🚀 Mobile Team Agent

v4.0.0 — Context-aware, memory-driven developer assistant with Jira + Git integration, smart ticket guidance, persistent preferences, intelligent workflow automation, Figma design-to-code, React Native project setup, deep code review, pre-PR merge prediction with risk scoring, unit test generation, prompt validation, context engineering, and session memory across days.

The Mobile Team Agent is an MCP (Model Context Protocol) server that plugs into Claude CLI. It gives Claude a full suite of tools for mobile developers — Jira ticketing, Git insights, Figma design reading, RN project scaffolding, code review, pre-PR merge analysis, unit test generation, AI efficiency gating, and persistent memory — all accessible via natural language.

📦 Installation

Option 1 — npm (Recommended)

npm install -g mobile-team-agent

Then run setup from inside your project directory:

cd /path/to/your/project
npx mobile-team-agent setup

⚠️ Always cd into your project first. Setup writes a CLAUDE.md into the current directory so Claude knows the agent rules for that project.

setup does three things automatically:

1. Registers the MCP server with Claude CLI globally (~/.claude/settings.json)
2. Installs agent instructions to ~/.claude/CLAUDE.md
3. Writes a CLAUDE.md into your current project folder with agent rules, daily workflow shortcuts, Change Safety Protocol, and memory shortcuts

If your project already has a CLAUDE.md, the agent block is safely appended between markers — your existing content is never touched.

To add the agent to another project later, just run setup from that project's root:

cd /path/to/another/project
npx mobile-team-agent setup

Option 2 — Clone & install

git clone <repo-url>
cd "Mobile Team Agent"
chmod +x install.sh && ./install.sh

Manual registration (fallback)

claude mcp add mobile-team-agent -- node $(npm root -g)/mobile-team-agent/Main/index.js

Requirements: Node.js ≥ 18, Claude CLI installed.

⚡ Quick Start

Once installed, open Claude CLI from any project directory:

claude

Activate the agent:

• "invoke mobile-team-agent" — activates, shows connection status, and surfaces your last session context
• "Good morning" / "hi" / "start my day" — morning standup
• "plan my day" — deep daily planning
• "end of day" / "EOD" — generate daily report + save session snapshot

🔌 Connecting Your Tools

Jira

"connect Jira"
"configure Jira for project MyApp"

Or set environment variables:

export JIRA_URL="https://your-org.atlassian.net"
export JIRA_EMAIL="[email protected]"
export JIRA_TOKEN="your-api-token"

Figma

"connect Figma"
"set up Figma"

The agent walks you through generating a Personal Access Token step by step. Once connected, paste the token (figd_...) and it's saved for all future sessions.

Git

Git is auto-detected from the current directory — no setup needed.

🗂 All Tools by Category

🎯 AI Efficiency & Prompt Validation

by Shekhar Manwar

Runs before every other skill — it decides whether the agent has enough information to implement correctly, and keeps it from reading half the repo to find out.

| Tool | What it does | Say | |------|-------------|-----| | analyze_request | Validate a request before implementing: goal, scope, inputs, expected output, constraints, edge cases. Classifies missing info as CRITICAL / IMPORTANT / OPTIONAL and returns a READY / PARTIALLY READY / NEED CLARIFICATION verdict | "check my prompt" / "is this enough information?" / /analyze <request> | | narrow_code_scope | Find the smallest relevant set of files for a feature or symbol, ranked, with the matched lines — so only those files get opened | "where is end_of_day_report implemented" / /narrow biometric login | | prompt_template | The standard prompt structure (goal, context, scope, requirements, constraints, inputs, expected output, validation) | "give me a prompt template" / /prompt add offline caching |

The gate the agent applies to every request:

1. Reuse context first — conversation → project files → ticket → saved memory → config. Never asks for what it already has.
2. Never assume file names, API contracts, business rules, error handling, defaults, auth behaviour, data models, or supported versions.
3. Classify the gap — CRITICAL (stop and ask) / IMPORTANT (ask only if a wrong guess causes rework) / OPTIONAL (never blocks).
4. Decide — READY → implement; PARTIALLY READY → implement and state assumptions; NEED CLARIFICATION → one grouped question set, critical first.
5. Narrow before reading — feature → files → functions; change only what was asked.

