guard-skills

# Agent Memory - Persistent Project Knowledge

> **Last Updated**: 2026-08-22 | **Session**: memory-improvement
> **Project**: Guard Skills | **Location**: C:\opencodes\guard skills
> **Auto-Sync**: Auto-generated by tools/build_index.py (225 skills)

---

## 0. TASK FOLDER RULE — MANDATORY (added 2026-08-24)

Follow on EVERY task, no exceptions:

1. **At task START**: analyze the task, choose a short descriptive folder name
   (kebab-case, English, max 4 words, e.g. `autocad-2024-install`, `plc-ladder-fix`).
2. **Create**: `H:\ai\<task-name>\` — if it exists, append `-2`, `-3`, ...
3. **Structure inside**:
   - `task.md` — what the task is, plan, progress updates
   - `outputs\` — every produced file (scripts, reports, configs)
   - `logs\` — command outputs, errors, diagnostics worth keeping
4. **Work FROM that folder**: temp files and intermediate results go there,
   not random temp locations.
5. **At task END**: update `task.md` with final summary (done, files, problems, next steps).

Never start a task without its folder. Never leave task data outside it.
First applied example: `H:\ai\autocad-2024-install\` (real data from the AutoCAD session).

---

## 1. Project Overview

| Field | Value |
|-------|-------|
| Name | Guard Skills |
| Type | Skills repository for coding agents |
| Total Skills | 220 (72 guards, 82 tools, 63 workflow + 3 special) |
| Root | `C:\opencodes\guard skills` |
| Skills Dir | `C:\opencodes\guard skills\skills\` |
| Last Verified | 2026-08-22 18:02 UTC+1 — filesystem counts match header |

## 2. User Profile

| Preference | Value |
|------------|-------|
| Language | English |
| Communication | Direct, concise |
| Style | Action over explanation |
| Browser | Brave (not Chrome) |
| Desktop MCP | Ghost (`ghost do "..."`) |
| Model (daily) | `opencode/deepseek-v4-flash-free` |
| Model (cheap) | `opencode/nemotron-3-ultra-free` |
| Model (vision) | `openrouter/inclusionai/ling-3.0-flash-vl:free` (probe-verified 2026-09-20; nemotron-nano-12b-v2-vl:free RETIRED) |
| Task endings (2026-09-14, standing rule) | ALWAYS end every task with 10+ recommended follow-up tasks in the same context |

## 3. All Projects (C:\opencodes) — Verified 2026-08-22

### Main Projects
| Project | Files | Description |
|---------|-------|-------------|
| guard skills | 52053* | Guard skills for coding agents (*incl. node_modules/cache) |
| automation | 7525 | OpenCode Automation Suite - OCR, PDF, folder watching, screenshots |
| instrumentation-list | 13910 | Analysis tooling for 19094-02 Instrumentation List |
| office expert | 3534 | Office expert tools |
| agents | 102 | 100 agents auto-mode orchestrators |
| OPENCODE IMPROVEMENT | 415 | Self-learning stack that persists state |
| tools | 294 | Skeleton Tool Generator - 80 skeleton tools (PS1) + BAT wrappers |
| tests | 156 | Test files |
| reports | 137 | Report files |
| scripts | 108 | Script files |

### PC/Diagnostic
| Project | Files | Description |
|---------|-------|-------------|
| PC DIAGNOSTIQUE | 3735* | 100-Point Windows Optimization Suite for Windows 11 (*was 65, growth 2026-07-28) |
| plc | 15429* | PLC tools (ABB conductivity flowcell simulator) (*incl. deps) |

### EPLAN/Engineering
| Project | Files | Description |
|---------|-------|-------------|
| Eplan | 9 | VEGAMET 141 EPLAN Symbol Package |
| Eplan_Symbol | 11 | EPLAN symbols |
| disgnodtique-autocad | 20 | Automotive Electrical Diagnostic System |

### Knowledge/Base
| Project | Files | Description |
|---------|-------|-------------|
| knowledge-base | 15 | Read-only documentation snapshots |
| problems | 22 | Problem-Solution Knowledge Base |
| improvement_report | 16 | Improvement reports |

### Life/Planning
| Project | Files | Description |
|---------|-------|-------------|
| LIFE-OS | 41 | Complete life planning system - 100% local, Markdown + PowerShell |

### OCR/Processing
| Project | Files | Description |
|---------|-------|-------------|
| OCR | 13 | OCR tools |
| opencode-custom | - | Custom opencode configuration |

### Other
| Project | Files | Description |
|---------|-------|-------------|
| opencodes | 47 | Main opencodes repository |
| opencodes-assistant-0.2.0 | 589 | OpenCode assistant package |
| ErrorReports | 3 | Error report files |
| office | 22 | Office files |
| plugins | 3 | Plugin files |
| shared-skills | 17 | Shared skill definitions |
| skills | 15 | Additional skills |

### Empty/Inactive
- KB-Invoice-Generator, KB-Invoice-Generator-New, openmythos, aguseron, aigent

> **Note**: File counts with * include dependencies (node_modules, __pycache__, .git). Clean counts available via `Get-ChildItem -Recurse -File | Where-Object { $_.FullName -notmatch "node_modules|__pycache__|\.git" }`

---

## 4. Skill Index (220 Skills - Auto-generated 2026-09-20)

> AUTO-GENERATED by tools/build_index.py - do not hand-edit. Run: python tools/build_index.py

