Skip to content

ai-pilot.md

Latest commit

 

History

History
121 lines (99 loc) · 5.07 KB

ai-pilot.md

File metadata and controls

121 lines (99 loc) · 5.07 KB

fuse-ai-pilot

APEX workflow orchestrator with sniper validation and research capabilities.

Agents

Agent Description
sniper 7-phase validation (DRY detection), zero linter errors
sniper-faster Quick validation, minimal output
explore-codebase Architecture discovery
research-expert Documentation with Context7/Exa
websearch Quick web research
seo-expert SEO/SEA/GEO optimization

Commands

Command Description
/apex Full APEX workflow
/apex-quick Skip Analyze, direct Execute
/research Technical research
/exploration Codebase discovery
/code-quality Linters, SOLID validation, DRY detection
/elicitation Self-review techniques

Skills

  • apex - Full APEX methodology
  • apex-quick - Quick flow
  • research - Research methodology
  • exploration - Discovery techniques
  • code-quality - Validation with DRY detection (jscpd)
  • elicitation - Self-review (75 techniques)
  • skill-creator - Create/restructure skills with SKILL.md + references/
  • agent-creator - Create expert agents with frontmatter, hooks, skills
  • react-effects-audit - Audit React useEffect anti-patterns (9 rules from "You Might Not Need an Effect")

Cache System (fusengine-cache)

4-level persistent cache to reduce redundant operations and save tokens (60-75% savings).

~/.claude/fusengine-cache/
├── explore/{project-hash}/    # Architecture snapshots
│   ├── metadata.json
│   └── snapshot.md
├── doc/{project-hash}/        # Documentation cache
│   ├── index.json
│   └── docs/{doc-hash}.md
├── lessons/{project-hash}/    # Sniper error patterns
│   └── {timestamp}.json
├── tests/{project-hash}/      # Test results cache
│   └── results.json
└── analytics/
    └── sessions.jsonl         # Cache hit/miss tracking

Cache Levels

Cache Source TTL Injection Savings
Explore explore-codebase agent 24h SubagentStart → explore-codebase ~85%
Documentation Context7/Exa synthesis 7d SubagentStart → research-expert ~90%
Lessons Sniper Edit corrections 30d SubagentStart → all agents ~50-70%
Tests Sniper test results 48h SubagentStart → sniper ~60%

Cache Scripts (TypeScript/Bun)

All scripts are in scripts/ with shared lib/ modules.

Script Hook Role
explore-cache-check.ts SubagentStart Inject cached architecture for explore-codebase
doc-cache-inject.ts SubagentStart Inject cached doc summaries for research-expert
cache-doc-from-transcript.ts SubagentStop Extract synthesis from research-expert transcript
lessons-cache-inject.ts SubagentStart Inject known error patterns for all agents
cache-sniper-lessons.ts SubagentStop Extract Edit corrections from sniper transcript
promote-global-lessons.ts Background Promote lessons seen 3+ times to _global/
test-cache-inject.ts SubagentStart Inject previous test results for sniper
cache-test-results.ts SubagentStop Save sniper test results with file hashes
cache-analytics-save.ts SessionEnd Save cache hit/miss analytics
inject-subagent-context.ts SubagentStart Inject general context to all subagents
inject-apex-context.sh PreToolUse Inject APEX context for Task tool
enforce-apex-phases.ts PreToolUse Enforce APEX phase ordering
detect-and-inject-apex.ts UserPromptSubmit Auto-detect APEX triggers
check-solid-compliance.sh PostToolUse SOLID validation on Write/Edit
check-solid-from-transcript.sh SubagentStop SOLID check from agent transcript
track-doc-consultation.sh PostToolUse Track documentation reads
sync-task-tracking.ts PostToolUse Sync TaskCreate/TaskUpdate

Lessons Format (per-timestamp)

Each sniper run creates a {timestamp}.json with Edit-extracted corrections:

{
  "project": "/path/to/project",
  "timestamp": "2026-02-09T01:14:44",
  "errors": [
    {"error_type":"missing_directive","pattern":"Component using hooks without 'use client'","fix":"Fix missing_directive in Dashboard.tsx","count":1,"files":["/path/Dashboard.tsx"],"code":{"line":["'use client'","","import React from 'react'"]}}
  ]
}

Hooks (14 entries)

Hook Type Count Scripts
UserPromptSubmit 1 detect-and-inject-apex
SubagentStart 5 inject-subagent-context, explore-cache-check, doc-cache-inject, lessons-cache-inject, test-cache-inject
PreToolUse 2 enforce-apex-phases, inject-apex-context
SubagentStop 4 cache-sniper-lessons, cache-test-results, cache-doc-from-transcript, check-solid-from-transcript
PostToolUse 3 check-solid-compliance, track-doc-consultation, sync-task-tracking
SessionEnd 1 cache-analytics-save

MCP Servers

  • Context7 (documentation)
  • Exa (web search, code context)
  • Sequential Thinking (complex reasoning)