnpm lint`
-   B. `npm run typecheck && npm run lint`
-   C. None (no automated checks)
-   D. Other: [specify]
-8. Based on scan: "Should this use {relevant workers}?"
-9. Does this need a new worker or skill?
-10. Repo path? (e.g. `repos/private/{name}`, or "none" if non-code)
-11. Branch name? (default: `feature/{project-name}`)
-
-## Step 5: Generate PRD
-
-Create `projects/{name}/` folder with two files.
-
-### Primary: projects/{name}/prd.json
-
-This is the **source of truth**. `/run-project` and `/execute-task` consume this file.
-
-```json
-{
-  "name": "{project-slug}",
-  "description": "{1-sentence goal}",
-  "branchName": "feature/{name}",
-  "userStories": [
-    {
-      "id": "US-001",
-      "title": "{Story title}",
-      "description": "{As a [user], I want [feature] so that [benefit]}",
-      "acceptanceCriteria": ["{Specific verifiable criterion}"],
-      "priority": 1,
-      "passes": false,
-      "labels": [],
-      "dependsOn": [],
-      "notes": ""
-    }
-  ],
-  "metadata": {
-    "createdAt": "{ISO8601}",
-    "goal": "{Overall project goal}",
-    "successCriteria": "{Measurable outcome}",
-    "qualityGates": ["{commands from Batch 3}"],
-    "repoPath": "{repos/private/repo-name or empty}",
-    "relatedWorkers": ["{worker-ids from scan}"],
-    "knowledge": ["{relevant knowledge paths}"]
-  }
-}
-```
-
-### Derived: projects/{name}/README.md
-
-Generate FROM the prd.json data. Human-friendly view.
-
-```markdown
-# {name from prd.json}
-
-**Goal:** {metadata.goal}
-**Success:** {metadata.successCriteria}
-**Repo:** {metadata.repoPath}
-**Branch:** {branchName}
-
-## Overview
-{description}
-
-## Quality Gates
-- `{metadata.qualityGates[0]}`
-
-## User Stories
-
-### US-001: {title}
-**Description:** {description}
-**Priority:** {priority}
-**Depends on:** {dependsOn or "None"}
-
-**Acceptance Criteria:**
-- [ ] {criterion 1}
-- [ ] {criterion 2}
-
-## Non-Goals
-{What's out of scope}
-
-## Technical Considerations
-{Constraints, dependencies}
-
-## Open Questions
-{Remaining questions}
-```
-
-## Step 6: Register with Orchestrator
-
-Read `workspace/orchestrator/state.json`. Append to `projects` array:
-
-```json
-{
-  "name": "{name}",
-  "state": "READY",
-  "prdPath": "projects/{name}/prd.json",
-  "updatedAt": "{ISO8601}",
-  "storiesComplete": 0,
-  "storiesTotal": "{N}",
-  "checkedOutFiles": []
-}
-```
-
-If project already exists in state.json, update it instead of duplicating.
-
-## Step 7: Sync to Beads
-
-```bash
-npx tsx scripts/prd-to-beads.ts --project={name}
-```
-
-Silent — just log success/failure.
-
-## Step 7.5: Capture Learning (Auto-Learn)
-
-Run `/learn` to register the new project in the learning system:
-```json
-{
-  "source": "build-activity",
-  "severity": "medium",
-  "scope": "global",
-  "rule": "Project {name} exists at projects/{name}/ with {N} stories targeting {repoPath or 'no repo'}",
-  "context": "Created via /prd"
-}
-```
-
-Also reindex: `qmd update 2>/dev/null || true`
-
-**Update INDEX.md:** Regenerate `projects/INDEX.md` per `knowledge/public/hq-core/index-md-spec.md`.
-
-## Step 8: Execution Choice
-
-Based on complexity, recommend execution path:
-
-**>3 stories OR dependencies OR multi-file:** recommend `/run-project`
-**1-3 simple stories:** recommend in-process or `/execute-task`
-
-Tell user:
-```
-Project **{name}** created with {N} user stories.
-
-Files:
-  projects/{name}/prd.json   (source of truth)
-  projects/{name}/README.md  (human-readable)
-
-Recommended execution:
-1. /run-project {name}  (orchestrator loop, crash recovery)
-2. /execute-task {name}/US-001  (run stories one at a time)
-3. Execute now (in this session, simple projects only)
-```
-
-## Story Guidelines
-
-- Each story completable in one AI session
-- Acceptance criteria must be verifiable (not "works correctly")
-- Order: schema → backend → UI → integration
-- Keep stories atomic (one deliverable each)
-- Every story starts with `passes: false`
-
-## Rules
-
-- Scan HQ first, ask questions second
-- Batch questions (don't overwhelm)
-- **prd.json is the source of truth** — README.md is derived from it, never the reverse
-- **All stories start with `passes: false`** — `/run-project` marks them true
-- **Do NOT use EnterPlanMode** — this skill IS planning
-- **Do NOT use TodoWrite** — PRD stories track tasks
-- **Do NOT implement** — just create the PRD