### Guards (72) - Quality Gates & Validation
| Skill | Purpose |
|-------|---------|
| analytics-guard | Analytics implementation validation, event tracking, and data privacy |
| api-contract-guard | API contract compliance validation between providers and consumers |
| api-security-guard | API security validation - auth, headers, CORS, input validation, error handling, logging |
| async-guard | Async/await pattern validation - proper usage, error handling, cancellation |
| build-guard | Build process validation and compilation safety checks |
| changelog-guard | Changelog format validation and completeness checking |
| ci-guard | CI/CD pipeline configuration validation and optimization |
| circuit-breaker-guard | Circuit breaker pattern validation - thresholds, recovery, fallbacks, monitoring |
| clean-arch-guard | Clean architecture layer dependency validation - entities, use cases, interfaces, infrastructure |
| clean-code-guard | ULTRA-ADVANCED Clean Code Guard with automated code analysis, security scanning, |
| comment-guard | Comment quality validation - meaningful vs noise, stale comments, missing docs |
| commit-guard | Commit message convention validation and changelog generation |
| concurrency-guard | Concurrency/threading safety validation and deadlock detection |
| config-guard | Configuration file validation and best practices |
| cors-guard | CORS configuration validation and security best practices |
| cqrs-guard | CQRS pattern validation - command/query separation, read/write models, separate stores |
| csrf-guard | CSRF protection validation - anti-forgery tokens, same-site cookies, custom headers |
| css-guard | CSS/Sass best practices, specificity management, and performance |
| cyclomatic-guard | Cyclomatic complexity validation - McCabe metrics, refactoring targets |
| ddd-guard | Domain-driven design validation - aggregates, entities, bounded contexts, repositories |
| dead-code-guard | Dead code detection - unused imports, functions, variables, unreachable code |
| dependency-injection-guard | DI/IoC validation - constructor injection, service locator, container setup, captive dependencies |
| deploy-guard | Deployment safety validation and rollback verification |
| docker-compose-guard | Docker Compose validation, service configuration, and networking checks |
| docker-guard | Dockerfile optimization, security best practices, and multi-stage builds |
| docs-guard | ULTRA-ADVANCED Docs Guard with automated documentation analysis, API verification, |
| duplicate-code-guard | Code duplication detection - exact, near-duplicate, DRY principle enforcement |
| encryption-guard | Encryption best practices - algorithm strength, key management, hashing, IV handling |
| env-check-guard | Environment variable validation and security checking |
| error-handling-guard | Error handling patterns and exception safety validation |
| event-driven-guard | Event-driven architecture validation - schema, channels, handlers, idempotency, ordering |
| event-sourcing-guard | Event sourcing pattern validation - event streams, aggregates, snapshots, replay |
| file-guard | File system security, path traversal detection, and permission validation |
| graceful-shutdown-guard | Graceful shutdown validation - signal handling, drain, cleanup, health check on stop |
| health-check-guard | Health check endpoint validation - liveness, readiness, startup probes, dependencies |
| helm-guard | Helm chart validation, template checking, and best practices |
| hexagonal-guard | Hexagonal/ports-and-adapters architecture validation - clean dependency flow |
| html-guard | HTML semantics, accessibility compliance, and best practices |
| iam-guard | IAM policy validation - least privilege, resource constraints, condition blocks |
| idor-guard | Insecure Direct Object Reference detection - authorization checks, predictable IDs, mass assignment |
| injection-guard | Injection prevention validation - OS command, LDAP, XPath, NoSQL, template injection |
| json-guard | JSON schema validation, structure checking, and best practices |
| jwt-guard | JWT implementation validation - token handling, algorithms, security patterns |
| kubernetes-guard | Kubernetes manifest validation, security contexts, and resource limits |
| large-class-guard | Large class detection - single responsibility, cohesion, coupling analysis |
| logging-pattern-guard | Logging patterns - structure, levels, correlation, sensitive data, sampling |
| long-method-guard | Long method detection - extraction opportunities, parameter count, nesting depth |
| lsp-guard | Language Server Protocol auto-detection validation, |
| markdown-guard | Markdown quality, consistency, and formatting checks |
| message-queue-guard | Message queue validation - producers, consumers, reliability, dead letter queues |
| migration-pattern-guard | Database migration validation - versioning, rollback, backward compat, expand-contract |
| naming-guard | Naming convention validation - consistency, readability, anti-patterns |
| nested-depth-guard | Nesting depth validation - conditionals, loops, callback nesting, refactoring targets |
| observer-pattern-guard | Observer/event pattern - subscription, notification, unsubscription, memory leaks |
| pr-guard | Pull request template validation, size checking, and review readiness |
| python-guard | Python best practices, PEP 8 validation, and idiomatic patterns |
| react-guard | React best practices, hooks rules, and component patterns |
| readme-guard | README completeness, installation instructions, and documentation quality |
| repository-pattern-guard | Repository pattern validation - abstraction, interface design, query methods |
| responsive-guard | Responsive design validation, breakpoint checking, and mobile compatibility |
| retry-guard | Retry pattern validation - exponential backoff, jitter, max retries, circuit breaker integration |
| saga-guard | Saga/choreography pattern validation - compensating transactions, step ordering, orchestration |
| seo-guard | SEO validation, meta tags, structured data, and accessibility impact |
| serialization-guard | Serialization/deserialization safety and injection prevention |
| ssrf-guard | Server-side request forgery prevention - URL validation, IP blocking, scheme restrictions |
| state-guard | State management validation and anti-pattern detection |
| strategy-pattern-guard | Strategy pattern validation - interchangeable algorithms, interface, context, registry |
| terraform-guard | Terraform configuration validation, security scanning, and best practices |
| test-guard | ULTRA-ADVANCED Test Guard with automated test analysis, coverage analysis, |
| typescript-guard | TypeScript best practices, strict mode validation, and type safety |
| xss-guard | XSS prevention validation - output encoding, CSP headers, dangerous API detection |
| yaml-guard | YAML validation, formatting consistency, and security checks |

### Tools (82) - Infrastructure & Utilities
| Skill | Purpose |
|-------|---------|
| accessibility-patterns | Accessibility patterns - ARIA, keyboard navigation, screen reader, color contrast |
| alerting-stack | Alerting stack - rules, routing, escalation, silencing, runbooks |
| analytics | Analytics implementation, event tracking, dashboards, and user behavior |
| analytics-setup | Analytics implementation - event tracking, taxonomy, dashboards, privacy |
| api-documentation | API documentation - OpenAPI/Swagger spec, docs generation, SDK generation, examples |
| api-rate-limiting | API rate limiting - algorithm selection, headers, throttling, quota management |
| api-versioning | API versioning strategy - URI, header, content negotiation, deprecation |
| audio-processing | Audio processing, speech recognition, and audio transformation |
| backup-strategy | Backup strategy - frequency, retention, offsite, encryption, recovery testing |
| batch-processing | Batch processing workflows, job scheduling, and parallel execution |
| cache-strategy | Cache strategy - TTL, invalidation, warming, hierarchy, distributed patterns |
| capacity-management | Capacity management - forecasting, monitoring, scaling triggers, planning |
| cdn | CDN configuration, edge caching, and content delivery optimization |
| cdn-strategy | CDN strategy - edge caching, origin shield, dynamic acceleration, compression |
| cicd-pipeline | CI/CD pipeline design - stages, gates, caching, optimization, security scanning |
| component-library | Component library management, versioning, and distribution |
| computer-vision | Computer vision, object detection, and image recognition |
| container-patterns | Container/Docker patterns - image optimization, multi-stage, security, layering |
| content-strategy | Content strategy, editorial planning, and content lifecycle management |
| contract-testing | Contract testing - consumer-driven contracts, provider verification, Pact |
| data-management | Data management - lifecycle, classification, governance, quality, lineage, retention |
| data-quality | Data quality framework - validation, monitoring, lineage, remediation, ownership |
| data-warehouse | Data warehouse design, star schema, and query optimization |
| database-optimization | Database optimization - query tuning, indexing, connection management, monitoring |
| database-scaling | Database scaling strategy - read replicas, sharding, partitioning, connection pooling |
| design-system | Design system creation - components, design tokens, documentation, distribution |
| ecommerce | E-commerce platform design, product catalog, cart, and checkout flow |
| email-systems | Email service design, transactional emails, and deliverability |
| error-boundary-patterns | Error boundary patterns - React error boundaries, graceful degradation, recovery |
| etl | ETL pipeline design, data transformation, and workflow orchestration |
| feature-flag-strategy | Feature flag strategy - toggle types, release management, cleanup, monitoring |
| freebuff-bridge | ULTRA-ADVANCED Freebuff Bridge — Communicate with Freebuff Desktop AI assistant. |
| full-text-search | Full-text search implementation, tokenization, and ranking algorithms |
| graph-database | Graph database design, query patterns, and relationship modeling |
| grpc | GRPC API design, protobuf definition, and bidirectional streaming |
| image-optimization | Image optimization - format selection, compression, responsive images, lazy loading |
| image-processing | Image processing pipeline, optimization, and transformation |
| incident-management | Incident management - detection, triage, severity, communication, escalation |
| index-design | Index strategy - B-tree, hash, composite, covering, partial, filtered indexes |
| information-architecture | Information architecture design, navigation patterns, and content organization |
| input-blocker | Block keyboard and mouse input on Windows until manually unblocked. |
| logging-patterns | Logging patterns - structured, levels, correlation ID, sampling, context |
| logging-stack | Logging stack design - collection, aggregation, storage, analysis, retention |
| makefile | Makefile best practices - phony targets, variables, dependencies, help |
| mfa | Multi-factor authentication design, TOTP, biometrics, and security keys |
| microservice-patterns | Microservice patterns - bounded context, API gateway, service discovery, database per service |
| monitoring-stack | Monitoring stack design - metrics collection, storage, visualization, alerting |
| nlp | Natural language processing, text analysis, and language models |
| nosql | NoSQL database design, document stores, and key-value patterns |
| notification | Notification system design, push/email/SMS, and delivery preferences |
| oauth | OAuth 2.0 implementation, authorization flows, and token management |
| payment | Payment integration, gateway selection, and PCI compliance |
| performance-monitoring | Performance monitoring setup, RUM, synthetic monitoring, and alerting |
| performance-optimization | Performance optimization - Core Web Vitals, loading, rendering, bundles |
| protocol-buffers | Protocol Buffers design - field numbering, versioning, backward compat, organization |
| pwa | Progressive web app development, service workers, and offline-first architecture |
| query-optimization | Query optimization - execution plans, rewriting, statistics, indexing hints |
| rbac | RBAC/ABAC design, permission modeling, and access control implementation |
| real-time | Real-time system design with WebSockets, SSE, and event-driven architecture |
| responsive-design | Responsive design patterns - breakpoints, fluid layouts, mobile-first, testing |
| rest-api | REST API design, resource modeling, and HATEOAS principles |
| scrcpy-install | ULTRA-ADVANCED Scrcpy Install & Usage — Android screen mirroring via ADB. |
| search-engine | Search engine integration, indexing strategy, and relevance tuning |
| seo-optimization | SEO optimization - technical SEO, content, structured data, performance, meta tags |
| serverless-patterns | Serverless patterns - function design, cold start, event sources, error handling |
| service-mesh | Service mesh - sidecar, mTLS, traffic routing, observability, security policies |
| shell-scripting | Shell scripting best practices - error handling, portability, security, quoting |
| skill-generator-tool | Creates, improves, and manages AI coding agent skill files at scale. Commands: list, create, improve |
| sso | Single sign-on implementation, SAML/OIDC integration, and federation |
| state-management | State management patterns - Redux, Context, signals, proper immutability |
| static-sites | Static site generation, SSG frameworks, and build optimization |
| streaming | Streaming data processing with Kafka, Flink, and event pipelines |
| terabox-integration | Cloud storage operations via browser automation and API |
| testing-pyramid | Testing pyramid strategy - unit, integration, e2e distribution, coverage goals |
| time-series | Time series data management, forecasting, and anomaly detection |
| tracing-stack | Distributed tracing - OpenTelemetry, instrumentation, sampling, correlation |
| vector-database | Vector database integration, embedding strategies, and similarity search |
| video-processing | Video processing, transcoding, streaming, and optimization. |
| webhooks | Webhook management, delivery guarantees, and payload verification |
| websockets | WebSocket management, real-time communication, and connection handling |
| windows-shortcut-repair | ULTRA-ADVANCED Windows Shortcut Repair — GUI + CLI with monitoring, export, backup, and scheduled ta |
| zero-downtime-deployment | Zero-downtime deployment - rolling, blue-green, canary, migration patterns |

