diff --git a/AGENTS.md b/AGENTS.md
index 2f856a10c..a06302d6c 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -1,10 +1,10 @@
-# AGENTS.md — Termux monorepo
+# AGENTS.md
 
-Instructions for coding agents (Grok, Claude, Codex, Devin, ChatGPT, local runners).
+This file provides guidance to agents when working with code in this repository.
 
-## Read first (in order)
+## Governance & Process (Read First)
 
-1. **This file** (`AGENTS.md`)
+1. **This file** (`AGENTS.md`) - Technical guidance and codebase overview
 2. [`docs/proposals/registry.yaml`](docs/proposals/registry.yaml) — what is active
 3. [`docs/proposals/PROCESS.md`](docs/proposals/PROCESS.md) — post / debate / consensus / close
 4. [`docs/PR-SUMMARY-PROCESS.md`](docs/PR-SUMMARY-PROCESS.md) — who may rewrite PR bodies (multi-agent)
@@ -15,7 +15,7 @@ Instructions for coding agents (Grok, Claude, Codex, Devin, ChatGPT, local runne
 
 Optional: `CLAUDE.md`, `CONTRIBUTING.md`.
 
-## Hard rules
+## Hard Rules
 
 - Target **`master-staging`**, not raw `master`, for integration work.
 - Both gates must pass before merge:
@@ -28,13 +28,7 @@ Optional: `CLAUDE.md`, `CONTRIBUTING.md`.
 - Unposted chat is not consensus — write Review log or DEBATE.md.
 - PR body rewrites: follow `docs/PR-SUMMARY-PROCESS.md` roster (not a single-agent monopoly).
 
-## Debate & close
-
-- Debate: MANIFEST Review log, optional DEBATE.md, linked PR/issue.
-- Close: all items terminal + Review log outcome + move `active/` → `closed/` + registry update.
-- Full rules: `docs/proposals/PROCESS.md` §§ consensus / closing.
-
-## Preferred execution loop
+## Preferred Execution Loop
 
 ```text
 registry.yaml → pick todo item → branch from master-staging
@@ -46,3 +40,526 @@ registry.yaml → pick todo item → branch from master-staging
 
 Credential rotation and history rewrite require Operator (human) authorization.
 See `docs/SECURITY-REMEDIATION.md`.
+
+---
+
+# Technical Documentation
+
+## Project Overview
+
+This is a **Termux-based multi-agent automation monorepo** running on Android. It orchestrates AI-powered development workflows through multiple specialized CLI tools and autonomous agents. The system integrates with DeepSeek API, manages conversation branching, performs code forensics, and maintains comprehensive knowledge indices.
+
+**Core Technologies:**
+- **Language:** Python 3 (primary), Node.js (supporting), Bash, Rust (selective components)
+- **Platform:** Termux on Android
+- **AI Integration:** DeepSeek API, multi-model support
+- **Architecture:** Microservices-style with shared workspace and indices
+
+**Codebase Statistics (via tokei):**
+```
+Total: 7,248 files | 2.1M lines
+- Code: 1.83M lines (87.4%)
+- Comments: 132K lines (6.3%)
+- Blanks: 133K lines (6.3%)
+
+Top Languages:
+- Rust: 1.2M lines (1,096K code + 98K blanks)
+- JSON: 575K lines (data/config)
+- Python: 96K lines (81K code, 775 files)
+- Shell: 19K lines (16K code, 243 files)
+- Zsh: 22K lines (15K code, 14 files)
+- TypeScript: 10K lines (5K code, 655 files)
+- JavaScript: 7K lines (6K code, 59 files)
+```
+
+**Key Architectural Components:**
+1. **ArchWiz** - Central automation cockpit with 28 tools across 7 categories
+2. **DeepCLI** - DeepSeek-oriented CLI for session management and streaming
+3. **Termux Multi-Agent** - Agent provisioning and orchestration
+4. **CLI Synthegration** - Conversation synthesis and branching
+5. **Harmonizer** - Unified automation interface
+6. **Central Mapper** - AST indexing and dependency graph generation
+
+## Navigation & Documentation
+
+**Primary Navigation (in priority order):**
+1. `archwiz/TOOL_INDEX.md` - 28 tools / 7 categories (cockpit, forensic, autonomous, verification)
+2. `archwiz/CONCEPT_INDEX.md` - Concepts + status + feature backlog
+3. `archwiz/REFERENCE_HUB.md` - Links to data flow, system map, indices
+4. `archwiz/METHODOLOGY_INDEX.md` - Approaches tried, failures, successes
+5. `archwiz/PROCEDURES.md` - Runbooks and active tasks
+6. `docs/RECON.md` - Branch/PR critique and proposals
+7. `README.md` - Recovery documentation and project inventory
+8. `workspace/llm_map/SYSTEM_MAP.md` - Complete project catalogue (8,239 files indexed)
+9. `archwiz/DATA_FLOW_MANIFEST.md` - Complete data flow manifest (1.5MB, 8,400 lines)
+
+**Quick Command Reference:**
+```bash
+# Research file history
+archaeo <file>
+
+# Check impact before changing
+oracle <file>
+
+# Make a change
+dispatch <task>
+
+# Validate & promote
+validate_promotion.py
+
+# Rebuild indices
+map-build && map-func && fore
+
+# Open cockpit
+python3 archwiz/archwiz.py
+```
+
+## Core Projects
+
+### 1. DeepCLI (`deepcli/`)
+DeepSeek-oriented CLI with session management, streaming, thinking mode, and export capabilities.