Clear requests still run immediately — "fix this function" with the function attached is never turned into a questionnaire.

🔧 Setup & Connection

| Tool | What it does | Say | |------|-------------|-----| | invoke_mobile_team | Activates the agent, scans project reports, restores last session context | "invoke mobile-team-agent" | | get_setup_status | Shows connected integrations and saved preferences | "check connection status" | | configure_service | Save Jira credentials (per project) | "configure Jira" | | switch_jira_project | Switch active Jira project context | "switch to MyApp project" | | health_check | Test all integrations | "are my connections healthy?" | | set_preferences | Save default project/sprint/assignee/name | "save these as defaults" | | jira_connection_test | Verify Jira token without reconfiguring | "test Jira connection" |

📋 Jira — Reading Tickets

| Tool | What it does | Say | |------|-------------|-----| | list_tickets | My open tickets (flexible filters) | "show my tickets" / "list PROJ tickets" | | smart_ticket_query | Categorized sprint board view | "show sprint board" | | fetch_jira_tickets | Raw JQL power queries | "JQL: project = PROJ AND status = 'In Progress'" | | get_ticket_details | Full details: description, comments, changelog | "tell me about PROJ-42" | | select_ticket | Pick a ticket + get an implementation plan | "PROJ-42" | | get_ticket_suggestions | AI-scored recommendations on what to work on | "what should I work on?" | | analyze_workload | Categorize all tickets: Done / In Progress / Blocked / Overdue | "analyze my workload" | | list_projects | List all Jira projects | "list Jira projects" | | list_sprints | List sprints for a project | "show sprints for PROJ" | | search_users | Find Jira users by name/email | "find user John" |

✏️ Jira — Writing & Actions

| Tool | What it does | Say | |------|-------------|-----| | transition_ticket | Move ticket status (To Do → In Progress → Done) | "move PROJ-42 to In Progress" | | add_comment | Comment on a ticket | "add comment to PROJ-42: done and tested" | | create_ticket | Create a new Jira ticket | "create a bug ticket in PROJ" | | assign_ticket | Assign ticket to a user | "assign PROJ-42 to me" | | log_work | Log time spent | "log 2h on PROJ-42" | | get_create_meta | Fetch required fields before creating a ticket | (auto-called internally) | | sync_offline_actions | Retry queued actions from when offline | "sync offline actions" |

🔀 Git & Commits

| Tool | What it does | Say | |------|-------------|-----| | get_recent_commits | Git log with Jira linking, file diff stats, work area analysis | "show my recent commits" / "what did I commit today?" | | check_branch_sync | Check the local branch against its remote before work starts — fetches refs (read-only), reports ahead/behind, and says when a git pull is needed. Never pulls (by Shekhar Manwar) | "am I up to date?" / "is my branch synced?" | | get_commit_details | Full commit deep-dive: patch, files changed, lines +/-, Jira tickets | "show changes in commit abc1234" |

🎯 Daily Workflow

| Tool | What it does | Trigger | |------|-------------|---------| | morning_standup | Today's tickets, recent commits, priorities | Greeting: "hi", "good morning", "start my day" | | plan_my_day | Deep plan: new/pending/blocked/overdue, comment context, code activity, yesterday's work | "plan my day" / "what should I focus on today?" | | end_of_day_report | Generate + save EOD summary and session snapshot | "end of day" / "EOD" / "wrap up" | | get_daily_report | Retrieve a saved report for a specific date | "show report for 2024-01-15" | | list_daily_reports | Browse all saved daily reports | "list my reports" | | weekly_summary | Weekly rollup across all work | "weekly summary" | | get_consolidated_summary | Cross-project daily summary from all project folders | "all projects today" |

Reports are saved per-project to ~/Documents/MobileTeamAgent/<ProjectName>/DD-MM-YYYY_updates.md