### Workflow (63) - Process & Methodology
| Skill | Purpose |
|-------|---------|
| a-b-testing | A/B testing design, statistical significance, and experiment analysis |
| acceptance-testing | Acceptance testing workflow - criteria definition, UAT, automation, sign-off process |
| api-design-workflow | API design workflow - contract-first, resource modeling, versioning, documentation |
| architecture-decision-record | ADR creation, review, and management workflow |
| atdd | Acceptance Test Driven Development - collaboration, automation, executable specifications |
| backlog-grooming | Backlog refinement, prioritization frameworks, and user story mapping |
| bdd-workflow | BDD workflow - Gherkin feature files, scenario structure, step definitions |
| changelog | Changelog management workflow with automated generation and release notes |
| chaos-engineering-workflow | Chaos engineering - experiments, steady-state hypothesis, blast radius control |
| code-migration | Code migration workflow with compatibility analysis and rollback planning |
| code-review-workflow | Code review process - review types, checklists, feedback patterns, turnaround time |
| compliance | Compliance validation workflow, regulatory checking, and audit preparation |
| continuous-improvement | Continuous improvement workflow - metrics, experiments, kaizen culture, feedback loops |
| cost-optimization | Cost optimization analysis, resource right-sizing, and waste reduction |
| data-migration | Data migration workflow with validation, rollback, and zero-downtime strategies |
| data-pipeline-workflow | Data pipeline workflow - ingestion, transformation, validation, monitoring, lineage |
| database-design-workflow | Database design workflow - schema modeling, normalization, indexing strategy |
| database-migration-workflow | Database migration workflow - expand-migrate-contract, rollback, zero-downtime strategies |
| decision-log | Decision log - ADR format, rationale, alternatives, impact, review |
| definition-of-done | Definition of Done - quality gates, verification, team agreement, enforcement |
| dependency-update | Dependency update workflow, breaking change analysis, and automation |
| diagnose | ULTRA-ADVANCED Diagnose skill with systematic debugging, root cause analysis, |
| disaster-recovery | Disaster recovery planning, backup strategy, and recovery testing |
| documentation-as-code | Documentation as code workflow with automated building and deployment |
| estimation | Story/task estimation techniques, planning poker, and effort sizing |
| event-storming | Event storming workshop - domain exploration, bounded context discovery, timeline ordering |
| example-mapping | Example mapping workshop - rules, examples, questions, acceptance criteria definition |
| exploratory-testing | Exploratory testing - charters, session notes, timeboxed discovery, debrief |
| grill-me | ULTRA-ADVANCED Grill Me skill with systematic questioning, decision tree analysis, |
| handoff | ULTRA-ADVANCED Handoff skill with comprehensive context preservation, |
| impact-mapping | Impact mapping - strategic planning, feature prioritization, assumption validation |
| incident-postmortem | Blameless post-mortem facilitation, root cause analysis, and action tracking |
| incident-response-workflow | Incident response workflow - detection, triage, mitigation, post-mortem, action items |
| load-testing | Load testing strategy, tool selection, and performance benchmarking |
| localization | Localization workflow, translation management, and i18n integration |
| ml-pipeline-workflow | ML pipeline workflow - data prep, training, evaluation, deployment, monitoring |
| pair-programming | Pair programming workflow - driver/navigator roles, rotation timing, collaboration patterns |
| pdf-ocr | ULTRA-ADVANCED PDF OCR Skill — Content stream scaling approach that produces |
| performance-audit | Performance audit - profiling, baseline, bottlenecks, load testing, regression |
| performance-budget | Performance budget creation, monitoring, and enforcement |
| performance-testing-workflow | Performance testing workflow - plan, script, execute, analyze, baseline comparison |
| refactoring-workflow | Refactoring workflow - code smells, safe steps, test coverage, incremental improvement |
| release-notes | Release notes creation with automated summaries and stakeholder communication |
| requirements-gathering | Requirements gathering - stakeholder interviews, user stories, non-functional, prioritization |
| retrospective | Sprint retrospective facilitation with multiple formats and action tracking |
| runbook | Runbook creation, incident response procedures, and automation |
| secrets-management | Secrets management workflow, rotation strategies, and vault integration |
| secrets-rotation-workflow | Secrets rotation workflow - lifecycle management, automation, emergency procedures |
| security-review | Security review workflow, threat modeling, and vulnerability assessment |
| security-testing-workflow | Security testing workflow - SAST, DAST, penetration testing, dependency scanning |
| semantic-versioning | Semantic versioning workflow with breaking change detection and automation |
| specification-by-example | Specification by Example - living documentation, executable specs, collaborative refinement |
| sprint-planning | Sprint planning workflow with capacity-based commitment and task breakdown |
| story-splitting | User story splitting - vertical slicing, SPIDR, independent value, INVEST |
| tdd | ULTRA-ADVANCED Test-Driven Development with systematic red-green-refactor, |
| technical-debt | Technical debt identification, quantification, and management strategy |
| technical-specification | Technical specification - architecture decisions, API contracts, data model, security review |
| terabox-improvement | Analyze and optimize TeraBox integration workflows |
| usability-testing | Usability testing workflow, test script creation, and results analysis |
| user-research | User research planning, interview techniques, and synthesis methods |
| user-story-mapping | User story mapping - backlog organization, release slicing, persona workflow discovery |
| value-stream-mapping | Value stream mapping - process optimization, waste identification, cycle time analysis |
| zoom-out | ULTRA-ADVANCED Zoom Out skill with comprehensive architecture analysis, |

