oh-my-claude-sisyphus 3.8.15 → 3.9.0

package/scripts/keyword-detector.mjs

@@ -73,10 +73,20 @@ function extractPrompt(input) {
   }
 }
 
-// Remove code blocks to prevent false positives
-function removeCodeBlocks(text) {
+// Sanitize text to prevent false positives from code blocks, XML tags, URLs, and file paths
+function sanitizeForKeywordDetection(text) {
   return text
+    // 1. Strip XML-style tag blocks: <tag-name ...>...</tag-name> (multi-line, greedy on tag name)
+    .replace(/<(\w[\w-]*)[\s>][\s\S]*?<\/\1>/g, '')
+    // 2. Strip self-closing XML tags: <tag-name />, <tag-name attr="val" />
+    .replace(/<\w[\w-]*(?:\s[^>]*)?\s*\/>/g, '')
+    // 3. Strip URLs: http://... or https://... up to whitespace
+    .replace(/https?:\/\/[^\s)>\]]+/g, '')
+    // 4. Strip file paths: /foo/bar/baz or foo/bar/baz (using lookbehind to avoid consuming leading char)
+    .replace(/(?<=^|[\s"'`(])(?:\/)?(?:[\w.-]+\/)+[\w.-]+/gm, '')
+    // 5. Strip markdown code blocks (existing)
     .replace(/```[\s\S]*?```/g, '')
+    // 6. Strip inline code (existing)
     .replace(/`[^`]+`/g, '');
 }
 
@@ -225,7 +235,7 @@ async function main() {
       return;
     }
 
-    const cleanPrompt = removeCodeBlocks(prompt).toLowerCase();
+    const cleanPrompt = sanitizeForKeywordDetection(prompt).toLowerCase();
 
     // Collect all matching keywords
     const matches = [];

package/scripts/persistent-mode.mjs

@@ -44,19 +44,37 @@ function writeJsonFile(path, data) {
 }
 
 /**
- * Read state file from local or global location, tracking the source.
+ * Staleness threshold for mode states (2 hours in milliseconds).
+ * States older than this are treated as inactive to prevent stale state
+ * from causing the stop hook to malfunction in new sessions.
  */
-function readStateFile(stateDir, globalStateDir, filename) {
-  const localPath = join(stateDir, filename);
-  const globalPath = join(globalStateDir, filename);
+const STALE_STATE_THRESHOLD_MS = 2 * 60 * 60 * 1000; // 2 hours
+
+/**
+ * Check if a state is stale based on its timestamps.
+ * A state is considered stale if it hasn't been updated recently.
+ * We check both `last_checked_at` and `started_at` - using whichever is more recent.
+ */
+function isStaleState(state) {
+  if (!state) return true;
+
+  const lastChecked = state.last_checked_at ? new Date(state.last_checked_at).getTime() : 0;
+  const startedAt = state.started_at ? new Date(state.started_at).getTime() : 0;
+  const mostRecent = Math.max(lastChecked, startedAt);
 
-  let state = readJsonFile(localPath);
-  if (state) return { state, path: localPath };
+  if (mostRecent === 0) return true; // No valid timestamps
 
-  state = readJsonFile(globalPath);
-  if (state) return { state, path: globalPath };
+  const age = Date.now() - mostRecent;
+  return age > STALE_STATE_THRESHOLD_MS;
+}
 
-  return { state: null, path: localPath }; // Default to local for new writes
+/**
+ * Read state file from local location only.
+ */
+function readStateFile(stateDir, filename) {
+  const localPath = join(stateDir, filename);
+  const state = readJsonFile(localPath);
+  return { state, path: localPath };
 }
 
 /**
@@ -122,7 +140,6 @@ function countIncompleteTodos(sessionId, projectDir) {
 function isContextLimitStop(data) {
   const reason = (data.stop_reason || data.stopReason || '').toLowerCase();
 
-  // Known context-limit patterns (both confirmed and defensive)
   const contextPatterns = [
     'context_limit',
     'context_window',
@@ -139,7 +156,6 @@ function isContextLimitStop(data) {
     return true;
   }
 
-  // Also check end_turn_reason (API-level field)
   const endTurnReason = (data.end_turn_reason || data.endTurnReason || '').toLowerCase();
   if (endTurnReason && contextPatterns.some(p => endTurnReason.includes(p))) {
     return true;
@@ -173,7 +189,6 @@ async function main() {
     const directory = data.directory || process.cwd();
     const sessionId = data.sessionId || data.session_id || '';
     const stateDir = join(directory, '.omc', 'state');
-    const globalStateDir = join(homedir(), '.omc', 'state');
 
