Repository navigation
Mobile Engineering Loop
The README overview shows the big picture. This guide explains what happens inside each stage and what evidence is needed to move forward. This is a guided agent workflow, not a claim that one command automatically runs every stage. SDKs, model access, design connections and device tools must be available for the selected checks.
flowchart TD
start(["Start a project or resume a session"])
subgraph planning["1. Understand and plan"]
context["Read repository instructions and saved context"]
inventory["Inspect code, architecture and available tools"]
reference["Identify requirements, design references and API contracts"]
task["Choose a matching issue or define a focused task"]
criteria["Record scope, risks and acceptance criteria"]
agreed{"Required product decisions resolved?"}
clarify["Clarify scope or obtain the missing decision"]
end
subgraph implementation["2. Implement one feature"]
branch["Use a focused branch and reuse existing components"]
build["Implement UI, state, navigation and API integration"]
states["Handle loading, empty, error and relevant offline states"]
checks["Run the build and relevant automated tests"]
end
subgraph verification["3. Audit and verify"]
audit["Audit architecture, code quality and model separation"]
security["Review authentication, secrets, storage and permissions"]
ui["Verify design match, layouts and accessibility"]
device["Run the user journey on a device or simulator"]
integration["When relevant, compare actual mobile API traffic"]
performance["Check relevant startup, memory and rendering risks"]
end
subgraph evidence["4. Report and preserve context"]
report["Record passed, failed and unverified checks with evidence"]
save["Save decisions, changed files, blockers and next action"]
outcome{"Required acceptance criteria satisfied?"}
repair["Fix confirmed issues and rerun affected checks"]
unblock["Resolve missing prerequisites without claiming a pass"]
end
subgraph delivery["5. Review, release and learn"]
pr["Prepare a reviewable PR and implementation summary"]
prepare["Prepare release notes, rollback and rollout checks"]
approved{"Release explicitly approved?"}
hold(["Keep work ready and resume when needed"])
release["Use the configured release pipeline"]
monitor["Review feedback, crashes and performance"]
next["Prioritize the next useful improvement"]
end
start --> context --> inventory --> reference --> task --> criteria --> agreed
agreed -->|"Not yet"| clarify --> criteria
agreed -->|"Yes"| branch --> build --> states --> checks
checks --> audit --> security --> ui --> device --> integration --> performance
performance --> report --> save --> outcome
outcome -->|"Failed checks"| repair --> build
outcome -->|"Missing evidence"| unblock --> inventory
outcome -->|"Yes"| pr --> prepare --> approved
approved -->|"Not yet"| hold
approved -->|"Yes"| release --> monitor --> next --> context
hold -.->|"Resume later"| context
classDef plan fill:#dbeafe,stroke:#2563eb,color:#172554
classDef work fill:#ede9fe,stroke:#7c3aed,color:#2e1065
classDef proof fill:#dcfce7,stroke:#15803d,color:#14532d
classDef decision fill:#fef3c7,stroke:#b45309,color:#78350f
class context,inventory,reference,task,criteria,clarify plan
class branch,build,states,repair work
class checks,audit,security,ui,device,integration,performance proof
class report,save,pr,hold,monitor,next proof
class agreed,outcome,unblock,prepare,approved,release decision
| Stage | Useful output | Existing resources |
|---|---|---|
| Planning | Scoped requirements, design direction, tasks, risks and acceptance criteria | APPFORGE, Mobile Harness, /grill-mobile, /feature-slice
|
| Implementation | A focused code change that preserves the app's existing stack | Platform agents and feature workflows |
| Code audit | Findings tied to files, architecture boundaries and maintainability | Platform reviewers, /clean-code-audit
|
| Security review | Findings about auth, secrets, storage, permissions and sensitive data | SENTINEL, /security-audit
|
| UI verification | Design/state comparison and accessibility findings | Figma handoff, /accessibility-audit, /prd-verification
|
| Device QA | Actual journey results, screenshots and reproducible failures |
/mobile-mcp-qa, Device Proof Reports |
| API verification | Observed request/response evidence compared with the intended contract |
/web-network-to-mobile, capture and contract commands |
| Performance | Relevant startup, memory, rendering or network findings | PERF, /perf-audit and platform investigation skills |
| Reporting | What changed, checks run, evidence, deviations and open gaps | Mobile Harness and PRD verification reports |
| Context saving | Durable decisions, current task, blockers and next action | Mobile Memory and Mobile Flight Recorder |
| Release | Reviewable changes, release notes and controlled rollout |
/release-prep, PIPELINE, SCRIBE |
| Improvement | Follow-up issues grounded in feedback and operational evidence | Crash, performance and store-listing workflows |
- Passed: The relevant check ran and its evidence supports the acceptance criterion.
- Failed: A confirmed problem needs correction and an appropriate rerun.
- Unverified: A check did not run or evidence is missing. It cannot count as passed.
- Model output and generated code alone do not prove that an app works. Screenshots alone do not prove API payload parity.
- Save context when blocked or stopping mid-stage as well as at normal handoffs; recording a report must not depend on completing the entire loop.
- Ask for required product decisions, unavailable access and explicit authorization for consequential actions. Existing authorization carries forward; routine work should not cause repeated approval requests.
- Release is a separate approval gate. Passing checks does not itself authorize publishing, production deployment or store submission.
Keep the approved requirements, current task, changed files, executed checks and evidence paths together with unresolved findings and the next concrete action. Preserve repository-specific conventions: beginner starter docs, Mobile Memory and Flight Recorder are complementary resources, not mandatory duplicate copies of the same history.
Start with Mobile Harness, use the Core Loops guide to choose specialists, and follow device setup when runtime proof is needed. If a referenced tool or runtime is unavailable, report the gap and continue independent authorized work.
Getting Started
Documentation
- All Documentation
- Installation
- Figma MCP and Design Handoff
- Figma Spec Tools
- Core Loops
- Mobile Engineering Loop
- Mobile Memory
- Performance Loop
Reference
- Agents
- Platform Guide
- Skills and Prompts
- Android Device QA Setup
- Mobile App Design Skill
- Mobile Harness
- Enterprise Mobile Harness
- Workflows
- Reference
Community
Repository github.com/salmanashraf/mobile-ai-agents
Version: v1.0.38 · MIT License