### Special (3) - Top-level cross-cutting skills
| Skill | Purpose |
|-------|---------|
| desktop-control-mcp | Enhanced Desktop Control MCP Server for advanced Windows PC control with AI/ML integration, enterpri |
| force-delete | Automates the removal of protected directories that throw **"Access is denied"** by: - Taking owners |
| self-learning | # Self-Learning Engine - ULTRA-ADVANCED v5.0 |


---

## 5. Session History

### Session 2026-06-30 (Memory Setup)
- Set up 3-layer memory system: AGENTS.md, HANDOFF.md, knowledge_base_v2.json
- Ran 13 improvement cycles on memory system
- **Outcome**: Memory system operational

### Session 2026-06-30 (OneDrive Download)
- Downloaded 8 folders from API-SOF-PRO SharePoint
- Downloaded bulk zip (~20GB) — corrupt, needs re-download
- Partial valid zips: FACTORY IO, SMICROSOFT, SOMOVE VFF SOFT
- Missing: ECOSTRUCTURE, SIMATIC STEP 7, STUDIO 5000 V33, TIA PORTAL V13, UNITY PRO V13
- GLM 5.2: PC cannot run (needs 241GB+ RAM, PC has 32GB)
- .NET Framework 2.0: Already installed
- **Bug**: write/edit tools had $ bug — workaround: bash commands (resolved later)

### Session 2026-07-01 (PLC Program Fix)
- Fixed Schneider SoMachine PLC program (`sdi program.smbp`)
- 2 manual fixes applied + 10 already correct in original
- **Fix 1**: SetCoil → Coil for %M1 (RINSE), %M2 (FILL), %M3 (BYPASS)
- **Fix 10**: SetCoil → Coil + S → ST for valves %M300-%M303, %M351
- Full PLC memory map documented (see Section 9)

### Session 2026-07-04 (PDF Report)
- Created PDF system report using fpdf2
- Pushed to GitHub: `Abdox-menara/opencode-reports`
- PowerShell commands for system data gathering documented

### Session 2026-07-05 (Memory Improvement)
- Rewrote AGENTS.md, HANDOFF.md, knowledge_base_v2.json
- Expanded skill index from 20 to 215+ skills
- Added structured session history
- **Outcome**: Memory system significantly improved

### Session 2026-07-07 (Autodesk Complete Removal)
- Removed all Autodesk products: AutoCAD 2024/2025/2026, Inventor 2027, DWG TrueView 2027, Autodesk Access, CER, AdODIS, AdskIdentityManager, Genuine Service, Material Library, App Manager, Featured Apps, REX Inventor, OpenStudio CLI, AutoCAD Open in Desktop, Save to Web and Mobile, Single Sign On Component
- Method: MsiExec /x (orphaned entries), manual registry cleanup, service removal, file deletion
- Services deleted: AdskLicensingService, Autodesk Access Service Host, Autodesk CER Service
- Registry cleaned: Uninstall keys (64/32-bit), Installer Products, HKLM\SOFTWARE\Autodesk, HKCU\Software\Autodesk
- Folders deleted: C:\Program Files\Autodesk, C:\Program Files (x86)\Autodesk, C:\ProgramData\Autodesk, AppData\Local\Autodesk, AppData\Roaming\Autodesk
- Fixed Windows Installer Error for AcademContentLP-CA (HKLM\SOFTWARE\Classes\Installer\Products\FEE98B82701900001620FCF3A3907BD7)
- **Outcome**: COMPLETED - No reboot required, all clean

### Session 2026-07-11 (Vercel Labs Research)
- Analyzed 318 Vercel Labs public repos
- Identified key architectural patterns: agent-first, skills as first-class, security-first
- Integrated findings into AGENTS.md and knowledge_base_v2.json
- Added 6 new patterns (p016-p020), 5 new learnings (l011-l015)
- **Outcome**: Knowledge base enhanced with external research

### Session 2026-07-13 (Ghost Snapshot Toolkit)
- Built `ghost_snapshot.py` v2.0 with 17 commands for desktop screenshot intelligence
- Fixed 1.333x DPI scale drift in both ghost_control.py and ghost_snapshot.py
- v1.0 (8 features): annotation, diff, OCR, multi-monitor, caching, element discovery, history
- v2.0 (9 new): UIA tree/find/click, EasyOCR/RapidOCR, template match, smart click, regions, watch, export
- Fixed: ctypes.wintypes import, pyautogui win32 GDI fallback, uia-find recursive search, smart-click UIA-first
- Created `ghost-snapshot` skill (SKILL.md) + PowerShell wrapper
- Updated Ghost memory.json to v6.0
- **Outcome**: All 17 commands tested and working

### Session 2026-07-15 (TeraBox Integration)
- Uploaded guard-skills.zip (171MB) to TeraBox cloud storage
- Created terabox-integration tool skill with browser automation
- Created terabox-improvement workflow skill for optimization
- Implemented Python scripts for API integration and analysis
- Updated AGENTS.md with new skills (217+ total)
- **Outcome**: TeraBox integration operational with improvement workflow

### Session 2026-07-28 (Ecosystem Cleanup — Major)
- **Scope**: C:\opencodes\, C:\opencodes\plc\, C:\cursor\, C:\Users\Abdox\.config\opencode\
- **Config fixes (5 files)**: Root `opencode.jsonc` (removed dead `INSTRUCTIONS.md` ref); `plc/opencode.json` (2 trailing commas → valid JSON); `Eplan/opencode.json` (added missing vision model); `C:\Users\Abdox\.config\opencode\opencode.jsonc` (2 trailing commas + broken script paths); `C:\cursor\opencode.jsonc` (3 trailing commas)
- **TS tool fixes (3 files)**: `test-runner.ts` (missing closing brace TS1005; removed dead `promisify(spawn)`); `secret-scanner.ts` (severity cascade TS7053 implicit any); `dependency-auditor.ts` (separated stdout/stderr, added descriptive error codes)
- **Python syntax fixes (3 files)**: `export_all_reports.py` (mismatched f-string parens); `plugin_hooks.py` (6 structural issues); `FastDownloader/_build_v5.py` (unescaped `'` breaking brackets)
- **NPM fix**: `brace-expansion` 5.0.6→5.0.8 — 0 vulnerabilities remaining
- **Windows compat**: replaced `rm -rf` with `if exist ... rd /s /q`; created `_audit.cjs` (ESM/CJS fix)
- **Filesystem cleanup (~400 items, ~75 MB)**: 287 Playwright dumps, 105 test_perms files, 7 stale HTML reports/scripts, 2 empty git repos, 2 dist/ dirs, 2 screenshots, 2 empty config subdirs, InputBlocker-v5.exe, 4 stale `.pyc`, 5 `index.lock` files
- **Git hygiene**: stale `index.lock` cleaned (LIFE-OS, OCR, office expert, opencodes, PC DIAGNOSTIQUE); nested `.git` inside `guard skills\.git\` identified (NTFS-locked, harmless ~1 MB)
- **.gitignore upgrades**: 20 sub-projects + `guard skills` updated (node_modules/, __pycache__/, dist/, *.pyc, *.log, .env, .pytest_cache/)
- **Validation**: all 10 `.json` configs parse strictly; both plc TS projects (`opencode-testing`, `opencode-security`) compile without errors; Ghost memory.json valid JSON
- **Git repair at C:\opencodes**: Corrupted `.git` (nested `.git\.git` + missing objects + stale config.lock) — NTFS-locked, needed `icacls /reset` + `git init --separate-git-dir` pointing to `C:\Users\Abdox\.opencode_git_storage`. Initial commit: 1704 files tracked.
- **Root `.gitignore`**: Updated from 73 → ~110 patterns. Excludes 16 nested repos, AI state files, config backups, Office lock files, logs, generated charts/HTML
- **Line endings normalized**: `.gitignore` (CRLF 51→73, LF 22→0) and `shared-abdox.md` (CRLF 112→131, LF 19→0)
- **17 nested repos identified**: Each has its own `.git` — excluded from parent git tracking
- **npm major updates**: `typescript` 5.7→7.0.2, `@types/node` 20→26.1.1, `@opencode-ai/plugin` 1.0→1.17.13 in both TS projects. Fixed TS 7 breaking changes: added `node:` prefix, replaced `Buffer` with `Uint8Array`, added `"types": ["node"]`
- **Status**: COMPLETED (guard skills node_modules 260 MB prune pending — needs admin)