+
+**Entry Points:**
+- `deepcli.py` - Main CLI launcher
+- `deepapi.py` - API interface
+- `deepseek_proxy.py` - Proxy layer
+- `ci_mode.py` - CI/CD integration entry point
+- `ci_agent.py` - GitHub Actions agent logic
+
+**Key Commands:**
+```bash
+./deepcli.py new                                    # Create new session
+./deepcli.py send "Your prompt"                     # Send message
+./deepcli.py send "Complex task" --thinking         # Enable thinking mode
+./deepcli.py list                                   # List sessions
+./deepcli.py history --session <session_id>         # View history
+./deepcli.py export --format json --output chat.json # Export session
+./deepcli.py fork --session <id> --message-id <msg> # Fork conversation
+```
+
+**CI/CD Integration:**
+- GitHub Actions workflow: `.github/workflows/deepseek-ci.yml`
+- Triggers on PR events (opened, synchronize, reopened)
+- Uses PoW WASM solver for authentication
+- Automated PR review and commenting
+
+**Related:** `deepcli-tui/` (TUI with conversation tree), `.deepcli/` (config), `deepseek-cli/` (includes deepterm)
+
+### 2. Termux Multi-Agent (`termux-multi-agent/`)
+Multi-agent orchestration system for provisioning, running, and managing autonomous agents.
+
+**Key Components:**
+- `provision_agent.py` - Agent provisioning
+- `run.py` - Main run loop
+- `dashboard.py` - Status/control UI
+- `patch_files.py` / `patch_files_final.py` - Patch application
+- `cedar-mcp-server.js` - CedarScript MCP server
+- `run_history.jsonl` - Run history tracking
+
+**Task Management:**
+- Tasks loaded from `~/workspace/llm_map/master_tasks.json`
+- Environment variables: `TASK_ID`, `REFACTOR_GOAL`, `TARGET_FILE`, `TASK_WORKSPACE`
+- Fallback: `current_task.txt` in workspace
+
+### 3. CLI Synthegration (`cli-synthegration/`)
+Conversation synthesis with branching, export, account/token management, and metrics.
+
+**Notable Modules:**
+- `branch_manager.py` - Conversation branch management
+- `conv_branching.py` - Branch/fork API implementation
+- `conv_explorer.py` - Conversation exploration
+- `conv_export_cli.py` - Export functionality
+- `account_manager.py` - Multi-account support
+- `token_provider.py` - Token management
+- `live_export.py` / `live_search.py` - Real-time operations
+- `sync_pipeline.py` - Synchronization
+- `Chronos/` - Time-loop acceleration
+- `workspace/` - Working directories
+- `metrics/` - Performance tracking
+
+**Data Storage:**
+- `synthegration_exports/` - 3,281 exported session files
+- `workspace/correlation/` - Session-to-file correlation index
+- `workspace/provenance/` - Version tracking and provenance
+- `codex/message_index.json` - Full-text message search index
+
+### 4. ArchWiz (`archwiz/`)
+Central automation cockpit with indexing, provenance tracking, and recovery tools.