🧠 Memory (Persistent Across Sessions)

| Tool | What it does | Say | |------|-------------|-----| | remember | Save a note (auto-links to ticket keys mentioned) | "remember: use MMKV for token storage" | | recall | Search saved notes | "what did I note about auth?" | | recall_ticket | Get all memory for a specific ticket | "recall notes for PROJ-42" | | journal | Add a real-time work log entry | "I just finished the login screen" | | show_journal | Show today's journal entries | "show my journal" | | add_decision | Record a team decision that persists until resolved | "we decided to use Zustand for state" | | show_decisions | List all active decisions | "what decisions are pending?" | | resolve_decision | Mark a decision resolved | "resolve decision about state management" | | forget | Delete a stored memory entry | "forget that note" | | memory_status | Show memory usage stats | "memory status" |

🔁 Session Snapshot (Next-Day Context)

At the end of every end_of_day_report and plan_my_day, the agent automatically saves a session snapshot to:

~/Documents/MobileTeamAgent/<ProjectName>/session_snapshot.md

It contains:

• 🟠 In-progress tickets (with last saved note)
• ✅ Completed today
• 📋 Pending / next up
• 🚫 Blocked tickets
• 💻 Today's commits
• 📓 Journal entries
• 🤝 Open decisions
• 🎯 Where to pick up tomorrow

When you say "invoke mobile-team-agent" or "good morning" next day, this snapshot is loaded automatically so Claude knows exactly where to pick up — no re-explaining needed.

🎨 Figma Design-to-Code

| Tool | What it does | Say | |------|-------------|-----| | configure_figma | One-time setup: save & validate Figma token | "connect Figma" / "set up Figma" | | figma_connection_test | Verify saved token without reconfiguring | "test Figma connection" | | list_figma_screens | List all top-level frames in a Figma file (names + dimensions) | "show Figma screens" / "list frames" | | read_figma_screen | Full design data for one screen: text, colors, fills, auto-layout, padding, child hierarchy + PNG URL | "read the Login screen" / "build the Home screen from Figma" | | suggest_figma_screens | Suggest screens not yet implemented in the project (5 at a time) | "suggest screens to implement" / "next 5 screens" |

Always call read_figma_screen before writing code for a screen — it provides the real design data including colors, spacing, and hierarchy.

Pagination for suggestions:

"show 5 more"       → offset=5
"page 2"            → offset=10
"refresh screens"   → refresh=true

⚛️ React Native Project Setup

| Tool | What it does | Say | |------|-------------|-----| | setup_rn_project | Scaffold new RN or Expo project with TS, ESLint, Prettier, Jest, and opinionated library stack | "set up a new RN project called MyApp" / "create an Expo app named ShopApp" | | analyze_rn_architecture | Audit existing RN project: missing folders, anti-patterns, installed libraries, architecture score | "review my project structure" / "is my RN architecture correct?" | | recommend_libraries | Opinionated library recommendation for a feature with install command + minimal setup | "what should I use for navigation?" / "recommend a library for auth" |

setup_rn_project parameters:

• name — project name (e.g. MyApp)
• type — cli or expo
• features — any of: navigation, state, networking, storage, forms, testing, ui, auth, analytics, crash

recommend_libraries feature keywords:

| Keyword | Recommends | |---------|-----------| | navigation | React Navigation v7 / Expo Router v4 | | state | Zustand / Redux Toolkit | | networking | Axios + TanStack Query | | storage | MMKV / AsyncStorage | | forms | React Hook Form + Zod | | testing | Jest + React Native Testing Library | | ui | NativeWind / Gluestack UI | | auth | Supabase Auth | | analytics | Segment | | crash | Sentry |

🔍 Code Review

| Tool | What it does | Say | |------|-------------|-----| | review_branch | Deep code review of current branch vs main. Detects RN issues, scores merge risk (LOW/MEDIUM/HIGH), lists must-fix and should-fix items, and automatically checks which changed files are missing unit tests | "review my code" / "review my branch before PR" | | compare_with_branch | Merge readiness report: changed files, native changes (rebuild?), dependency changes, config changes, files by risk level, commit list | "compare with main" / "what did I change?" | | check_breaking_changes | What could break on merge: major package bumps, deleted files, type changes, nav route changes, native code, service/store changes | "will this break anything?" / "is it safe to merge?" | | detect_rn_issues | Scan a file or full branch diff for RN anti-patterns | "scan for RN issues in LoginScreen.tsx" / "detect issues in my branch" |