package/template/.claude/commands/reanchor.md

@@ -1,51 +0,0 @@
----
-description: Force reanchoring conversation before next task
-allowed-tools: Read, Bash, AskUserQuestion
-argument-hint:
-visibility: public
----
-
-# /reanchor - Pause and Realign
-
-Stop and realign before continuing. Use when you want to ensure the agent is on track.
-
-## Process
-
-1. **Load INDEX context**
-   - Read root `INDEX.md` (directory map, recent threads, workers)
-   - Read `workspace/orchestrator/INDEX.md` (active projects + progress)
-   - Read `workspace/threads/handoff.json` if exists (last handoff context)
-
-2. **Check git state**
-   ```bash
-   git log --oneline -5
-   ```
-
-3. **Display combined summary**
-   ```
-   HQ State:
-   - Active projects: {from orchestrator INDEX — name + progress %}
-   - Recent threads: {from root INDEX — last 5}
-   - Last handoff: {from handoff.json — summary + timestamp}
-   - Recent commits: {from git log}
-   ```
-
-4. **Ask for focus**
-   Use AskUserQuestion:
-   - "What's the focus for the next stretch?"
-   - Options: active projects from orchestrator INDEX + handoff next_steps + "Something else"
-
-5. **Load company context if needed**
-   If user picks a company-specific focus, read that company's `knowledge/INDEX.md`:
-   - `companies/{company}/knowledge/INDEX.md`
-
-6. **Wait for response** before proceeding
-
-## Why Reanchor
-
-Reanchoring prevents:
-- Context drift (agent goes off-track)
-- Wasted work (building wrong thing)
-- Compounding errors
-
-Core Ralph methodology: small loops with human checkpoints.

package/template/.claude/commands/remember.md

@@ -1,126 +0,0 @@
----
-description: Capture learnings when things don't work right
-allowed-tools: Read, Write, Edit, Bash, Glob, Grep, AskUserQuestion
-argument-hint: [what went wrong]
-visibility: public
----
-
-# /remember - Capture Learnings
-
-When something doesn't work right, capture the learning as a rule in the relevant file.
-
-**User's input:** $ARGUMENTS
-
-## Core Principle
-
-Don't store learnings in a separate database. Inject rules directly into:
-- **worker.yaml** `instructions:` block — worker-specific
-- **`.claude/commands/*.md`** `## Rules` section — command-specific
-- **`.claude/CLAUDE.md`** — global rules
-- **skill files** — skill-specific
-
-## Step 1: Gather Context
-
-If $ARGUMENTS empty, ask: "What happened that you want to capture?"
-
-Then ask: "What's the fix or rule to prevent this?"
-
-## Step 2: Detect Injection Target
-
-Analyze the learning and conversation context to determine target:
-
-1. **Check for active worker/command/skill** in recent conversation
-2. **Parse the learning** for keywords (worker names, command names, file types)
-3. **Default:** If unclear, target `.claude/CLAUDE.md`
-
-Present suggestion:
-```
-Target: {file path}
-Section: {instructions: | ## Rules | relevant heading}
-
-Proposed rule:
-- {NEVER/ALWAYS/condition}: {rule}
-
-Is this the right place? [Y/n/specify different target]
-```
-
-## Step 3: Check for Duplicates
-
-Search for similar existing rules:
-```bash
-qmd vsearch "{rule description}" --json -n 5
-```
-
-If similar found:
-```
-Similar rule found in {file}:
-  "{existing rule}"
-
-Options:
-1. Skip (already captured)
-2. Merge (combine rules)
-3. Add anyway (different enough)
-```
-
-## Step 4: Inject Rule via /learn
-
-Delegate to the `/learn` pipeline:
-
-Run `/learn` with:
-- **rule**: The rule from steps 1-2
-- **source**: `user-correction`
-- **severity**: `high`
-- **scope**: Detected from step 2 (worker:{id}, command:{name}, or global)
-
-`/learn` injects the rule directly into the file it governs (worker.yaml, command .md, knowledge file, or CLAUDE.md). User corrections **ALWAYS** also promote to CLAUDE.md `## Learned Rules`.
-
-`/learn` handles: dedup, injection, event logging, qmd reindexing.
-
-## Step 5: Report
-
-```
-Rule captured via /learn:
-  Rule: {the rule}
-  Injected: {target file path}
-  Global: promoted (user correction)
-  Event: workspace/learnings/learn-{timestamp}.json
-
-Search with: /search "{keywords}"
-```
-
-## Rule Format
-
-Match existing style in target file. Common patterns:
-- `- NEVER: {anti-pattern}`
-- `- ALWAYS: {pattern}`
-- `- {condition} → {action}`
-- `- **{keyword}**: {rule}`
-
-## Examples
-
-### Worker-specific
-User: "The CFO worker kept trying to write to Stripe instead of just reading"
-Target: `workers/private/cfo-{company}/worker.yaml`
-Rule: `- NEVER: Write or modify data in Stripe. All operations are read-only.`
-
-### Command-specific
-User: "/prd kept asking too many questions at once"
-Target: `.claude/commands/prd.md`
-Rule: `- Limit discovery batches to 3-4 questions maximum`
-
-### Global
-User: "Claude kept creating new files instead of editing existing ones"
-Target: `.claude/CLAUDE.md`
-Rule: `- ALWAYS: Edit existing files. Never create new files unless explicitly required.`
-
-## Ralph Integration
-
-Workers can invoke `/remember` after back-pressure failures:
-- Test failure → capture what broke
-- Lint error → capture style rule
-- Build failure → capture constraint
-
-Automatically capture learnings when:
-1. A task fails verification
-2. User corrects behavior
-3. Workaround discovered