Reactive Skills turns a long instruction file into a workflow your agent runs one step at a time. It sees only the current step, moves forward when a check passes, and leaves a run history you can inspect and resume.
RED_SPEC, passed
exit_code != 0
GREEN_CODE, running
exit_code === 0
REFACTOR, waiting
signal AUDIT_PASSED
AUDIT_VERIFY, waiting
exit_code === 0
COMPLETED, waiting
terminal
npx skills add Reactive-Skills/skills --skill tdd-refactorPublished workflows you can install today. Each one runs as explicit states, so your agent works one step at a time and moves on when a check passes. Checks counts the transitions that carry a guard.
~7% of a skill's instructions delivered at each step298 explicit states across 21 published workflowsEvery transition and failed check lands in a replayable run history
| Kind of check | Version | |||
|---|---|---|---|---|
| TDD RefactorRefactoring Red, green, and refactor run as separate states, and reported test exit codes gate each phase. | exit code | 11 | 6 | v2.0.6 |
| PR TriageCode review Collects PR metadata, classifies risk and ownership, checks readiness, then stops for a human decision. | payload guard | 9 | 13 | v1.0.3 |
| Release NotesReleases Scopes changes, classifies entries, drafts notes, validates formatting, and waits for approval. | exit code | 9 | 13 | v1.0.2 |
| Security ScanSecurity Scans staged changes for secrets and credentials, then hands back a remediation checklist. | exit code | 12 | 8 | v1.0.1 |
| API ContractGeneral OpenAPI/Swagger spec vs client code drift detector — parses spec, walks fetch/axios calls, detects mismatches | exit code | 16 | 11 | v1.0.2 |
| Browser VerifierGeneral Automated browser verification reactive skill that launches headless browser sessions, verifies DOM states, intercepts console errors, and captures visual artifacts. | payload guard | 10 | 2 | v1.0.2 |
| Build AdvisorGeneral Develop worthwhile products, challenge existing ideas and experiences, or advise builders on product, leadership, team, business, and career decisions using contextual principles from Tony Fadell's Build. | payload guard | 22 | 27 | v1.0.3 |
| CI/CD AutomationGeneral Automated CI/CD pipeline generator and linter that scaffolds production-ready workflows with caching, matrix builds, and security gates. | payload guard | 9 | 2 | v1.0.1 |
| Docs ArchitectMetaprogramming & Lifecycle Documentation architecture workflow that discovers source truth, designs information architecture, authors living documentation, embeds diagrams, validates links, and gates publication. | exit code | 11 | 17 | v1.0.3 |
| Experiment LoopGeneral Run bounded, evidence-backed improvement experiments with fixed evaluators, repeated baselines, protected confirmation, budgets, and recovery. | signal only | 17 | 0 | v1.0.2 |
| JSM WorkflowMetaprogramming & Lifecycle Reactive SDLC coordinator taking software changes from intake to final context sync, implementing the JS Mastery Engineering Workflow (https://jsmastery.com/skills). | payload guard | 20 | 3 | v1.5.1 |
| Llm Eval DesignGeneral Define measurable success criteria for an LLM application and design the evaluations that test them. Produces an approved plan covering criteria, eval methods, graders, and test-case coverage. Optionally generates an eval harness and runs it, each behind its own human approval gate. | signal only | 16 | 0 | v1.0.1 |
| Mutation TesterTesting & Quality Technology-agnostic mutation testing reactive skill powered by high-performance Go CLI engine. | exit code | 9 | 2 | v1.0.1 |
| Onboarding MapGeneral Codebase onboarding workflow: intake role and scope, scan architecture and ownership, map modules and dependencies, design a personalized learning path, review with stakeholders, and approve completion. | payload guard | 8 | 5 | v1.0.5 |
| PEP 8 ReviewGeneral Reactive Python style review using PEP 8, project-specific style rules, and relevant companion conventions for docstrings and type annotations. | payload guard | 7 | 4 | v1.0.1 |
| Product ManagerMetaprogramming & Lifecycle Event-driven product management and opportunity realization engine. Orchestrates opportunity discovery & worthwhileness research (new green-field projects vs existing product feature enhancements), strategic goal alignment, SMART requirement scoping, Eisenhower matrix prioritization, and lean vertical slice architecture. | exit code | 14 | 4 | v1.0.4 |
| Research Design PlannerGeneral One reactive entry point that helps users explore, choose, plan, or review a quantitative, qualitative, or mixed methods research design. | payload guard | 34 | 22 | v1.0.1 |
| Resume ManagerCareer & Automation Consolidated reactive career management engine. Orchestrates master profile bootstrapping, continuous achievement maintenance, multi-track profile specialization (Senior Technical, Practical Mid-Level, Operations Management, Client Solutions, and Adjacent Industry), and job application tailoring with automated overqualification and anti-flight risk calibration. | exit code | 22 | 15 | v2.9.0 |
| Skill ManagerMetaprogramming & Lifecycle Full CRUD lifecycle management for reactive skills - CREATE UPDATE DELETE MIGRATE_LEGACY and MIGRATE_REACTIVE with pre-COMMIT approval best-effort rollback and manifest snapshot projection | exit code | 24 | 21 | v1.4.3 |
| Systems DiagnosisGeneral Diagnose recurring or surprising behavior in a system, explain it through observed patterns and structure, and develop a testable intervention. | payload guard | 9 | 6 | v1.0.1 |
| Test Coverage GateTesting & Quality Test coverage quality gate that establishes a clean baseline, measures statement, branch, function, and line coverage, analyzes gaps, and blocks release below explicit thresholds. | exit code | 9 | 13 | v1.0.1 |
Median across 21 registry skills: one state file versus SKILL.md plus every state file, measured in bytes at build time. SKILL.md loads once when the skill starts, and earlier steps stay in the conversation.
A passive skill leaves the model to police itself. A reactive skill runs explicit states with checks on every transition, scoped instructions, and an append-only event log.
A passing demo proves little. A rejected check shows that the workflow can stop, keep its state, and continue from evidence.
Live run
Refactor workflow
states/verify.md
Only active instructions are mounted.
exit_code == 0Run the check. The guard decides whether the workflow advances.
Mount
The agent receives only VERIFY instructions.
Reject
A failed check blocks progress and preserves state.
Resume
A passing result advances the workflow and records why.
Use a published skill, author one for your team, or connect the runtime to your agent host.
Install a workflow someone already made, then run it from your agent or terminal.
Browse the registrynpx skills add Reactive-Skills/skills --skill <name>Turn a repeatable process into explicit states, evidence gates, and durable outputs.
Read authoring guideInstall first: npx skills add Reactive-Skills/skills --skill skill-manager
"Use skill-manager to create a reactive skill named my-workflow"Ask your agent. Recommended.
Terminal alternative: npx -y @reactive-skills/axi init my-workflow
Select a compatible local AXI or MCP transport during startup. AXI uses shell commands, while MCP connects host tool panels.
Compare AXI and MCPnpx -y @reactive-skills/axi mcpReactive Skills is not required for every prompt. It earns its place when a workflow spans multiple steps, tools, or sessions.
Install the official skills, then invoke one by name in your agent. It starts in its first state and tells the agent what to do next.
npx skills add Reactive-Skills/skills