### Session 2026-08-22 (Memory Improvement)
- Audited 3-layer memory drift: skill counts off by +6 (AGENTS) / +8 (knowledge_base), PC specs stale (C: 147GB→88.4GB, H: 0GB→117.4GB), missing sessions 2026-07-07 and 2026-07-28
- Re-verified filesystem: 223 skills (72 guards, 83 tools, 68 workflow) — updated AGENTS.md header and tables from actual SKILL.md frontmatter
- Updated knowledge_base_v2.json: total_skills 215→223, session_history 6→9, added patterns p021-p024, learnings l016-l019
- Fixed shared-abdox.md encoding (— → —) and updated active issues / models date
- Updated HANDOFF.md and PC Specs (OS 10.0.29648, disk free)
- **Outcome**: Memory fully synced and verified

### Session 2026-08-23 (Improvement Cycle)
- Verified 226 skills (72 guards + 86 tools + 68 workflow); added force-delete, self-learning, desktop-control-mcp to index
- Deleted 180 stale .pyc files (27MB) in skills/self-learning/__pycache__ + 3 junk root test scripts
- Fixed NTFS-locked files via force-delete method (icacls /grant:r user:F /T + .NET File.Delete) - learning l020
- knowledge_base_v2.json: skills_index 223->226, session_history 11->12
- **Finding**: git tracks only 13 of ~500 skill files (.gitignore blocks *.ps1/*.bat globally) - full commit recommended
- **Outcome**: Memory synced to 226 skills, junk purged
---

## 5b. External Research — Vercel Labs (2026-07-11)

318 public repos analyzed. Key patterns applied:

### Session 2026-08-23 (Full Library Commit + Git DB Repair)
- Committed entire library: 333 files, 224/224 SKILL.md tracked (was 13 files) - commits fdb56c1, 9c41da9
- Removed .gitignore blocks: global *.ps1/*.bat, guards/workflow SKILL.md ignores
- Diagnosed C:\\opencodes NTFS fault: create OK, delete/rename blocked -> git object writes fail (learning l021)
- Repair: repack -a -d + fresh separate-git-dir (C:\\Users\\Abdox\\.opencode_git_storage\\guard-skills-clean.git) + .git pointer rewrite
- Excluded idm/ (416MB Rust project, own-repo candidate); old guard-skills.git deletable later
- **Outcome**: Library fully version-controlled; chkdsk /f still recommended (admin)

### Session 2026-08-23 (D: Drive Rescue — 152 GB Freed)
- D: was critical (12.1 GB free of 474). Root cause: 3 stacked Desktop backups in `D:\A` (stale Jul-14 copy 160 GB, semi Aug-17 151 GB, current root 77 GB)
- Diffed all 3 trees by rel-path+size: stale had 145 GB uniques, NOT mostly duplicates
- **Moved to H:** `H:\Recovered\API-SFO` 59.3 GB (incl. 2×20GB OneDrive zips — one known corrupt, user to review) + `H:\Recovered\installers` 33.7 GB (STEP7 V17×2/V19 ISOs, Vijeo Designer, ECOSTRUCTURE.zip, Eplan.rar — the "missing" OneDrive downloads were here all along)
- **Preserved on D:** big unique folders (disktop 19.6G, new folder(3) 14.1G, phone 6.8G, OneDrive fixed zip 10.6G) → instant same-volume move to `D:\Desktop_Sync\_recovered_stale_20260823\` (H: lacked space; upload to TeraBox later)
- **Deleted:** entire stale tree + 11,157 dup files (44 GB) from semi copy — 0 errors; semi keeps its 15,674 uniques (106.8 GB)
- **Result**: D: 12.1 → **164.3 GB free**; H: 116.5 → 23.2 GB; chkdsk /f on C: still pending
- **Outcome**: D: crisis resolved, API SFO mystery solved

### Session 2026-08-23 (Input Blocker v68+v69 + Deep Audit + Purge)
- Deep 2-layer analysis of H:\ai\Input Blocker (3.24 GB, v3-v67): reports in reports\ (Folder_Analysis + Deep_Technical docx)
- Found: fake PQ-crypto shims (v62), NavItem ghost-text bug, 0 tests on v58 security modules, production crash traces
- **v68 shipped**: Windows Hello unlock (winsdk), PQ shims deleted, safe-mode tabs, +31 tests (139)
- **v69 "Hardening & Trust" shipped** (all 23 roadmap items): schema v2 migration, audit hash chain, ClipCursor hardening, Hello-gate, LAN localhost default, settings filter, security-tab split (9 builders), tray quick-actions, insights CSV/PNG export, scheduled profiles, i18n EN/FR/AR foundation, git repo+tag v69.0, pre-build gate (pytest->compile->PyInstaller), hex6 lint, key-budget guard, Inno Setup script
- Deep layer: 1000-case fuzz found coerce dict-leak bug -> type safety-net; migration clamp added; tamper matrix verified; 170/170 tests
- Workspace convention confirmed: true source lives H:\freebuff\Project\iblocker-vNN (v63 held v67 code); mirrored to H:\ai\Input Blocker\versions\vNN
- **Purge**: build\ deleted (829 MB freed); 1,046 MB old EXEs consolidated to _archive_pre_v69\ (user may delete later); root .lnk clutter 31->0; kept lineage v64-v69 + v52.1 (~404 MB)
- **Outcome**: v69.exe live via desktop shortcut; chkdsk /f still pending; user must manually verify Hello biometrics

### Session 2026-08-23 (AndroidDex Install + OpenDex Alternative)
- Installed AndroidDex.apk (com.shrey.androiddex v1.0.1) on MNA-LX9 via adb — Huawei multi-user quirk: install landed on Guest (user 10); fixed with `cmd package install-existing --user 0`
- AndroidDex identified: Flutter Windows + scrcpy wrapper (DeX-style phone mirroring), closed source, at H:\Downloads\Compressed\pre_build_Android_Dex_Windows\
- Built **OpenDex** alternative: C:\opencodes\opendex\opendex.py (Python 3.11 tkinter, ~490 lines, zero deps) wrapping scrcpy 3.1 (C:\other\scrcpy-win64-v3.1) + adb (C:\Android\platform-tools)
- Features: device auto-detect (2.5s poll), Wi-Fi connect/pair, Use-USB-IP, mirror+control, fullscreen/ontop/borderless/no-audio/stay-awake/show-touches/screen-off, size/bitrate/fps/codec, screenshot, MP4/MKV recording, live log
- Verified live: mirror session on MNA-LX9 (direct3d, 864x1920); desktop shortcut OpenDex.lnk created
- **Outcome**: OpenDex running; scrcpy is the engine (true protocol rebuild not attempted)