+
+**Tool Categories (28 tools total):**
+
+**Cockpit & Pipeline:**
+- `archwiz.py` - Dashboard with 16 options + 3 mode toggles
+- `activity_listener.py` - Auto-executes assistant code blocks
+- `live_view.py` - Review panel with /exec, /skip, /send commands
+- `debug_daemon.py` - Watches failures, auto-fixes with ruff/shellcheck
+- `listener_control.py` - PID-based safe start/stop
+
+**Forensic & Version Control:**
+- `forensic_toolchain.py` - Fragment matcher, similarity scan, correlation scout
+- `correlation_scout.py` - Traces file-path changes across versions
+- `fragment_matcher.py` - Function-level provenance
+- `restore_version.py` - Provenance-based code resurrection
+
+**Autonomous Operation:**
+- `autonomous_runner.py` - Dispatches pending tasks with memory awareness
+- `dispatch_task.py` - Sandboxed execution with Sentinel verification
+- `task_builder.py` - Interactive task creation
+- `auto_repair.py` - Auto-fixes Sentinel REVIEW issues
+
+**Verification & Testing:**
+- `sentinel.py` - 5-gate verification (file integrity, naming, duplicate, probe, shockwave)
+- `probe.py` - Syntax/import/test validation
+- `mirror.py` - Self-critique for task hygiene, index freshness
+- `dangle_detector.py` - Cross-ecosystem broken reference scanner
+
+**Knowledge & Memory:**
+- `archivist.py` - Local query engine across all indices
+- `tasque_declare.py` - Declares completion to taDone.md
+- `timeline_editor.py` - Full DB editor + archaeologist
+- `narrative.py` - Chronological feed of pipeline events
+- `lexicon_harvest.py` - Session scanning for novel terms
+- `name_forge.py` - Grimoire-powered tool naming
+
+**Documentation Pipeline:**
+- `session_digest.py` - Scans exported sessions for structured features
+- `structural_scanner.py` - Fast chunked-correlation scanner
+- `export_status.py` - Shows cached vs exported sessions
+- `pointer_index.py` - Builds hash→location map of code blocks
+
+### 5. Central Mapper (`central_mapper_v420.py`, `workspace/llm_map/`)
+Comprehensive codebase indexing and dependency analysis.
+
+**Key Indices:**
+- `llm_index.jsonl` - Full LLM-optimized index
+- `llm_index_compact.jsonl` - Compact version (1.99MB)
+- `func_index.jsonl` - Function-level index (287KB)
+- `central_enriched.jsonl` - Enriched central index
+- `task_files_index.json` - All task/sprint/todo files
+- `file_graph.json` - Dependency graph
+- `deps.jsonl` - Dependency relationships
+
+**Build Scripts:**
+- `build_all.py` - Complete index rebuild
+- `build_llm_index.py` - LLM index generation
+- `build_ast_index_from_existing.py` - AST index from existing data
+- `build_graph_fast.py` - Fast graph generation
+
+### 6. Harmonizer (`harmonizer-prod_cli/`)
+Production Harmonizer CLI providing unified DeepSeek automation: sessions, export, search, sync.
+
+### 7. Supporting Projects
+- `harmony_hub/` - Harmony hub integration
+- `multi-ai-cli/` - Multi-model CLI surface
+- `commingle-swarm/` - Template/scavenge-only (not first-class runtime)
+- `colab-cli/` - Colab CLI tooling
+- `exchanges/` - Exchange/market API code
+- `appliedSxi/maxc/` - Applied Sxi/Max work
+
+## Development Workflow
+
+### Building and Running
+
+**Prerequisites:**
+```bash
+# Install base requirements
+pip install -r requirements-base.txt
+
+# Core dependencies: curl-cffi, requests, websockets
+```
+
+**Initialize Database:**
+```bash
+cd termux-multi-agent
+python run.py  # First run creates workspace and dummy files
+```
+
+**Start ArchWiz Cockpit:**
+```bash
+python3 archwiz/archwiz.py
+```
+
+**Run Autonomous Agent:**
+```bash
+cd termux-multi-agent
+export TASK_ID="your-task-id"  # Or set REFACTOR_GOAL
+./run_agent.sh
+```
+
+**Rebuild Indices:**
+```bash
+# Central mapper
+python3 central_mapper_v420.py
+
+# Provenance indices
+cd cli-synthegration/workspace/provenance
+python3 final_provenance.py
+python3 comprehensive_fast.py
+
+# LLM map (large, may be slow)
+cd workspace/llm_map
+python3 build_all.py
+```
+
+### Testing
