Skip to content

GuillemRoca/agent-skills-android / skills/using-agent-skills

using-agent-skills

Verified

Use when starting a session or selecting a skill for a task. Meta-skill defining five core agent behaviors: assumption surfacing, confusion management, constructive pushback, simplicity enforcement, and scope discipline. Includes the skill discovery flowchart for all 30 Android skills.

Using Agent Skills (Android)

Overview

This is the meta-skill — it defines how agents should work, not what they should build. Five core operating behaviors ensure agents produce reliable, high-quality Android code that matches project conventions and user intent.

Five Core Operating Behaviors

1. Surface Assumptions Explicitly

Before implementing non-trivial work, state your assumptions:

  • "I'm assuming we're using MVVM, not MVI, for this screen"
  • "I'm assuming minSdk 26 since that's in build.gradle.kts"
  • "I'm assuming offline-first since the project uses Room as source of truth"

Why: Unstated assumptions are the #1 cause of wasted agent work. A wrong assumption in step 1 makes steps 2–10 wrong.

2. Manage Confusion

When encountering inconsistencies, stop and clarify:

  • "The spec says use DataStore but the existing code uses SharedPreferences — which should I follow?"
  • "TaskRepository has two different patterns — which is the intended one?"
  • Name the specific confusion. Don't proceed with a guess.

3. Push Back When Warranted

Don't default to agreement. When a request has concrete downsides, say so:

  • "Adding that dependency would increase APK size by ~2MB for a feature used by <1% of users"
  • "Skipping tests here is risky because this ViewModel handles payment state"
  • Provide the tradeoff, let the human decide.

4. Enforce Simplicity

Before writing code, ask: "Can this be done with less?"

  • Resist unnecessary abstractions
  • Don't add features beyond what's requested
  • Three similar lines of code is better than a premature abstraction
  • Follow project conventions, don't introduce new patterns

5. Maintain Scope Discipline

Only modify what's requested:

  • Don't refactor neighboring code "while you're here"
  • Don't add error handling for impossible scenarios
  • Don't add comments to code you didn't change
  • Don't upgrade dependencies unrelated to the task

Skill Discovery Flowchart

Match your current task to the right skill:

What are you trying to do?
│
├─ "I have a vague idea"
│   └─ idea-refine → spec-driven-development
│
├─ "Requirements are unclear or contradictory"
│   └─ interview-me → spec-driven-development
│
├─ "I need to write a spec"
│   └─ spec-driven-development
│
├─ "No quality bar is written down" / "the agent keeps silencing checks"
│   └─ constraint-driven-development
│
├─ "I need to plan implementation"
│   └─ planning-and-task-breakdown
│
├─ "I need to build a feature"
│   ├─ Architecture decisions? → android-architecture
│   ├─ UI with Compose? → android-ui-engineering
│   ├─ Data persistence? → android-data-persistence
│   ├─ API/interface design? → api-and-interface-design
│   ├─ Work outside the visible UI (sync, uploads)? → android-background-work
│   ├─ High-stakes, hard-to-reverse decision? → doubt-driven-development
│   └─ General implementation → incremental-implementation + test-driven-development
│
├─ "I need to test"
│   ├─ Unit/integration tests → test-driven-development
│   ├─ Device/UI tests → android-device-testing
│   ├─ End-to-end proof on a device → android-e2e-verification
│   └─ Accessibility testing → android-accessibility
│
├─ "I need to review code"
│   ├─ Code quality → code-review-and-quality
│   ├─ Security review → security-and-hardening
│   └─ Performance review → performance-optimization
│
├─ "I need to debug"
│   └─ debugging-and-error-recovery
│
├─ "I need to simplify code"
│   └─ code-simplification
│
├─ "I need to ship"
│   ├─ CI/CD setup → ci-cd-and-automation
│   ├─ Version/release → git-workflow-and-versioning
│   ├─ Logging, crash reporting, monitoring → observability-and-instrumentation
│   └─ Launch checklist → shipping-and-launch
│
├─ "I need to migrate/deprecate"
│   └─ deprecation-and-migration
│
├─ "I need to write docs"
│   └─ documentation-and-adrs
│
├─ "I need to set up context for AI"
│   └─ context-engineering
│
├─ "I need to check official docs"
│   └─ source-driven-development
│
└─ "I'm not sure"
    └─ Start with idea-refine or context-engineering

Quick Reference: All 30 Skills

PhaseSkillTrigger
DEFINEinterview-meRequirements vague or contradictory
DEFINEidea-refineVague idea needs sharpening
DEFINEspec-driven-developmentNeed a spec before coding
DEFINEcontext-engineeringSetting up AI context for project
DEFINEconstraint-driven-developmentWriting down and guarding the quality bar
PLANplanning-and-task-breakdownBreaking work into tasks
PLANandroid-architectureArchitecture decisions
BUILDincremental-implementationBuilding feature step by step
BUILDtest-driven-developmentWriting tests before code
BUILDandroid-ui-engineeringBuilding Compose UI
BUILDandroid-data-persistenceRoom, DataStore, offline-first
BUILDandroid-background-workWorkManager, foreground services, alarms
BUILDapi-and-interface-designDesigning interfaces and contracts
BUILDsource-driven-developmentUsing official docs for framework code
BUILDdoubt-driven-developmentStress-testing a high-stakes decision
BUILDcode-simplificationMaking code easier to understand
BUILDdocumentation-and-adrsWriting docs and ADRs
VERIFYandroid-device-testingInstrumented tests, emulator, ADB
VERIFYandroid-e2e-verificationMaestro end-to-end flows on device
VERIFYandroid-accessibilityAccessibility compliance
VERIFYdebugging-and-error-recoveryFixing bugs systematically
VERIFYperformance-optimizationMeasuring and improving performance
REVIEWcode-review-and-qualityReviewing code quality
REVIEWsecurity-and-hardeningSecurity review and hardening
SHIPci-cd-and-automationCI/CD pipeline setup
SHIPgit-workflow-and-versioningGit workflow and versioning
SHIPshipping-and-launchRelease checklist and rollout
SHIPobservability-and-instrumentationLogging, Crashlytics, release health
SHIPdeprecation-and-migrationDeprecating and migrating code
METAusing-agent-skillsHow agents should operate

Skills Are Workflows, Not Suggestions

  • Follow steps in order — each step depends on the previous
  • Never skip verification — "it looks right" is not evidence
  • Produce tangible artifacts — specs, task lists, test results, not just prose
  • Anti-rationalizations are real — if you're about to skip a step, check the "Common Rationalizations" table
  • Assumption-making without validation is the most common failure mode

Verification

  • Assumptions stated explicitly before implementation
  • Confusion surfaced (not silently resolved with guesses)
  • Pushback provided when warranted (with tradeoffs)
  • Simplicity enforced (no unnecessary abstractions)
  • Scope maintained (only requested changes made)
  • Correct skill loaded for the current task
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 468kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 321kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 321kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 321kUpdated Claude CodeCodexVerifiedInstall