agy-skills Documentation
Welcome to the documentation for agy-skills, agent skills for Google Antigravity (AGY) maintained by Fred de Ruiter.
| Skills Index | Workflow Topology | AI Coding Dictionary |
AI Coding Dictionary
Explore the comprehensive AI Coding Dictionary, the shared vocabulary for agentic software development, context management, prompt engineering, and tool execution.
- The Model: Parameters, inference, effort, tokens, next-token prediction, harnesses.
- Sessions, Context Windows & Turns: Context, window limits, stateful vs stateless execution, system prompts.
- Tools & Environment: Tool calls, MCP servers, permissions, sandboxing.
- Failure Modes: Sycophancy, hallucination, parametric knowledge, attention degradation, smart zone limits.
- Handoffs: Clearing, handoffs, compaction, specs, tickets, primary vs secondary sources.
- Memory & Steering: AGENTS.md, progressive disclosure, context pointers, skills, subagents.
- Patterns of Work: Human-in-the-loop (HITL), AFK execution, automated checks, grilling, prototyping, vibe coding.
Workflow Topology
The Main Flow: Idea to Ship
The primary pipeline taken by new features, architectural refactors, and planned enhancements.
flowchart TD
Start(["New Idea or Feature Requirement"]) --> Grill["/grill-with-docs\n(Interview, CONTEXT.md, ADRs)"]
Grill --> PrototypeCheck{"Can every design\nquestion be settled\nin conversation?"}
PrototypeCheck -- "No (Needs runnable validation)" --> HandoffOut["/handoff (Out)"]
HandoffOut --> Proto["/prototype\n(Throwaway code on prototype/* branch)"]
Proto --> HandoffIn["/handoff (Back)"]
HandoffIn --> ScopeCheck
PrototypeCheck -- "Yes" --> ScopeCheck{"Is this a multi-session\nor multi-ticket build?"}
ScopeCheck -- "Yes" --> ToSpec["/to-spec\n(Create formal specification)"]
ToSpec --> ToTickets["/to-tickets\n(Tracer bullets with blocking edges)"]
ToTickets --> ExecChoice{"Execution Strategy"}
ExecChoice -- "Concurrent Subagents" --> ImplementSpec["/implement-spec\n(Orchestrate subagents on branch)"]
ExecChoice -- "Focused Sessions" --> ImplementTickets["/implement\n(One ticket per clean session)"]
ScopeCheck -- "No" --> ImplementDirect["/implement\n(Same session implementation)"]
ImplementSpec --> TDD["/tdd\n(Red-green-refactor slices)"]
ImplementTickets --> TDD
ImplementDirect --> TDD
TDD --> CodeReview["/code-review\n(Two-axis review: Standards + Spec)"]
CodeReview --> Ship(["Commit & Merge to Main"])
On-Ramps onto the Main Flow
Structured entry points (bug triage, regressions, or massive greenfield initiatives) that prepare work before merging onto the main flow.
flowchart TD
subgraph TriageQueue ["Incoming Issues & Requests"]
RawIssues["Incoming Bug Reports / Requests"] --> Triage["/triage\n(Classify, assign roles, label)"]
Triage --> AgentReady["Agent-Ready Tickets\n(ready-for-agent)"]
end
subgraph BugDiagnosis ["Difficult Regressions & Defects"]
BugFound["Intermittent Flake or Complex Bug"] --> Diag["/diagnosing-bugs\n(Tight feedback loop & regression test)"]
Diag --> DiagPostMortem["Post-Mortem Findings"]
end
subgraph Wayfinding ["Large / Ambiguous Greenfield Efforts"]
FoggyIdea["Massive Greenfield Effort\n(Too big for single session)"] --> Wayfinder["/wayfinder\n(Chart decision map on tracker)"]
Wayfinder --> DecisionTickets["Resolve Decision Tickets\n(Decisions, not deliverables)"]
DecisionTickets --> ClearMap["Clear Strategic Map"]
end
AgentReady --> Implement["/implement (Main Flow)"]
DiagPostMortem -- "Architectural Seam Missing" --> ImproveArch["/improve-codebase-architecture"]
DiagPostMortem -- "Direct Bug Fix" --> CodeReview["/code-review (Main Flow)"]
ClearMap --> ToSpec["/to-spec (Main Flow)"]
Codebase Upkeep & Architecture Health
Continuous improvement loops that maintain codebase seams and tune agent instructions over time.