+
+**Validation Pipeline:**
+```bash
+# Sentinel 5-gate verification
+python3 archwiz/sentinel.py <file>
+
+# Probe syntax/import/test validation
+python3 archwiz/probe.py <file>
+
+# Mirror self-critique
+python3 archwiz/mirror.py
+```
+
+**Agent Test Runs:**
+- Test verdicts tracked in `termux-multi-agent/run_history.jsonl`
+- Dashboard available via `termux-multi-agent/dashboard.py`
+
+## CI/CD & GitHub Actions
+
+### Active Workflows
+
+**DeepSeek CI (`.github/workflows/deepseek-ci.yml`):**
+- Triggers: PR opened, synchronize, reopened, manual dispatch
+- Features: Automated PR review, code analysis, comment posting
+- Authentication: PoW WASM solver (deepseek.wasm + pow_solver.js)
+- Session caching for performance
+- Artifact upload for results
+
+**Other Workflows:**
+- `agent-continuous-ops.yml` - Continuous agent operations
+- `agent-feedback-linear-sync.yml` - Linear issue sync
+- `agent-jules-on-issues.yml` - Jules agent for issues
+- `agent-review-auto-jules.yml` - Automated Jules reviews
+- `gemini-*.yml` - Gemini integration workflows
+- `peer-review-orchestrator.yml` - Multi-peer review coordination
+- `publish-wiki.yml` - Wiki publishing
+
+### GitHub Actions Best Practices
+- Use `OPERATOR_TOKEN` secret for authenticated operations
+- Cache session data to reduce API calls
+- Fetch WASM files from correct commit hash
+- Use metadata-only artifacts for large outputs
+- Implement proper error handling and timeouts
+
+## Development Conventions
+
+### Code Style
+- **Python:** Follow PEP 8, use `ruff` for linting and auto-fix
+- **Shell:** Use `shellcheck` for validation
+- **Logging:** Use existing patterns or language standard logging library/facade
+- **Performance:** Check performance implications of changes
+- **Security:** Review security considerations, especially for API tokens and credentials
+
+### File Organization
+- **Indices:** Stored in `workspace/llm_map/`, `archwiz/`, `cli-synthegration/workspace/`
+- **Configuration:** Project-specific configs in respective directories
+- **Backups:** `.bak` files created automatically, large JSONL indices backed up externally
+- **Ignore Patterns:** Respect `.gitignore` and `.bobignore` (see root `.bobignore` for exclusions)
+
+### Commit Practices
+- Use feature branches (`feature/*`) for all changes
+- Never commit directly to `master` without review
+- Reference task IDs in commit messages when applicable
+- Keep commits focused and atomic
+- Follow conventional commit format when possible
+
+### Security & Secrets
+- **Never commit:** API keys, tokens, cookies, session stores
+- **Sensitive files:** `cookies_2.json`, `*_api_key`, `*.env`, `*.pem`, `*.key`
+- **Session stores:** Must not be tracked in Git (see PR #3 `agent/repository-hygiene`)
+- **Backup:** Large JSONL indices externally before destructive rebuilds
+- **Browser data:** `deepcli/browser-data*/` and `deepseek-cli/browser-data*/` contain sensitive cookies
+
+### Autonomous Operation
+- **Task Sources:** `master_tasks.json`, `TASK_ID` env var, `REFACTOR_GOAL` env var, `current_task.txt`
+- **Execution:** Sandboxed with Sentinel verification
+- **Memory:** Memory-aware dispatching with crash recovery
+- **Validation:** Auto-repair for simple issues, manual review for complex ones
+
+### Index Management
+- **Pointer Index:** CID-style bookmarking (`archwiz/pointer_index.json`)
+- **Correlation Index:** Session-to-file links (`cli-synthegration/workspace/correlation/`)
+- **True Versions:** Version hash tracking (`cli-synthegration/workspace/provenance/true_versions.json`)
+- **Message Index:** Full-text session messages (`cli-synthegration/codex/message_index.json`)
+- **Task Files:** All task/sprint/todo files (`workspace/llm_map/task_files_index.json`)
+- **Data Flow Manifest:** Complete file writer tracking (`archwiz/DATA_FLOW_MANIFEST.md`)
+
+### Forensic Recovery
+- **Staged Blocks:** `archwiz/staging_blocks.json`
+- **Pipeline:** Extract with forensic toolchain → restore with `restore_version.py`