What detect_rn_issues checks:

| Severity | Category | Examples | |----------|----------|---------| | CRITICAL | Navigation | Untyped navigation calls (raw string routes) | | CRITICAL | Async | AsyncStorage without await, async setState after unmount | | CRITICAL | Promises | .then() without .catch() | | HIGH | Hooks | useEffect with empty deps (stale closure), direct API in useEffect | | HIGH | Lists | FlatList without keyExtractor | | HIGH | Logs | console.log/debug/info/verbose left in code | | HIGH | Errors | Empty catch blocks, JSON.parse without try-catch | | HIGH | UX | Missing loading/error UI, async onPress without disabled state | | HIGH | Placement | API calls directly in screen files | | HIGH | Libraries | Raw fetch() instead of axios client, full lodash import, moment.js | | MEDIUM | Styles | Inline style objects, hardcoded colors | | MEDIUM | Types | TypeScript any type used | | MEDIUM | Reuse | Magic numbers, duplicated string literals | | MEDIUM | State | Multiple setState calls in one handler | | MEDIUM | Lists | Missing empty state in FlatList/ScrollView | | LOW | Typos | 30+ common spelling mistakes in identifiers/strings | | LOW | Logic | Hardcoded booleans, unreachable code after return | | LOW | Debt | TODO/FIXME comments |

review_branch automatically includes a Unit Test Coverage check:

• Scans every changed .ts/.tsx/.js/.jsx file for a matching test file
• Reports how many changed files have tests vs are missing tests
• Flags CRITICAL if none of the changed files have any tests
• Suggests generate_unit_tests inline if coverage gaps are found — no extra command needed

🚀 Pre-PR Check — Merge Prediction & Risk Scoring

by Devashish Kukade

Run pre_pr_check before raising a pull request or to review an existing PR. It is entirely read-only — the working tree is never touched.

| Tool | What it does | Say | |------|-------------|-----| | pre_pr_check | Predict merge outcome, score risk 0–100, map feature impact, list conflicts | "pre pr check into dev" / "check before I raise a PR" / /pre-pr dev | | pre_pr_check (PR mode) | Same checks on an existing PR — pass the link | "review this PR: https://github.com/org/repo/pull/42" / /review-pr <link> |

The tool reports four sections in priority order:

1. Mergeable

Will the branch merge cleanly into the target? Uses git merge-tree --write-tree (requires git ≥ 2.38) to predict the merge in memory without touching any files.

2. Risk Score — LOW / MEDIUM / HIGH (0–100)

A deterministic score where every point is explained. Nine weighted drivers, each capped so no single signal overwhelms:

| Driver | Points | Cap | What it catches | |--------|--------|-----|-----------------| | Native code | 25 | 25 | iOS/Android build inputs (Podfile, .gradle, Info.plist, AndroidManifest) — forces a clean rebuild | | Shared fan-in | 2/dep | 20 | Shared files imported by 5+ other files — a one-line change that breaks six screens | | Dependency bumps | 12/major | 25 | Major version bumps in package.json — APIs break by definition | | Test gap | 3/file | 18 | Changed source files with no matching test file (disabled if project has < 3 test files) | | Migration/config | 15 | 15 | Schema migrations, CI workflows, .env files, app config, build config | | Merge conflicts | 2/file | 15 | Files that will not auto-merge | | Staleness | 1/10 commits | 10 | How far behind the target branch this branch has drifted | | Fragile paths | 5/file | 10 | Files that have been reverted or hot-fixed in recent history | | Churn | 1/150 lines | 10 | Total lines changed (generated/lockfiles excluded) |

| Score | Band | |-------|------| | 0–29 | LOW | | 30–59 | MEDIUM | | 60–100 | HIGH |

Each driver includes evidence (file paths, importer counts, version bumps) and actionable mitigations.