     // CRITICAL: Never block context-limit stops.
     // Blocking these causes a deadlock where Claude Code cannot compact.
@@ -189,14 +204,14 @@ async function main() {
       return;
     }
 
-    // Read all mode states (local-first with fallback to global)
-    const ralph = readStateFile(stateDir, globalStateDir, 'ralph-state.json');
-    const autopilot = readStateFile(stateDir, globalStateDir, 'autopilot-state.json');
-    const ultrapilot = readStateFile(stateDir, globalStateDir, 'ultrapilot-state.json');
-    const ultrawork = readStateFile(stateDir, globalStateDir, 'ultrawork-state.json');
-    const ecomode = readStateFile(stateDir, globalStateDir, 'ecomode-state.json');
-    const ultraqa = readStateFile(stateDir, globalStateDir, 'ultraqa-state.json');
-    const pipeline = readStateFile(stateDir, globalStateDir, 'pipeline-state.json');
+    // Read all mode states (local-only)
+    const ralph = readStateFile(stateDir, 'ralph-state.json');
+    const autopilot = readStateFile(stateDir, 'autopilot-state.json');
+    const ultrapilot = readStateFile(stateDir, 'ultrapilot-state.json');
+    const ultrawork = readStateFile(stateDir, 'ultrawork-state.json');
+    const ecomode = readStateFile(stateDir, 'ecomode-state.json');
+    const ultraqa = readStateFile(stateDir, 'ultraqa-state.json');
+    const pipeline = readStateFile(stateDir, 'pipeline-state.json');
 
     // Swarm uses swarm-summary.json (not swarm-state.json) + marker file
     const swarmMarker = existsSync(join(stateDir, 'swarm-active.marker'));
@@ -208,34 +223,37 @@ async function main() {
     const totalIncomplete = taskCount + todoCount;
 
     // Priority 1: Ralph Loop (explicit persistence mode)
-    if (ralph.state?.active) {
+    // Skip if state is stale (older than 2 hours) - prevents blocking new sessions
+    if (ralph.state?.active && !isStaleState(ralph.state)) {
       const iteration = ralph.state.iteration || 1;
       const maxIter = ralph.state.max_iterations || 100;
 
       if (iteration < maxIter) {
         ralph.state.iteration = iteration + 1;
+        ralph.state.last_checked_at = new Date().toISOString();
         writeJsonFile(ralph.path, ralph.state);
 
         console.log(JSON.stringify({
           decision: 'block',
-          reason: `[RALPH LOOP - ITERATION ${iteration + 1}/${maxIter}] Work is NOT done. Continue. When complete, output: <promise>${ralph.state.completion_promise || 'DONE'}</promise>\n${ralph.state.prompt ? `Task: ${ralph.state.prompt}` : ''}`
+          reason: `[RALPH LOOP - ITERATION ${iteration + 1}/${maxIter}] Work is NOT done. Continue working.\nWhen FULLY complete (after Architect verification), run /oh-my-claudecode:cancel to cleanly exit ralph mode and clean up all state files. If cancel fails, retry with /oh-my-claudecode:cancel --force.\n${ralph.state.prompt ? `Task: ${ralph.state.prompt}` : ''}`
         }));
         return;
       }
     }
 
     // Priority 2: Autopilot (high-level orchestration)
-    if (autopilot.state?.active) {
+    if (autopilot.state?.active && !isStaleState(autopilot.state)) {
       const phase = autopilot.state.phase || 'unknown';
       if (phase !== 'complete') {
         const newCount = (autopilot.state.reinforcement_count || 0) + 1;
         if (newCount <= 20) {
           autopilot.state.reinforcement_count = newCount;
+          autopilot.state.last_checked_at = new Date().toISOString();
           writeJsonFile(autopilot.path, autopilot.state);
 
           console.log(JSON.stringify({
             decision: 'block',
-            reason: `[AUTOPILOT - Phase: ${phase}] Autopilot not complete. Continue working.`
+            reason: `[AUTOPILOT - Phase: ${phase}] Autopilot not complete. Continue working. When all phases are complete, run /oh-my-claudecode:cancel to cleanly exit and clean up state files. If cancel fails, retry with /oh-my-claudecode:cancel --force.`
           }));
           return;
         }
@@ -243,18 +261,19 @@ async function main() {
     }
 