+- **Backups:** Automatic `.bak` creation before modifications
+- **Restore Points:** Up to 5 per file (0 = initial state)
+
+## Key Concepts
+
+### Core Abstractions
+- **TasQue:** Task completion declaration system (ta'Done)
+- **Sentinel:** 5-gate verification (file, naming, duplicate, probe, shockwave)
+- **Archivist:** Local-only query engine across all indices
+- **Probe:** Syntax/import/test validation
+- **Mirror:** Self-critique for task hygiene and index freshness
+- **Dangle Detector:** Cross-ecosystem broken reference scanner
+- **Pointer Index:** CID-style bookmarking of messages, tables, data
+- **Narrative Feed:** Chronological event stream of pipeline events
+
+### Reserved Concepts (Not Yet Implemented)
+- **Spellbook:** Library of system abilities
+- **Rune:** Short hash pointer (CID-style)
+- **Sigil:** Substitution engine for runtime compression
+- **ChronoMancer:** Time-loop agent with success-only trunks
+- **Self-healing Sandbox:** Detect error → request fix → validate → promote
+
+### Methodology Evolution
+| Phase | What We Tried → What Stuck |
+|-------|---------------------------|
+| **Listener lifecycle** | `nohup` → `pkill -f` → `Popen` → `listener_control.py` (PID file) |
+| **Chat feedback** | `deepcli_send.py` (new session) → `send_message` (missing fields) → `stream_completion` (TUI pipe) |
+| **Session cache** | `synthegration export` → `manifest.json` → `get_history()` direct |
+| **Block tracking** | Message-ID → per-block hash (MD5 first 12 chars) |
+| **Live View** | Curses → text loop → throttled redraw + `/send` single-line |
+| **Hang avoidance** | `stdin=DEVNULL` + `start_new_session=True` + PID-file controller |
+
+## Workspace Structure
+
+```
+~/
+├── archwiz/              # Central automation cockpit (28 tools)
+├── deepcli/              # DeepSeek CLI (205 files)
+├── termux-multi-agent/   # Agent orchestration (40 files)
+├── cli-synthegration/    # Conversation synthesis (1,355 files)
+├── harmonizer-prod_cli/  # Production harmonizer
+├── workspace/            # Shared workspace
+│   └── llm_map/         # LLM indices (8,239 files indexed)
+├── refTemplates/         # Reference templates (metadata-only)
+├── _1-Projects/          # Project tree (222 files)
+│   ├── a/               # Arbitrage projects (54 files)
+│   └── b/               # BSC, Yobit, eggshell (168 files)
+├── exchanges/            # Exchange APIs
+├── sandbox/              # Experimental code
+├── src/                  # Shared sources
+├── bin/                  # Binaries
+├── config/               # Configuration
+├── synthegration_exports/ # 3,281 exported sessions
+├── deepseek_harvest_work/ # 1,417 harvested code files
+└── deepseek-cli/         # 1,463 files (includes browser data)
+```
+
+## Installed Tools
+
+- **ruff** - Python linter + auto-fix
+- **shellcheck** - Shell script analysis
+- **ripgrep (rg)** - Fast recursive search
+- **fd** - Fast `find` alternative
+- **jq** - JSON processor
+- **entr** - File watcher
+- **tokei** - Code statistics and line counting
+
+## Open Work & Priorities
+
+### Active PRs (as of 2026-08-01)
+1. **PR #1 `critical-proposal`** - Critical eval + roadmap (mergeable as docs)
+2. **PR #2 `timerloggedout-spec-patch-1`** - GHA Rust (narrow scope before merge)
+3. **PR #3 `agent/repository-hygiene`** (draft) - Untrack session stores (**priority security**)
+
+### Active Branches (Critical Evaluation)
+
+**High Priority:**
+- `recreate/refTemplates-skeleton` - Full metadata tree (merge to master)
+- `agent/repository-hygiene` - Session store security (PR #3, priority)
+- `mistral/fixes-config-security` - config.py + security baseline
+
+**Under Review:**
+- `feat/gh-actions/deepseek-integrates-itself` - DeepSeek CI workflow (fixed, ready)
+- `critical-proposal` - Documentation and architecture critique (PR #1)
+- `vibe/mistralai-vibe-code-wrapper-*` - Mistral CLI + harvester
+
+**Branch Health Notes (from RECON.md):**