3. Features Impacted

Maps changed files to the project's feature structure (auto-detected from src/domain, src/features, src/modules, or src/screens):

• Direct — changed file lives inside a feature folder
• Indirect — changed file is shared code (services, redux, navigation); its importers are traced one hop to find which features are affected
• App-wide — shared code imported by 50+ files (reported separately)

Uses a reverse-import index built from the project's source files, honouring tsconfig path aliases.

4. Conflicts

Exact list of files that will not auto-merge — fix these on your branch before opening the PR.

Requirements:

• Git ≥ 2.38 (for git merge-tree --write-tree)
• Not a shallow clone (merge-base would be wrong — run git fetch --unshallow first)
• gh CLI installed and authenticated (only for PR-mode via URL)

🧪 Unit Tests

| Tool | What it does | Say | |------|-------------|-----| | generate_unit_tests | Generate test files for changed files vs main. Auto-detects Jest/Vitest/Mocha, places tests correctly, runs them immediately | "generate unit tests" / "generate tests for LoginScreen.tsx" | | check_test_coverage | Run full test suite with coverage, flag files below threshold | "check test coverage" / "what's my coverage?" | | run_tests | Run the test suite or a specific file | "run tests" / "run tests for LoginScreen" |

How generate_unit_tests works:

1. Finds all changed .ts/.tsx/.js/.jsx files vs main (or a specific file you name)
2. Auto-detects your test framework from package.json (Jest → Vitest → Mocha)
3. Checks if a test file already exists — skips if so
4. Generates a test file with: render tests, snapshot, interaction tests (from testIDs), async tests, hook tests, utility function tests
5. Runs the generated tests immediately and shows pass/fail
6. Asks you to fill in the TODO sections with real expected values

Test file placement — auto-detected from project structure:

• Co-located: LoginScreen.test.tsx next to LoginScreen.tsx
• Or in __tests__/ folder if that pattern exists in the project

Coverage thresholds (check_test_coverage): | Coverage | Status | |----------|--------| | 100% | ✅ Full | | 80–99% | 🟡 Partial | | 50–79% | 🟠 Low | | < 50% | 🔴 CRITICAL — generate tests |

🔧 Build & Branch Sync Rules

by Shekhar Manwar

The agent never builds your project. No npm run build, expo prebuild, pod install, xcodebuild, gradlew, or run-ios/run-android — when a build or native rebuild is needed it tells you and hands you the exact command. Running tests (run_tests, generate_unit_tests, check_test_coverage) is not a build and still happens automatically after development.

The agent checks remote sync before it starts. At session start, right after activation, it runs check_branch_sync:

| Result | What the agent does | |--------|--------------------| | In sync | Says the branch is up to date, then continues | | Behind / diverged | Tells you to run git pull first and makes no edits until you have | | No upstream | Says so and continues |

It never runs git pull, git merge, or git rebase itself — only a read-only fetch of remote-tracking refs.

🛡 Change Safety Protocol

Every file change Claude makes goes through a mandatory safety flow:

Before every edit

Claude states the risk level inline and proceeds immediately — no stopping:

🔍 Risk: 🟡 MEDIUM — modifying WorkflowSkill.js end_of_day_report output format

| Level | When | |-------|------| | 🟢 LOW | Docs, comments, README, non-functional text | | 🟡 MEDIUM | Logic in one skill/tool, new optional param, new file | | 🔴 HIGH | Tool signature change, shared service, index.js, package.json, CI/CD |

After all edits are done

Claude provides a full impact summary:

• What changed and why
• What stays the same
• Side effects on other tools
• Automated test steps (npm run validate, npm test)
• Manual test steps (exact Claude CLI phrases to verify)
• Test cases table (happy path, edge case, failure case)

Then asks: "Save to TESTING.md? And shall I commit?" — and never commits without your explicit yes.

This protocol is enforced via PreToolUse and PostToolUse hooks installed into ~/.claude/settings.json during setup.

Slash Commands

Slash commands are auto-installed to ~/.claude/commands/ when you run npx mobile-team-agent setup. New commands from v4.0.0 use the mta: prefix to avoid conflicts with user commands.