     // Priority 3: Ultrapilot (parallel autopilot)
-    if (ultrapilot.state?.active) {
+    if (ultrapilot.state?.active && !isStaleState(ultrapilot.state)) {
       const workers = ultrapilot.state.workers || [];
       const incomplete = workers.filter(w => w.status !== 'complete' && w.status !== 'failed').length;
       if (incomplete > 0) {
         const newCount = (ultrapilot.state.reinforcement_count || 0) + 1;
         if (newCount <= 20) {
           ultrapilot.state.reinforcement_count = newCount;
+          ultrapilot.state.last_checked_at = new Date().toISOString();
           writeJsonFile(ultrapilot.path, ultrapilot.state);
 
           console.log(JSON.stringify({
             decision: 'block',
-            reason: `[ULTRAPILOT] ${incomplete} workers still running. Continue.`
+            reason: `[ULTRAPILOT] ${incomplete} workers still running. Continue working. When all workers complete, run /oh-my-claudecode:cancel to cleanly exit and clean up state files. If cancel fails, retry with /oh-my-claudecode:cancel --force.`
           }));
           return;
         }
@@ -262,17 +281,18 @@ async function main() {
     }
 
     // Priority 4: Swarm (coordinated agents with SQLite)
-    if (swarmMarker && swarmSummary?.active) {
+    if (swarmMarker && swarmSummary?.active && !isStaleState(swarmSummary)) {
       const pending = (swarmSummary.tasks_pending || 0) + (swarmSummary.tasks_claimed || 0);
       if (pending > 0) {
         const newCount = (swarmSummary.reinforcement_count || 0) + 1;
         if (newCount <= 15) {
           swarmSummary.reinforcement_count = newCount;
+          swarmSummary.last_checked_at = new Date().toISOString();
           writeJsonFile(join(stateDir, 'swarm-summary.json'), swarmSummary);
 
           console.log(JSON.stringify({
             decision: 'block',
-            reason: `[SWARM ACTIVE] ${pending} tasks remain. Continue working.`
+            reason: `[SWARM ACTIVE] ${pending} tasks remain. Continue working. When all tasks are done, run /oh-my-claudecode:cancel to cleanly exit and clean up state files. If cancel fails, retry with /oh-my-claudecode:cancel --force.`
           }));
           return;
         }
@@ -280,18 +300,19 @@ async function main() {
     }
 
     // Priority 5: Pipeline (sequential stages)
-    if (pipeline.state?.active) {
+    if (pipeline.state?.active && !isStaleState(pipeline.state)) {
       const currentStage = pipeline.state.current_stage || 0;
       const totalStages = pipeline.state.stages?.length || 0;
       if (currentStage < totalStages) {
         const newCount = (pipeline.state.reinforcement_count || 0) + 1;
         if (newCount <= 15) {
           pipeline.state.reinforcement_count = newCount;
+          pipeline.state.last_checked_at = new Date().toISOString();
           writeJsonFile(pipeline.path, pipeline.state);
 
           console.log(JSON.stringify({
             decision: 'block',
-            reason: `[PIPELINE - Stage ${currentStage + 1}/${totalStages}] Pipeline not complete. Continue.`
+            reason: `[PIPELINE - Stage ${currentStage + 1}/${totalStages}] Pipeline not complete. Continue working. When all stages complete, run /oh-my-claudecode:cancel to cleanly exit and clean up state files. If cancel fails, retry with /oh-my-claudecode:cancel --force.`
           }));
           return;
         }
@@ -299,16 +320,17 @@ async function main() {
     }
 
     // Priority 6: UltraQA (QA cycling)
-    if (ultraqa.state?.active) {
+    if (ultraqa.state?.active && !isStaleState(ultraqa.state)) {
       const cycle = ultraqa.state.cycle || 1;
       const maxCycles = ultraqa.state.max_cycles || 10;
       if (cycle < maxCycles && !ultraqa.state.all_passing) {
         ultraqa.state.cycle = cycle + 1;
+        ultraqa.state.last_checked_at = new Date().toISOString();
         writeJsonFile(ultraqa.path, ultraqa.state);
 
         console.log(JSON.stringify({
           decision: 'block',
-          reason: `[ULTRAQA - Cycle ${cycle + 1}/${maxCycles}] Tests not all passing. Continue fixing.`
+          reason: `[ULTRAQA - Cycle ${cycle + 1}/${maxCycles}] Tests not all passing. Continue fixing. When all tests pass, run /oh-my-claudecode:cancel to cleanly exit and clean up state files. If cancel fails, retry with /oh-my-claudecode:cancel --force.`
         }));
         return;
       }
@@ -316,7 +338,7 @@ async function main() {
 