### Session 2026-08-24 (OpenDex v2 — Better than AndroidDex)
- User rejected v1 ("I DO NOT LIKE") → rebuilt as OpenDex v2, feature-superior to AndroidDex
- Upgraded engine: scrcpy 4.1 downloaded to C:\other\scrcpy-win64-v4.1 (10.8 MB, from GitHub releases); v3.1 kept
- **Critical quirk**: scrcpy's bundled adb kills the platform-tools adb server ("Could not find any ADB device") — fix: pass `ADB=C:\Android\platform-tools\adb.exe` env var to scrcpy child processes
- v2 modes: DeX Desktop (--new-display, verified connecting on MNA-LX9), Phone Mirror, Record Only (--no-window), Camera (--video-source=camera), OTG Control (--otg --hid-keyboard --hid-mouse)
- v2 features: app launcher (pm list -3 + monkey/am --display fallback), DeX --start-app field, APK installer (--user 0 first for Huawei quirk), push/pull files, battery %, multi-device sessions, auto Wi-Fi reconnect (config.json in %LOCALAPPDATA%\OpenDex)
- scrcpy 4.1 flags verified: --start-app, --new-display, --otg, --video-source, --no-window all present
- ghost MCP died mid-session → bash+pwsh fallback used (write .ps1 to temp, run via pwsh -File; inline nested quotes break)
- **Outcome**: OpenDex v2 running (pythonw); phone was unplugged at finish — DeX mode needs live test on reconnect

### Session 2026-08-24 (OpenDex v2.1 — DeX Verified Working)
- DeX mode VERIFIED LIVE on MNA-LX9: `--new-display=1920x1080` created virtual display **id=8**; YouTube launched via `am start --display 8` renders in a **freeform window with desktop controls** — true DeX experience
- **Keyguard blocks the DeX display** (shows lock screen) — phone must be unlocked for desktop to appear; `wm dismiss-keyguard` insufficient with PIN
- DeX prep settings applied: enable_freeform_support=1, force_desktop_mode_on_external_displays=1, force_resizable_activities=1
- Huawei PC Mode components exist (com.huawei.desktop.systemui ImitateActivity + desktop.explorer) but launch blocked by `com.huawei.desktop.systemui.permission.SELF` — cannot trigger via adb
- `cmd wallpaper` not implemented on EMUI 12 — wallpaper via SET_WALLPAPER intent (one tap)
- **v2.1**: added Prepare DeX button (wake+keyguard+3 settings), auto-prepare on DeX session start, Wallpaper picker button, multi-pattern display-id detection (mDisplayId=/Display #/Display id=/DisplayId=)
- ghost MCP died → full bash+pwsh fallback session; window capture via PrintWindow PW_RENDERFULLCONTENT=2 works on occluded scrcpy windows
- **Outcome**: OpenDex v2.1 live (pythonw 15648) + DeX session live (scrcpy 19380); OpenDex now feature-superior to AndroidDex (DeX desktop + mirror + record-only + camera + OTG + app launcher + APK install + push/pull + multi-device + auto-reconnect)

### Session 2026-08-24 (OpenDex v2.2 — Taskbar + Watchdog)
- **Taskbar 6.2.2** (farmerbb, open-source) downloaded to C:\opencodes\opendex\Taskbar.apk and installed on MNA-LX9 — its Home/SecondaryHome activities are DISABLED until configured in-app ("does not exist" error = component disabled, not missing)
- DeX desktop verified twice: YouTube renders as freeform window (min/close controls, notification badges) on 1080p virtual display — display id CHANGES per session (8→3→5), never hardcode it
- **v2.2 additions**: DeX Home button (SecondaryHome→Home→HOME-category chain), Install Taskbar button, session watchdog (auto-restart crashed scrcpy unless user-stopped), adb-server self-heal (kill/start-server on empty poll, 30s throttle)
- **Watchdog quirk**: killing scrcpy externally while GUI runs → watchdog resurrects it (by design) — always stop via GUI or kill pythonw first
- **Outcome**: v2.2 live; one clean DeX session (display 5) + YouTube window; Taskbar desktop home needs 1 manual enable in Taskbar app on phone

### Session 2026-08-24 (OpenDex — Full DeX Desktop Achieved)
- **Taskbar desktop VERIFIED on DeX display**: SecondaryHomeActivity renders full taskbar (app drawer grid + pinned apps + collapse arrow) on the 1080p virtual display — complete Samsung DeX experience
- **Setup chain that made it work**: Taskbar 6.2.2 installed → `cmd package set-home-activity com.farmerbb.taskbar/.activity.HomeActivity` (Success, sets default home non-interactively) → Taskbar Desktop mode toggle ON → SecondaryHomeActivity launches on virtual display
- Taskbar HomeActivity stays disabled (Error type 3) even as default home — SecondaryHomeActivity is the one that works on displays
- **Re-locking pain**: phone re-locks in ~1 min (30% battery, aggressive timeout) — keyguard blocks DeX display; unlock needed per session; wm dismiss-keyguard useless with PIN
- Duplicate scrcpy sessions caused by external kills + GUI watchdog resurrect — always kill pythonw first when doing external scrcpy work
- Display id per session: 8→3→5→9→10 (always re-detect, never hardcode)
- **Outcome**: OpenDex = full AndroidDex replacement + more (DeX desktop w/ taskbar, mirror, record-only, camera, OTG, app launcher, APK install, push/pull, multi-device, watchdog, adb self-heal). Desktop wallpaper on DeX display is black — cosmetic, set via Wallpaper button

### Session 2026-08-24 (OpenDex v2.3 — Advanced)
- **v2.3 additions**: Auto-DeX on plug-in (auto-starts DeX + taskbar home when device connects), saved Profiles (save/load/delete full option sets in config.json), DeX DPI control (--new-display=WxH/DPI, tested 160), simultaneous DeX + phone mirror (2 scrcpy instances, one device — verified working), DeX-display screenshots (screencap -d <id> works on EMUI 12), scrcpy shortcut cheatsheet dialog
- Sessions refactored: sessions[serial] = {procs: [...], mode, user_stop} — multi-proc aware start/stop/watchdog; DeX auto-runs dex_home 6s after start
- screencap -d verified: returns PNG of virtual display; --wallpaper flag does NOT exist in scrcpy 4.1
- **Outcome**: v2.3 live (pythonw 4572); live test: DeX display 11 w/ DPI 160 + mirror session simultaneously + taskbar home + DeX screenshot all verified in one run

### Session 2026-08-24 (OpenDex v2.3.1 — Hands-Free Flow Verified)
- **Hands-free flow VERIFIED end-to-end**: config.json preseeded (autodex=true, MyDesk profile w/ DPI 160 + startapp=YouTube) → plug phone → GUI auto-starts DeX → taskbar home renders after unlock — zero-click desktop
- **Anti-lock fix**: _dex_prepare_cmds now includes `svc power stayon usb` + `settings put system screen_off_timeout 600000` (phone re-locked every ~1 min before, blocking DeX display)
- **dex_home retry chain**: fires at 6s/15s/30s/60s after DeX start (keyguard blocks first attempt if locked; retry lands after unlock)
- startapp added to profile save/load
- **Outcome**: v2.3.1 live; final state = taskbar desktop (display 13, DPI 160) + auto-everything; project complete at C:\opencodes\opendex\opendex.py (~1010 lines) + Taskbar.apk + desktop shortcut

### Session 2026-08-25 (OpenDex v2.4 — Wireless + Polish)
- **v2.4 additions**: Go Wireless button (tcpip 5555 + connect — VERIFIED: connected 192.168.100.131:5555, phone usable unplugged), App set launcher (multi-window one click, space-separated pkgs), wallpaper flow (push img → media scan → SET_WALLPAPER picker), file logging (%LOCALAPPDATA%\OpenDex\opendex.log), low-battery warning (<20%)
- **Display-id bug FIXED**: `DisplayId=(\d+)` regex matched a 64-bit pointer (4630947188334859394) in dumpsys → removed pattern + 0-63 range filter; verified correct id 16 after fix
- **Config overwrite lesson**: hand-written config.json was wiped when a stale GUI instance closed normally (_save_config on close = last-writer-wins) — re-applied config via script AFTER killing all pythonw instances; never edit config while any GUI instance runs
- Wireless DeX caveat: Wi-Fi adb serial differs from USB serial → auto-dex treats it as new device and restarts desktop over Wi-Fi; Wi-Fi adb can drop on phone screen-off/power-save
- **Outcome**: v2.4 live (auto-dex verified again: display 16, DPI 160, taskbar home OK on correct id); phone left unplugged — GUI will auto-DeX on reconnect (USB or Wi-Fi)