New in v4.0.0 (use /mta: prefix):

| Command | What it does | Author | |---------|-------------|--------| | /mta:pre-pr <branch> | Run pre_pr_check — first word is the branch to merge INTO | Devashish Kukade | | /mta:review-pr <link> | Run pre_pr_check on an existing pull request URL | Devashish Kukade | | /mta:analyze <request> | Validate a vague request before implementing | Shekhar Manwar | | /mta:narrow <feature> | Find the smallest relevant set of files for a feature or symbol | Shekhar Manwar | | /mta:prompt <task> | Show the standard prompt template | Shekhar Manwar |

Existing commands (no prefix change):

| Command | What it does | |---------|-------------| | /standup | Morning standup | | /eod | End-of-day report | | /plan | Deep daily plan | | /tickets | List open tickets | | /ticket <key> | Select a ticket | | /commits | Recent git commits | | /remember / /recall | Memory save/search | | /journal | Work log entry | | /decide / /decisions | Track decisions | | /suggest | AI-scored ticket suggestions | | /weekly | Weekly summary | | /workload | Workload analysis | | /health | Integration health check | | /status | Setup status | | /memory | Memory usage stats |

Context Engineering (Optional Add-on)

by Aatif

The context-engineering/ folder contains an advanced Claude Code workflow system for long-running sessions. It is self-contained and can be adopted independently.

What it provides:

• Slash commands: /checkpoint, /context, /handoff, /resume-handoff, /unknowns
• Hooks: git-guard.sh (blocks destructive git ops), pre-compact.sh, session-start.sh, statusline.js/sh (live context usage %), turn-context.js/sh
• Skills: context-handoff — structured save/resume of task state across sessions
• Docs: 8-chapter guide (overview, getting started, commands, context engineering, git safety, model safety, reference, troubleshooting)

Install:

cd context-engineering
chmod +x install.sh && ./install.sh

Key concepts: Checkpoints (mid-task state save), Context Handoffs (persist across sessions), Unknowns tracking (unresolved decisions surfaced via /unknowns), Context budget bands (under 60% normal, 80%+ checkpoint recommended, 90%+ handoff strongly recommended), Git safety hooks.

Team rules in context-engineering/CLAUDE.md: no assumptions (label certainty as Known/Found/Inferred/Unknown), understand before you change, follow existing patterns, verify (never fake certainty), keep context small.

Natural Language Examples

# Morning
"Good morning"                            -> morning standup + last session context
"plan my day"                             -> deep daily plan

# Tickets
"show my tickets"                         -> list open tickets
"PROJ-42"                                 -> full details + implementation plan
"what should I work on?"                 -> AI-scored suggestions
"move PROJ-42 to In Progress"            -> transition status

# Git
"show my recent commits"                 -> git log with Jira links
"am I up to date?"                       -> check branch sync status

# Pre-PR & Merge Safety
"pre pr check into dev"                  -> merge prediction + risk score
"review this PR: <github-link>"         -> review an existing PR by URL

# AI Efficiency
"check my prompt"                        -> analyze request completeness
"where is end_of_day_report"            -> narrow code scope

# Figma
"build the Login screen from Figma"     -> read design + generate code

# Code Review
"review my branch"                       -> full review with risk score
"scan LoginScreen.tsx for RN issues"    -> file-level issue scan