     // Priority 7: Ultrawork - ALWAYS continue while active (not just when tasks exist)
     // This prevents false stops from bash errors, transient failures, etc.
-    if (ultrawork.state?.active) {
+    if (ultrawork.state?.active && !isStaleState(ultrawork.state)) {
       const newCount = (ultrawork.state.reinforcement_count || 0) + 1;
       const maxReinforcements = ultrawork.state.max_reinforcements || 50;
 
@@ -330,25 +352,29 @@ async function main() {
       ultrawork.state.last_checked_at = new Date().toISOString();
       writeJsonFile(ultrawork.path, ultrawork.state);
 
-      // Build continuation message
-      let reason = `[ULTRAWORK #${newCount}] Mode active - continue working.`;
+      let reason = `[ULTRAWORK #${newCount}/${maxReinforcements}] Mode active.`;
+
       if (totalIncomplete > 0) {
         const itemType = taskCount > 0 ? 'Tasks' : 'todos';
-        reason = `[ULTRAWORK #${newCount}] ${totalIncomplete} incomplete ${itemType}. Continue working.`;
+        reason += ` ${totalIncomplete} incomplete ${itemType} remain. Continue working.`;
+      } else if (newCount >= 3) {
+        // Only suggest cancel after minimum iterations (guard against no-tasks-created scenario)
+        reason += ` If all work is complete, run /oh-my-claudecode:cancel to cleanly exit ultrawork mode and clean up state files. If cancel fails, retry with /oh-my-claudecode:cancel --force. Otherwise, continue working.`;
+      } else {
+        // Early iterations with no tasks yet - just tell LLM to continue
+        reason += ` Continue working - create Tasks to track your progress.`;
       }
+
       if (ultrawork.state.original_prompt) {
         reason += `\nTask: ${ultrawork.state.original_prompt}`;
       }
 
-      console.log(JSON.stringify({
-        decision: 'block',
-        reason: reason
-      }));
+      console.log(JSON.stringify({ decision: 'block', reason }));
       return;
     }
 
     // Priority 8: Ecomode - ALWAYS continue while active
-    if (ecomode.state?.active) {
+    if (ecomode.state?.active && !isStaleState(ecomode.state)) {
       const newCount = (ecomode.state.reinforcement_count || 0) + 1;
       const maxReinforcements = ecomode.state.max_reinforcements || 50;
 
@@ -359,18 +385,23 @@ async function main() {
       }
 
       ecomode.state.reinforcement_count = newCount;
+      ecomode.state.last_checked_at = new Date().toISOString();
       writeJsonFile(ecomode.path, ecomode.state);
 
-      let reason = `[ECOMODE #${newCount}] Mode active - continue working.`;
+      let reason = `[ECOMODE #${newCount}/${maxReinforcements}] Mode active.`;
+
       if (totalIncomplete > 0) {
         const itemType = taskCount > 0 ? 'Tasks' : 'todos';
-        reason = `[ECOMODE #${newCount}] ${totalIncomplete} incomplete ${itemType}. Continue working.`;
+        reason += ` ${totalIncomplete} incomplete ${itemType} remain. Continue working.`;
+      } else if (newCount >= 3) {
+        // Only suggest cancel after minimum iterations (guard against no-tasks-created scenario)
+        reason += ` If all work is complete, run /oh-my-claudecode:cancel to cleanly exit ecomode and clean up state files. If cancel fails, retry with /oh-my-claudecode:cancel --force. Otherwise, continue working.`;
+      } else {
+        // Early iterations with no tasks yet - just tell LLM to continue
+        reason += ` Continue working - create Tasks to track your progress.`;
       }
 
-      console.log(JSON.stringify({
-        decision: 'block',
-        reason: reason
-      }));
+      console.log(JSON.stringify({ decision: 'block', reason }));
       return;
     }
 

package/scripts/test-mutual-exclusion.ts

@@ -7,7 +7,7 @@ import { mkdirSync } from 'fs';
 
 // Import the hooks
 import { startUltraQA, clearUltraQAState, isRalphLoopActive } from '../src/hooks/ultraqa/index.js';
-import { createRalphLoopHook, clearRalphState, isUltraQAActive } from '../src/hooks/ralph-loop/index.js';
+import { createRalphLoopHook, clearRalphState, isUltraQAActive } from '../src/hooks/ralph/index.js';
 