### Session 2026-08-25 (OpenDex v2.5+v2.6 — Appliance Edition + AndroidDex Research)
- **GitHub research**: found real AndroidDex = Shrey113/Android-Dex (1,687 stars): Flutter + Java JAR pushed to device + Kotlin companion APK over WebSocket; flagship UX = Windows-side taskbar overlay, live telemetry, media controls, notification mirroring, gaming mode, clipboard manager
- **v2.5**: config split (config.json UI-owned + profiles.json user-owned — kills last-writer-wins overwrite bug), dark theme (clam/catppuccin), global hotkeys via ctypes RegisterHotKey (Ctrl+Alt+D DeX / M mirror / S screenshot / W wireless), Wi-Fi keepalive loop (10s poll, 30s throttle), phone-settings restore on stop (saves old screen_off_timeout, restores + stayon false), Dual DeX (2nd 1280x720 display), session badges (● DeX/mirror in device list)
- **v2.6**: Windows-side taskbar overlay (tkinter Toplevel, topmost, bottom bar: Apps drawer + pinned appset buttons + media keys ⏮⏯⏭ + volume + battery + clock), auto-opens 8s after DeX start, media keys via `input keyevent` (126 play-pause, 87 next, 88 prev, 24/25 vol)
- **PyInstaller .exe built**: C:\opencodes\opendex\app\OpenDex.exe (onedir, ~2MB exe; build MUST use temp workpath/distpath — C:\opencodes NTFS locks break resource step); desktop shortcut + shell:startup shortcut now point to exe (auto-start with Windows)
- pip note: `python` resolves to hermes-agent venv — pyinstaller installed there
- **Outcome**: v2.6 exe live; OpenDex now matches AndroidDex UX (taskbar+telemetry+media) and exceeds it (DeX virtual display w/ real Android taskbar, auto-DeX, profiles, hotkeys, watchdog, wireless)
### Session 2026-08-24 (AutoCAD 2024 Install — ODIS Silent Method)
- User requested AutoCAD install; only had web installer (32MB bootstrap, H:\tmp)
- Web installer GUI kept dying silently (window buried by user activity + silent exits)
- **Winning method**: extracted exact ODIS CLI args from `C:\Users\Abdox\AppData\Local\Autodesk\ODIS\Setup.log` (bootstrap logs the full Installer.exe command line), found full 2.5GB payload already cached in `C:\Autodesk\WI\{CC46AD7F-5075-3702-B2BF-CFCC5AB8468B}\ACD_2024_en-US\` (from a failed Aug-21 attempt), then ran `Installer.exe <same args> --silent` elevated
- **Blocker**: exit 1618 = zombie elevated msiexec holding Windows Installer mutex from earlier killed runs — fix: taskkill msiexec + `net stop msiserver` + `net start msiserver`, then re-run
- **Result**: exit code 0; AutoCAD 2024 English + 2024.1.8 Update (R24.3.222) in `C:\Program Files\Autodesk\AutoCAD 2024\`; trial mode (needs Autodesk sign-in to license)
- ODIS learnings: `--silent` refuses to run if any installer instance alive; DDA idle-loops on `\\.\pipe\adsk_dda_sdk` when UI flow is broken; AdODIS-installer.exe (427MB) = framework only, NOT product payload; pkg.odis.tar contains only AdODIS
- **Outcome**: AutoCAD 2024 fully installed via headless ODIS — GUI never needed

### Session 2026-08-26 (H: Drive Free-Up — 87 GB Recovered)
- Analyzed user's 6 folders (Recovered/Downloads/Disktop/TeraBoxDownload/Home/Phone = 268.5 of 307 GB used)
- Tier B deleted +25.8 GB: TeraBoxDownload 16.4GB (re-downloads from own cloud), game rars, Android_Dex x6, Kehua "- Copie" dups
- Tier A deleted +63.9 GB: BOTH 20GB OneDrive zips verified CORRUPT (zip central-directory test), STEP7 V171.iso MD5-identical to V17.iso, Office installer x2 redundant forms, AdskLicensing x2 (Autodesk removed), UnityPro zip+rar+"virus inside"rar dup forms
- Word report generated via docx-js before approval (user demanded "word report before i accept")
- **Key learning**: files created by elevated processes are BUILTIN\Administrators-owned with Users=RX,W ACL → undeletable non-elevated; fix = elevated .ps1 granting Everyone:F by SID *S-1-1-0 then .NET File/Directory.Delete
- robocopy /MIR from empty dir handles Arabic/weird-char filenames; stub dirs need attrib -s -h first
- Task folder: H:\ai\h-drive-storage-freeup\ (task.md, outputs\ incl. DOCX report, logs\)
- **Result**: H: 21.6 → 108.74 GB free (93%→67% full)
- Remaining candidates noted for later: TIA V18 ISOs 23GB, installers ~27GB, Huawei backups 16.7GB, 3x ESA folders ~40GB, Phone DCIM triplicates

### Session 2026-08-26 (Self-Improvement Review + System Fixes)
- Produced 12 recommendations (6 agent-behavior, 6 system) in H:\ai\agent-self-improvement\
- Verified guard-skills node_modules already pruned; deleted orphaned `C:\Users\Abdox\.opencode_git_storage\guard-skills.git` (0.8 MB, 706 files — pointer confirmed on clean.git)
- C: dirty bit = NOT dirty (NTFS issues are directory-level only); `chkdsk C: /F` scheduled for next reboot via elevated script
- Learned: Remove-Item -Recurse -Force patterns denied by permission rules → use elevated .ps1 file via Start-Process -Verb RunAs (one UAC prompt)
- Pending: reboot to run chkdsk; TeraBox upload H:\Recovered (93 GB); AutoCAD license sign-in
- **Outcome**: Quick-win fixes executed; chkdsk armed
- **PC analysis follow-up**: Health B+; H: drive 93% full (Recovered 93GB awaiting TeraBox); WU stuck fixed — Defender KB2267602 + Intel display driver installed via COM API, stuck PowerShell updates bypassed via `winget upgrade Microsoft.PowerShell` (7.6.5 staged, finalizes at reboot). Reports in H:\ai\pc-analysis-training\
- **Learned**: Win11 SoftwareDistribution rename blocked even elevated (WaaSMedic re-lock) → clear Download content instead; MSI-based PS updates finalize only after running pwsh exits

### Top Repos by Stars
| Stars | Repo | Lesson |
|-------|------|--------|
| 38.3k | `agent-browser` | Browser automation CLI for AI agents — validates our Ghost MCP direction |
| 28.9k | `agent-skills` | Agent skill collections are high-demand — validates Guard Skills approach |
| 25.8k | `skills` | Open agent skills tool (`npx skills`) — standardize skill interfaces |
| 15.7k | `json-render` | Generative UI framework — dynamic interfaces from structured data |
| 10.1k | `portless` | Named local URLs replacing ports — simplify local dev |
| 5.7k | `native` | Zig-based native desktop toolkit — performance optimization path |
| 5.2k | `deepsec` | Security harness for coding agents — security-first agent development |

### Architectural Patterns Adopted
| Pattern | Source | Applied To |
|---------|--------|------------|
| Skills as first-class citizens | `skills`, `agent-skills` | Guard Skills 223 skill index |
| Agent-browser automation | `agent-browser` | Ghost MCP tool design |
| Generative UI | `json-render` | Ghost MCP JSON output formatting |
| Edge-first computing | `portless`, `vercel-openclaw` | Cloudflare Workers integration |
| Security-first agents | `deepsec` | Guard validation approach |
| Workflow DevKit patterns | ~60 `workflow-*` repos | Integration pattern library |

