Daniel Perussina
Community@dperussina · Unknown
Design-review skills for building LLM agent stacks: tool synthesis, MCP export surfaces, sandboxing, memory tiers, and Spec Kit feature-specification phases.
Agent Skills by Daniel Perussina
Showing 28 vetted skills indexed across 1 GitHub repositories.
dperussinaspeckit-checklist
Generate requirements-quality checklists for spec-kit feature specifications.
dperussinatool-synthesis-from-code
Guides curation of analyzed source code into consolidated, outcome-named LLM tool sets.
dperussinamcp-export-design
Designs dual-era MCP server export surfaces projected from a typed internal tool IR.
dperussinacontract-derived-verification
Derives agent verification signals from code contracts instead of model self-assessment.
dperussinaagent-tool-design
Reviews and designs LLM tool schemas, descriptions, error messages, and return shapes.
dperussinaspeckit-constitution
Create or update the project constitution from interactive or provided principle inputs.
dperussinaspeckit-implement
Execute implementation plans by processing tasks defined in tasks.md.
dperussinaprovider-abstraction
Defines boundaries for model-provider abstraction layers and opaque continuation state handling.
dperussinaspeckit-converge
Assess codebase against spec, plan, and tasks, then append remaining work to tasks.md.
dperussinaspeckit-taskstoissues
Convert spec-kit tasks into dependency-ordered GitHub issues with deduplication.
dperussinacontext-engineering
Budgets and manages which tokens enter an agent's context window across prompts, retrieval, and memory.
dperussinagraph-vs-loop-decision
Decides between bare loops, state machines, and graphs for agent control flow.
dperussinaspeckit-clarify
Detects underspecified areas in feature specs and records targeted clarification answers.
dperussinaharness-selection
Decides whether to adopt an agent framework or build the harness for agent runtimes.
dperussinaspeckit-plan
Generates implementation plans and design artifacts from feature specifications using the spec-kit plan template.
dperussinaspeckit-analyze
Analyze spec.md, plan.md, and tasks.md for inconsistencies, gaps, and constitution violations.
dperussinaexperiment-design
Designs falsifiable experiments and validation benchmarks for agent systems with pre-registered kill criteria.
dperussinaagent-safety-and-sandboxing
Classifies agent tool calls into effect tiers and audits sandbox configurations for exfiltration risk.
dperussinamulti-agent-topology-review
Reviews multi-agent architecture proposals against a five-gate test with cost and failure-rate evidence.
dperussinaintegration-surface-design
Designs HTTP/SSE and embeddable-iframe agent surfaces as separate capability tiers.
dperussinaspeckit-specify
Generate a validated feature specification from a natural language feature description.
dperussinaresearch-doc-conventions
Applies house style conventions to research documents in the research/ directory.
dperussinacredential-and-env-injection
Designs credential injection and secret isolation for generated agent runtimes and synthesized tools.
dperussinaknowledge-graph-memory
Designs memory tiers, substrates, and hygiene rules for agent knowledge stores.
Frequently Asked Questions About Daniel Perussina
FAQPage SchemaWhat tasks can I accomplish with dperussina's skills?▼
You can design and review LLM agent systems: curate tool sets from source code, design MCP export surfaces, choose between loops and graph control flow, budget context windows, place sandboxing and credential-injection boundaries, and run Spec Kit phases (specify, plan, tasks, implement, analyze) for feature development.
Who are these skills intended for?▼
Engineers and architects building or reviewing LLM agent stacks — people writing tool schemas and MCP servers, evaluating frameworks like LangGraph or the OpenAI Agents SDK, designing multi-agent topologies, or driving specification-first feature development with GitHub's Spec Kit templates.
What prerequisites do the Spec Kit skills require?▼
The speckit-* skills require an existing spec-kit project structure with a .specify/ directory, as stated in their compatibility frontmatter. The agent-design skills (tool synthesis, sandboxing, memory, context) are design-review guidance and carry no stated runtime dependency.
How do the safety and credential skills work in practice?▼
agent-safety-and-sandboxing classifies effects into read-only, reversible-write, and irreversible-destructive tiers, then places isolation (container, gVisor, microVM) and human approval gates. credential-and-env-injection keeps secrets out of model context via credential_ref handles, brokers, and redaction in traces and memory.
Do these skills cover multi-agent decomposition?▼
Yes, but as v2-scoped design review. codebase-decomposition and tool-synthesis-from-code explicitly describe v2 planning only; multi-agent-topology-review challenges subagent, swarm, and orchestrator proposals with cost and failure-rate evidence, arguing against splitting a single agent without a named, measured constraint.