 // Test utilities
 function printTest(testName: string, passed: boolean) {
@@ -36,7 +36,7 @@ async function runTests() {
     const ralphStarted = ralphHook.startLoop(
       'test-session-1',
       'test task',
-      { maxIterations: 5, completionPromise: 'TASK_COMPLETE' }
+      { maxIterations: 5 }
     );
 
     if (!ralphStarted) {
@@ -86,7 +86,7 @@ async function runTests() {
     const ralphStarted2 = ralphHook2.startLoop(
       'test-session-4',
       'test task',
-      { maxIterations: 5, completionPromise: 'TASK_COMPLETE' }
+      { maxIterations: 5 }
     );
 
     if (ralphStarted2) {

package/skills/AGENTS.md

@@ -1,9 +1,9 @@
 <!-- Parent: ../AGENTS.md -->
-<!-- Generated: 2026-01-28 | Updated: 2026-01-28 -->
+<!-- Generated: 2026-01-28 | Updated: 2026-01-31 -->
 
 # skills
 
-32 skill definitions for workflow automation and specialized behaviors.
+37 skill definitions for workflow automation and specialized behaviors.
 
 ## Purpose
 
@@ -15,62 +15,71 @@ Skills are reusable workflow templates that can be invoked via `/oh-my-claudecod
 ## Key Files
 
 ### Execution Mode Skills
+
 | File | Skill | Purpose |
-|------|-------|---------|
-| `autopilot.md` | autopilot | Full autonomous execution from idea to working code |
-| `ultrawork.md` | ultrawork | Maximum parallel agent execution |
-| `ralph.md` | ralph | Persistence until verified complete |
-| `ultrapilot.md` | ultrapilot | Parallel autopilot with file ownership |
-| `swarm.md` | swarm | N coordinated agents with task claiming |
-| `pipeline.md` | pipeline | Sequential agent chaining |
-| `ecomode.md` | ecomode | Token-efficient parallel execution |
-| `ultraqa.md` | ultraqa | QA cycling until goal met |
+|-----------|-------|---------|
+| `autopilot/SKILL.md` | autopilot | Full autonomous execution from idea to working code |
+| `ultrawork/SKILL.md` | ultrawork | Maximum parallel agent execution |
+| `ralph/SKILL.md` | ralph | Persistence until verified complete |
+| `ultrapilot/SKILL.md` | ultrapilot | Parallel autopilot with file ownership |
+| `swarm/SKILL.md` | swarm | N coordinated agents with task claiming |
+| `pipeline/SKILL.md` | pipeline | Sequential agent chaining |
+| `ecomode/SKILL.md` | ecomode | Token-efficient parallel execution |
+| `ultraqa/SKILL.md` | ultraqa | QA cycling until goal met |
 
 ### Planning Skills
+
 | File | Skill | Purpose |
-|------|-------|---------|
-| `plan.md` | plan | Strategic planning with interview workflow |
-| `ralplan.md` | ralplan | Iterative planning (Planner+Architect+Critic) |
-| `review.md` | review | Review plan with Critic |
-| `analyze.md` | analyze | Deep analysis and investigation |
-| `ralph-init.md` | ralph-init | Initialize PRD for structured ralph |
+|-----------|-------|---------|
+| `plan/SKILL.md` | plan | Strategic planning with interview workflow |
+| `ralplan/SKILL.md` | ralplan | Iterative planning (Planner+Architect+Critic) |
+| `review/SKILL.md` | review | Review plan with Critic |
+| `analyze/SKILL.md` | analyze | Deep analysis and investigation |
+| `ralph-init/SKILL.md` | ralph-init | Initialize PRD for structured ralph |
 
 ### Code Quality Skills
+
 | File | Skill | Purpose |
-|------|-------|---------|
-| `code-review.md` | code-review | Comprehensive code review |
-| `security-review.md` | security-review | Security vulnerability detection |
-| `tdd.md` | tdd | Test-driven development workflow |
-| `build-fix.md` | build-fix | Fix build and TypeScript errors |
+|-----------|-------|---------|
+| `code-review/SKILL.md` | code-review | Comprehensive code review |
+| `security-review/SKILL.md` | security-review | Security vulnerability detection |
+| `tdd/SKILL.md` | tdd | Test-driven development workflow |
+| `build-fix/SKILL.md` | build-fix | Fix build and TypeScript errors |
 