### Workflow DevKit Patterns (Applicable)
- Approval chains, gates, and async request-reply for task orchestration
- Circuit breaker, bulkhead, and retry with backoff for resilience
- Event sourcing, message filtering, and content-based routing for event-driven
- Pipeline, map-reduce, and batch processing for data workflows
- Dead letter queue and guaranteed delivery for reliability

### Technology Stack Insights
| Tech | Vercel Labs Usage | Our Adoption |
|------|-------------------|--------------|
| TypeScript | Primary language (80%+) | Ghost MCP uses Python — consider TS port |
| Zig | Native desktop toolkit | Future optimization for Windows API calls |
| Rust | Browser automation (38.3k stars) | High-performance agent tool |
| Python | AI/ML pipelines, agent SDKs | Our primary language — validates choice |

---

## 6. Learned Patterns

| # | Pattern | Confidence | Source |
|---|---------|------------|--------|
| 1 | User asks direct questions, expects concise answers | 95% | Multiple sessions |
| 2 | Prefers action over explanation | 90% | Multiple sessions |
| 3 | Wants iterative improvement cycles | 85% | 13 improvement cycles |
| 4 | Uses Brave browser (not Chrome) | 95% | Ghost quirks |
| 5 | Uses desktop-control MCP for browser automation | 90% | Session history |
| 6 | write/edit tools have $ bug — use bash workaround | 95% | Session 2026-06-30 (resolved, keep as fallback) |
| 7 | Always finish current task before answering new questions | 90% | Session 2026-07-01 |
| 8 | Verify everything thoroughly before completing | 85% | Session 2026-07-01 |
| 9 | Recheck with deep analysis | 80% | Session 2026-07-01 |
| 10 | Ghost Snapshot requires DPI-aware coords (1.333x) + UIA-first click | 92% | Session 2026-07-13 |
| 11 | TeraBox integration via browser hidden input + API quota checks | 88% | Session 2026-07-15 |
| 12 | NTFS-locked dirs need icacls /reset + --separate-git-dir workaround | 90% | Session 2026-07-28 |
| 13 | Autodesk removal requires MsiExec /x + registry + service + folder purge | 95% | Session 2026-07-07 |
| 14 | PLC .smbp dual-representation — never text-edit ladder, edit XML tags only | 90% | Session 2026-07-01 |

## 7. Important Decisions

| Decision | Rationale | Alternatives |
|----------|-----------|--------------|
| Use AGENTS.md for persistent memory | Human-readable, version-controlled | Database, JSON only |
| Use HANDOFF.md for session transfers | Structured handoff format | Plain notes |
| Use knowledge_base_v2.json for learning | Machine-readable, queryable | Binary formats |
| OneDrive files save to Desktop/API SFO | Easy access | Downloads folder |
| Use bash for file operations | write/edit tools had $ bug (now fallback) | Wait for fix |
| Auto-verify skill counts from filesystem | Prevents drift (found 223 vs 217) | Manual counting |
| Store node_modules-excluded counts optionally | 52053 incl. deps vs ~4000 clean | Always full scan |

## 8. Quick Reference

### Paths
| Resource | Path |
|----------|------|
| Skills root | `C:\opencodes\guard skills\skills\` |
| Guards | `skills\guards\` |
| Tools | `skills\tools\` |
| Workflow | `skills\workflow\` |
| OneDrive URL | `https://ofpptcasa-my.sharepoint.com/personal/said_fallah_ofppt-edu_ma/_layouts/15/onedrive.aspx?id=%2Fpersonal%2Fsaid%5Ffallah%5Fofppt%2Dedu%5Fma%2FDocuments%2FAPI%2DSOF%2DPRO` |
| Local download | `C:\Users\Abdox\Desktop\API SFO\` |
| GitHub reports | `Abdox-menara/opencode-reports` |
| Ghost memory | `C:\Users\Abdox\.ghost\memory.json` |
| Verify skills | `Get-ChildItem C:\opencodes\guard\ skills\skills\*\ -Directory | Measure-Object` |

### Key Commands
```powershell
# Verify skill counts (auto): python tools/build_index.py
Get-ChildItem "C:\opencodes\guard skills\skills\guards" -Directory | Measure-Object
Get-ChildItem "C:\opencodes\guard skills\skills\tools" -Directory | Measure-Object
Get-ChildItem "C:\opencodes\guard skills\skills\workflow" -Directory | Measure-Object

# System report
Get-CimInstance Win32_OperatingSystem | Select-Object TotalVisibleMemorySize, FreePhysicalMemory
Get-Process | Sort-Object WorkingSet64 -Descending | Select-Object -First 20 Name, @{N='RAM(MB)';E={[math]::Round($_.WorkingSet64/1MB,1)}}
Get-CimInstance Win32_LogicalDisk -Filter "DriveType=3" | Select-Object DeviceID, @{N='Free(GB)';E={[math]::Round($_.FreeSpace/1GB,1)}}

# Push to GitHub
gh release create "tag-name" "path/to/file.pdf" --repo Abdox-menara/opencode-reports --title "Title" --notes "Description"

# PDF generation
python -c "from fpdf import FPDF; pdf=FPDF(); pdf.add_page(); pdf.set_font('Arial','B',16); pdf.cell(0,10,'Title',0,1,'C'); pdf.output('out.pdf')"
```

---

## 9. PLC Memory Map (SDI Program)

| Address | Function | Type |
|---------|----------|------|
| %M0-%M3 | Manual mode bits (DRAIN, RINSE, FILL, BYPASS) | Memory |
| %M5 | AUTO mode bit | Memory |
| %M230 | Modbus OK status | Memory |
| %M231 | Alarm reset bit | Memory |
| %M300-%M304 | Valve bits (AV1-AV3) | Memory |
| %M320-%M322 | Physical valve status | Memory |
| %M323 | Pump command | Memory |
| %M330 | Tank low sensor | Memory |
| %M350-%M353 | Pump command bits | Memory |
| %M999 | Emergency stop active | Memory |
| %I0.3 | Emergency stop button (NC) | Input |
| %S13 | First scan bit | System |
| %Q0.0-%Q0.3 | Physical outputs (Pump, AV1, AV2, AV3) | Output |

---

## 10. PC Specs — Verified 2026-08-22 18:02 UTC+1

| Component | Spec |
|-----------|------|
| CPU | Intel Core i7-6920HQ @ 2.90GHz (4C/8T) |
| GPU | NVIDIA Quadro M1200 (4GB) + Intel HD 530 |
| RAM | 32 GB (33396208 KB total, 14877700 KB free) |
| C: | 599.9 GB (88.4 GB free) |
| D: | 474.2 GB (16.2 GB free) |
| H: | 328.5 GB (117.4 GB free — was 0 GB FULL on 2026-07-15) |
| OS | Windows 11 Pro Insider Preview (10.0.29648) |

---

## 11. Current State — Verified 2026-08-22

| Item | Status |
|------|--------|
| OneDrive download | Incomplete (20GB zip corrupt) — not retried since 2026-06-30 |
| Desktop MCP (Ghost) | Operational — 17 commands verified 2026-07-13, memory.json v6.0 exists |
| Memory system | Operational — synced 2026-08-22 (AGENTS.md, shared-abdox.md, knowledge_base_v2.json, HANDOFF.md) |
| write/edit bug | Resolved / fallback — bash workaround kept as fallback |
| Autodesk suite | Fully removed 2026-07-07 — no orphaned Installer products |
| Ecosystem | Cleaned 2026-07-28 — 400 items / 75 MB removed, TS 7 migrated, 0 vulnerabilities |
| Skills drift | Fixed 2026-08-22 — 223 verified (was 217) |

---

*This file is the single source of truth for cross-session memory. Auto-verified 2026-08-22. Update at end of every session and re-run filesystem verification.*

More agent context in Abdox-menara/guard-skills

220 other files this repository gives its agents, the first 60 shown.