+- `master` @ `6ef0e2f` - Protected, recovery README + live inventory
+- Legacy paths (`export_poller.sh`, `activity_listener.py`) are candidates for archive
+- Prefer `dispatch_pipeline` on cache write over dual maintenance
+
+### Feature Requests
+- Real-time chat feedback from listener
+- Cross-session idea harvester
+- ChronoMancer branch UI (visual tree)
+- Listener auto-scribe (consolidate notes)
+- Multi-account probing with image upload
+- Tab completion for /sessions
+- Expert-mode session creation
+
+### Known Issues & Technical Debt
+1. **Termux path coupling** - Absolute paths in REFERENCE_HUB need portability
+2. **Silent `except: pass`** - In dispatch (deepcli + multi-ai-cli)
+3. **Broken root symlinks** - Path coupling issues
+4. **refTemplates gaps:**
+   - Category 15 (Reverse Engineering) removed, needs metadata-only restore
+   - Uncategorized: Haven/, Interpreted-Context-Methdology_fork/
+5. **Session store hygiene** - PR #3 addresses credential-adjacent risk
+
+## Troubleshooting
+
+### Common Issues
+
+**Listener Hangs:**
+- Use `listener_control.py` for safe start/stop
+- Check PID file in `.deepcli/`
+- Ensure `stdin=DEVNULL` and `start_new_session=True` in subprocess calls
+
+**Index Staleness:**
+- Run `mirror.py` to check index freshness
+- Rebuild with `map-build && map-func && fore`
+- Check backup age before destructive rebuilds
+
+**Task Not Found:**
+- Verify `TASK_ID` in `master_tasks.json`
+- Check `REFACTOR_GOAL` env var
+- Fallback to `current_task.txt` in workspace
+
+**Sentinel Failures:**
+- Review 5-gate output (file, naming, duplicate, probe, shockwave)
+- Use `auto_repair.py` for simple fixes
+- Manual review for complex issues
+
+**Broken References:**
+- Run `dangle_detector.py` for cross-ecosystem scan
+- Check correlation index for file-path changes
+- Use forensic toolchain for recovery
+
+**CI/CD Workflow Failures:**
+- Verify `OPERATOR_TOKEN` secret is configured
+- Check WASM file paths and commit hash
+- Review workflow logs in GitHub Actions
+- Ensure session cache is properly configured
+
+## Additional Resources
+
+- **Full Documentation:** See `archwiz/REFERENCE_HUB.md` for comprehensive links
+- **Methodology:** `archwiz/METHODOLOGY_INDEX.md` documents what worked and what didn't
+- **Procedures:** `archwiz/PROCEDURES.md` contains runbooks
+- **Recovery:** `README.md` has detailed recovery procedures and refTemplates snapshot
+- **Recon:** `docs/RECON.md` contains branch/PR critique and proposals
+- **System Map:** `workspace/llm_map/SYSTEM_MAP.md` - Complete project catalogue
+- **Data Flow:** `archwiz/DATA_FLOW_MANIFEST.md` - Complete data flow manifest
+
+## Notes for AI Agents
+
+1. **Always check indices first** - Use `archivist.py` to query existing knowledge before making changes
+2. **Respect the verification pipeline** - Run Sentinel and Probe before promoting changes
+3. **Use feature branches** - Never commit directly to master
+4. **Maintain provenance** - Update correlation and version indices when modifying files
+5. **Check for existing patterns** - Review similar code before implementing new features
+6. **Security first** - Never expose tokens, keys, or credentials
+7. **Document decisions** - Update relevant indices and documentation
+8. **Test incrementally** - Use sandbox for experiments, validate before promoting
+9. **Respect ignore patterns** - Honor `.gitignore` and `.bobignore` exclusions
+10. **Prefer metadata-only** - For refTemplates, use depth-1 sparse checkout with metadata only
+11. **Check RECON.md** - Review branch health and known issues before major changes
+12. **Use tokei** - Run `tokei --sort code` to understand codebase composition
+13. **Follow CI/CD patterns** - Use existing GitHub Actions workflows as templates
+14. **Cache aggressively** - Session caching reduces API calls and improves performance
+15. **Monitor data flow** - Check DATA_FLOW_MANIFEST.md for file writer relationships