 ### Exploration Skills
+
 | File | Skill | Purpose |
-|------|-------|---------|
-| `deepsearch.md` | deepsearch | Thorough codebase search |
-| `deepinit.md` | deepinit | Generate hierarchical AGENTS.md |
-| `research.md` | research | Parallel scientist orchestration |
+|-----------|-------|---------|
+| `deepsearch/SKILL.md` | deepsearch | Thorough codebase search |
+| `deepinit/SKILL.md` | deepinit | Generate hierarchical AGENTS.md |
+| `research/SKILL.md` | research | Parallel scientist orchestration |
 
 ### Utility Skills
+
 | File | Skill | Purpose |
-|------|-------|---------|
-| `learner.md` | learner | Extract reusable skill from session |
-| `note.md` | note | Save notes for compaction resilience |
-| `cancel.md` | cancel | Cancel any active OMC mode |
-| `hud.md` | hud | Configure HUD display |
-| `doctor.md` | doctor | Diagnose installation issues |
-| `omc-setup.md` | omc-setup | One-time setup wizard |
-| `mcp-setup.md` | mcp-setup | Configure MCP servers |
-| `help.md` | help | Usage guide |
+|-----------|-------|---------|
+| `orchestrate/SKILL.md` | orchestrate | Core multi-agent orchestration (always active) |
+| `learner/SKILL.md` | learner | Extract reusable skill from session |
+| `note/SKILL.md` | note | Save notes for compaction resilience |
+| `cancel/SKILL.md` | cancel | Cancel any active OMC mode |
+| `hud/SKILL.md` | hud | Configure HUD display |
+| `doctor/SKILL.md` | doctor | Diagnose installation issues |
+| `omc-setup/SKILL.md` | omc-setup | One-time setup wizard |
+| `mcp-setup/SKILL.md` | mcp-setup | Configure MCP servers |
+| `help/SKILL.md` | help | Usage guide |
+| `learn-about-omc/SKILL.md` | learn-about-omc | Usage pattern analysis |
+| `skill/SKILL.md` | skill | Manage local skills |
 
 ### Domain Skills
+
 | File | Skill | Purpose |
-|------|-------|---------|
-| `frontend-ui-ux.md` | frontend-ui-ux | Designer-developer aesthetic |
-| `git-master.md` | git-master | Git expertise, atomic commits |
-| `project-session-manager.md` | psm | Isolated dev environments |
-| `writer-memory.md` | writer-memory | Agentic memory for writers |
-| `release.md` | release | Automated release workflow |
-| `local-skills-setup.md` | local-skills-setup | Manage local skills |
+|-----------|-------|---------|
+| `frontend-ui-ux/SKILL.md` | frontend-ui-ux | Designer-developer aesthetic |
+| `git-master/SKILL.md` | git-master | Git expertise, atomic commits |
+| `project-session-manager/SKILL.md` | psm | Isolated dev environments |
+| `writer-memory/SKILL.md` | writer-memory | Agentic memory for writers |
+| `release/SKILL.md` | release | Automated release workflow |
+| `local-skills-setup/SKILL.md` | local-skills-setup | Manage local skills |
 
 ## For AI Agents
 
@@ -121,10 +130,13 @@ Any configurable options.
 
 #### Creating a New Skill
 
-1. Create `new-skill.md` with YAML frontmatter
+1. Create `new-skill/SKILL.md` directory and file with YAML frontmatter
 2. Define purpose, workflow, and usage
 3. Add to skill registry (auto-detected from frontmatter)
 4. Optionally add activation triggers
+5. Create corresponding `commands/new-skill.md` file (mirror)
+6. Update `docs/REFERENCE.md` (Skills section, count)
+7. If execution mode skill, also create `src/hooks/new-skill/` hook
 
 ### Common Patterns
 
@@ -148,7 +160,10 @@ Any configurable options.
 
 ### Testing Requirements
 
-Skills are tested via integration tests that invoke skills and verify behavior.
+- Skills are verified via integration tests
+- Test skill invocation with `/oh-my-claudecode:skill-name`
+- Verify trigger keywords activate correct skill
+- For git-related skills, follow `templates/rules/git-workflow.md`
 
 ## Dependencies
 

package/skills/autopilot/SKILL.md

@@ -161,8 +161,6 @@ rm -f .omc/state/autopilot-state.json
 rm -f .omc/state/ralph-state.json
 rm -f .omc/state/ultrawork-state.json
 rm -f .omc/state/ultraqa-state.json
-rm -f ~/.claude/ralph-state.json
-rm -f ~/.claude/ultrawork-state.json
 ```
 
 This ensures clean state for future sessions.