# Unit Tests
"generate unit tests"                    -> tests for all changed files

# Memory
"remember: auth token stored in MMKV"
"we decided to use Zustand"              -> saved decision

# End of day
"end of day"                             -> EOD report + session snapshot saved

📁 Project Structure

Mobile Team Agent/
├── Main/
│   ├── index.js              # MCP server entry point
│   └── SkillRegistry.js      # Tool registration
├── Skills/
│   ├── Core/BaseSkill.js
│   ├── EfficiencySkill.js    # analyze_request, narrow_code_scope, prompt_template
│   ├── SetupSkill.js         # invoke, get_setup_status, configure_service, health_check
│   ├── JiraReadSkill.js      # list_tickets, get_ticket_details, analyze_workload, ...
│   ├── JiraWriteSkill.js     # transition_ticket, add_comment, create_ticket, ...
│   ├── GitSkill.js           # get_recent_commits, get_commit_details, check_branch_sync
│   ├── WorkflowSkill.js      # morning_standup, plan_my_day, end_of_day_report, ...
│   ├── FigmaSkill.js         # configure_figma, list_figma_screens, read_figma_screen, ...
│   ├── MemorySkill.js        # remember, recall, journal, add_decision, ...
│   ├── CodeReviewSkill.js    # review_branch, detect_rn_issues, pre_pr_check, ...
│   ├── RNProjectSkill.js     # setup_rn_project, analyze_rn_architecture, recommend_libraries
│   ├── UnitTestSkill.js      # generate_unit_tests, check_test_coverage, run_tests
│   └── prompts/              # Markdown prompt templates per skill
├── Services/
│   ├── jira-client.js
│   ├── figma-client.js
│   ├── config-manager.js
│   ├── memory-manager.js     # includes session snapshot save/load
│   ├── report-manager.js
│   └── offline-queue.js
├── Constants/constants.js    # version, risk weights, risk bands, timeouts
├── Utils/
│   ├── git-utils.js          # commit analysis + pre-PR merge plumbing + branch sync
│   ├── feature-map.js        # reverse-import index, feature impact analysis
│   ├── gh-utils.js           # GitHub CLI (gh) wrapper for PR metadata
│   ├── risk-model.js         # deterministic risk scoring (pure, no I/O)
│   ├── ticket-utils.js
│   └── validators.js
├── commands/                  # Slash commands (installed to ~/.claude/commands/ by setup)
│   ├── standup.md, eod.md, plan.md, tickets.md, ...  # existing
│   ├── mta:pre-pr.md         # /mta:pre-pr — pre-PR merge check
│   ├── mta:review-pr.md      # /mta:review-pr — review existing PR
│   ├── mta:analyze.md        # /mta:analyze — request validation
│   ├── mta:narrow.md         # /mta:narrow — code scope narrowing
│   └── mta:prompt.md         # /mta:prompt — prompt template
├── setup.js                  # npx mobile-team-agent setup entry point
├── install.sh                # Clone-based installer
├── package.json
└── CLAUDE.md                 # Agent instructions for Claude

.claude/commands/
├── pre-pr.md                 # /pre-pr slash command
├── review-pr.md              # /review-pr slash command
├── analyze.md                # /analyze slash command
├── narrow.md                 # /narrow slash command
└── prompt.md                 # /prompt slash command

context-engineering/           # Optional add-on (see Context Engineering section)
├── .claude/commands/          # /checkpoint, /context, /handoff, /resume-handoff, /unknowns
├── .claude/hooks/             # git-guard, pre-compact, session-start, statusline, turn-context
├── .claude/skills/            # context-handoff skill
├── docs/guide/                # 8-chapter documentation
├── install.sh                 # Self-contained installer
├── CLAUDE.md                  # Team engineering rules
└── README.md

📊 Storage Layout

~/Documents/MobileTeamAgent/
├── MyApp/
│   ├── 31-07-2026_updates.md       ← EOD report
│   ├── 30-07-2026_updates.md
│   └── session_snapshot.md         ← next-day context (auto-updated)
├── ShopApp/
│   ├── 31-07-2026_updates.md
│   └── session_snapshot.md

invoke_mobile_team scans this folder on startup and surfaces all projects with their latest report date and session snapshot.

🛠 Troubleshooting

Agent not responding to tools?

npx mobile-team-agent setup   # re-register + refresh CLAUDE.md
claude mcp list               # verify registration

CLAUDE.md not in my project?

cd /path/to/your/project
npx mobile-team-agent setup   # writes CLAUDE.md into current directory

Jira not connecting?

"health check"                # run health_check tool
"configure Jira"              # re-run configure_service

Hook blocking Claude mid-task?

npx mobile-team-agent setup   # updates hooks to non-blocking version

Manual MCP registration:

claude mcp add mobile-team-agent -- node $(npm root -g)/mobile-team-agent/Main/index.js

🤝 Contributing

The repo is source-only. Install from npm or clone and run the installer. Do not commit node_modules/, dist/, .env, or .DS_Store.

To test locally:

npm run validate    # syntax-check all source files
npm start           # run the MCP server directly