flowchart TD
TriggerArch(["Periodic Upkeep or Refactoring"]) --> ImproveArch["/improve-codebase-architecture\n(Scan for shallow modules and missing seams)"]
ImproveArch --> Candidates["Deepening Opportunities"]
Candidates --> CodebaseDesign["/codebase-design\n(Apply deep-module vocabulary and seam patterns)"]
CodebaseDesign --> MainFlow["Feed into /grill-with-docs"]
TriggerRetro(["Session Complete or Milestone"]) --> Retro["/retro\n(Analyze conversation transcripts)"]
Retro --> AuditResults["Audit Transcripts across 7 Categories"]
AuditResults --> EnvUpdates["Update AGENTS.md, Rules,\nAutomated Checks, & Navigation"]
Writing and Thought Shaping
Productivity workflows for mining raw fragments, shaping arguments paragraph by paragraph, and assembling narrative journeys.
flowchart LR
RawThoughts["Raw Unstructured Thoughts or Voice Memos"] --> WriteFrag["/writing-fragments\n(Mine raw notes into single quarry file)"]
WriteFrag --> Quarry["Structured Quarry File"]
Quarry --> WriteShape["/writing-shape\n(Shape article block by block,\ndebate format choices)"]
WriteShape --> Draft["Structured Draft"]
Draft --> WriteBeats["/writing-beats\n(Assemble narrative journey of beats,\nground concepts before leaning on them)"]
WriteBeats --> FinalProse(["Finished Article or Doc"])
Standalone Utilities Matrix
flowchart TD
subgraph StandaloneSkills ["Standalone Utilities"]
GrillMe["/grill-me\n(Stateless interview outside a repository)"]
Grilling["/grilling\n(Reusable core interview engine)"]
Conflicts["/resolving-merge-conflicts\n(Intent-traced hunk resolution)"]
Research["/research\n(Background research on primary sources)"]
Questionnaire["/to-questionnaire\n(Asynchronous survey for stakeholders)"]
Wizard["/wizard\n(Interactive script for human-only operational steps)"]
WaitWhat["/wait-what\n(Mid-session re-explanation and vocabulary alignment)"]
Teach["/teach\n(Multi-session learning workspace)"]
WritingForAgents["/writing-for-agents\n(Reference for authoring agent documents)"]
end
subgraph EnvironmentSetup ["Environment & Preconditions"]
SetupAgy["/setup-agy-skills\n(Configure issue tracker and labels)"]
SetupMcp["/setup-mcp\n(Configure and verify MCP servers)"]
SetupDeep["/setup-ts-deep-modules\n(Enforce deep module boundaries)"]
SetupPyDeep["/setup-python-deep-modules\n(Enforce Python import-linter boundaries)"]
NewSkill["/new-skill\n(Scaffold new skill & doc page)"]
end
Engineering Skills
User-invoked
- ask-fred: Router over the skills in this repository.
- grill-with-docs: Grilling session that builds your domain model in CONTEXT.md and ADRs.
- triage: Move issues through a state machine of triage roles.
- improve-codebase-architecture: Scan codebase for deepening opportunities.
- setup-agy-skills: Configure a repo for the engineering skills.
- setup-mcp: Configure Model Context Protocol (MCP) servers in Google Antigravity.
- setup-ts-deep-modules: Enforce deep module boundaries with dependency-cruiser.
- to-spec: Turn conversation into a spec and publish to issue tracker.
- to-tickets: Break specs/plans into tracer-bullet tickets with blocking edges.
- implement: Build work described by spec/tickets test-first and review before committing.
- implement-spec: Orchestrate concurrent subagents to implement a full specification across its ticket task graph.
- wayfinder: Chart and resolve large efforts as decision tickets.
- retro: Session retrospective and transcript audit for environment tuning.
Model-invoked
- prototype: Build throwaway prototypes to answer design questions.
- diagnosing-bugs: Feedback-loop driven debugging.
- research: Investigate questions against primary sources via background agent.
- tdd: Test-driven development with red-green-refactor loop.
- domain-modeling: Build and maintain project domain model and ADRs.
- codebase-design: Deep module and clean seam design discipline.
- code-review: Two-axis review (Standards + Spec) of diffs.
- resolving-merge-conflicts: Intent-traced git conflict resolution.
- wizard: Interactive wizard script for human-only operational steps.
Productivity Skills
User-invoked
- grill-me: Relentless interview to resolve design decisions.
- handoff: Compact conversation into handoff document for another session/agent.
- teach: Stateful multi-session learning workspace.
- to-questionnaire: Turn decisions into questionnaires for external stakeholders.
- wait-what: Clarify and re-pitch misunderstood agent responses.
- writing-beats: Assemble raw writing fragments into a narrative journey of beats.
- writing-fragments: Mine raw heterogeneous writing fragments into a quarry document.
- writing-shape: Shape raw writing fragments into a structured article paragraph by paragraph.
Model-invoked
- grilling: Reusable interview primitive.
- writing-for-agents: Discipline for authoring agent-readable documents.