From product discovery and architecture to implementation, code review, impact analysis and delivery governance.
Plan better. Understand repositories deeper. Review code harder. Ship with evidence.
Explore the skills · Choose a workflow · How to use · Contribute
gpt-sw-dev-skills is a curated collection of 32 packaged AI skills for software-development work.
The repository is built around a simple idea: coding agents become much more useful when they are given specialized, reusable operating methods instead of being asked to improvise every task from a blank prompt.
The collection spans the software lifecycle:
Idea / Product Value
↓
Project & Delivery Planning
↓
Architecture & Repository Understanding
↓
API / Middleware / MCP Design
↓
Implementation & Agent Delivery
↓
Code Review / Impact / Gap Analysis
↓
Evidence, Governance & Retrospective
These are not presented here as magic prompts or a replacement for engineering judgment. They are reusable skill packages intended to make AI-assisted software work more structured, inspectable and repeatable.
AI coding agents can generate code quickly. The harder problems are everything around the generation step:
- understanding an unfamiliar repository before changing it;
- defining the right problem and avoiding feature drift;
- planning a bounded implementation slice;
- assessing architectural and downstream impact;
- reviewing correctness, security, reliability and maintainability;
- distinguishing plausible output from verified evidence;
- coordinating coding agents, GitHub work and delivery gates;
- documenting decisions so the next session does not start from zero.
This repository collects skills aimed at those problems.
This collection is especially relevant if you work with:
- AI coding agents and agentic software-development workflows;
- GitHub repositories, pull requests and code review;
- product ownership, agile delivery and project orchestration;
- software architecture, APIs, middleware and MCP servers;
- repository audits, gap analysis and change-impact analysis;
- LLM evaluation, evidence-first delivery and anti-hallucination controls;
- frontend/UI implementation where targeted interaction effects are useful.
You do not need every skill. The value is in choosing the smallest combination that matches the job.
| Skill | What it is useful for |
|---|---|
value-brainstorming |
Value-first ideation, problem discovery, alternatives and MVP convergence. |
gpt-project-management |
Project planning and orchestration for GPT/agent-based software work. |
scrum-product-owner-orchestrator |
Product Owner workflows, backlog/sprint steering and value-centered delivery. |
plumbline-agileteam-intake-builder |
Structured intake, product vision, PRD-oriented preparation and traceability. |
software-project-start-readiness-orchestrator |
Project-start readiness, architecture/security/quality foundations and go/no-go thinking. |
claude-goal-forge |
Bounded software objectives and acceptance-oriented coding-agent goals. |
claude-cowork-project-orchestrator |
Project orchestration and structured handoff patterns for Claude/Cowork-style workflows. |
| Skill | What it is useful for |
|---|---|
codebase-architecture-skill |
Understanding and describing software architecture from a codebase. |
repo-architecture-visualization-harness |
Repository-aware architecture maps, visual models and implementation handoffs. |
deep-repository-impact-analysis |
Change blast-radius analysis across code, contracts, data, tests and architecture. |
software-gap-analysis-agent |
Source-inspected gap analysis against target behavior or architecture. |
adaptive-boilerplate-architecture-builder |
Production-oriented starter architecture and project foundation design. |
| Skill | What it is useful for |
|---|---|
api-developer-builder |
API design, testing, schema validation, debugging and developer-platform architecture. |
middleware-api-architect |
Contract-first middleware, gateways, adapters, webhooks and event/API architecture. |
middleware-project-orchestrator-harness |
Orchestrating middleware/API projects and agent-ready integration harnesses. |
mcp-builder |
Designing and building Model Context Protocol servers and external-tool integrations. |
| Skill | What it is useful for |
|---|---|
evidence-first-coding-agent-delivery |
Coordinating coding agents with verification, WIP control, review and merge gates. |
enterprise-code-review |
Evidence-first review for correctness, security, reliability, architecture, tests and more. |
ultrathink-craftsmanship |
Deeper engineering decision discipline for non-trivial implementation and refactoring. |
deep-reasoning-analysis-enterprise |
Multi-path technical analysis, verification and failure-mode review. |
model-response-audit-comparator |
Comparing multiple model outputs with rubrics, uncertainty and failure gates. |
gpt-self-reflect |
Retrospective analysis of substantial GPT/agent work to extract reusable lessons. |
modell-retrospektive |
Structured after-action review of long agent/coding sessions without relying on hidden reasoning. |
| Skill | What it is useful for |
|---|---|
plumbline-project-manager |
Truth-oriented product/project governance for the Plumbline repository and workflow. |
plumbline-agent-observatory |
Comparing coding-agent behavior, regressions, false-green results and longitudinal quality. |
confluence-project-writer |
Controlled project documentation and validated Confluence page operations. |
research-backed-skill-builder |
Building skills from research and source material with explicit evidence strategy. |
enterprise-skill-creator |
Creating and validating enterprise-ready reusable skill packages. |
ultimate-enterprise-prompt-architect |
Production-oriented prompt systems, evaluator prompts, guardrails and handoffs. |
| Skill | What it is useful for |
|---|---|
enterprise-workforce-planning-builder |
Enterprise workforce, capacity, scheduling and field-service platform delivery. |
reactbits-inspired-ui-effects |
Selecting and integrating premium React interaction and animation effects. |
sw-doce-skill |
Packaged software-development documentation skill. Inspect the bundle for its exact supported workflow. |
Note: The descriptions above are a navigation layer for the repository. The individual package contents remain the source of truth for each skill's exact triggers, requirements, tools and limitations.
The repository becomes more useful when skills are combined by workflow, not simply installed in bulk.
value-brainstorming → software-project-start-readiness-orchestrator → adaptive-boilerplate-architecture-builder → evidence-first-coding-agent-delivery
Use this path when the risk is building the wrong thing or starting implementation before the foundations are clear.
codebase-architecture-skill → repo-architecture-visualization-harness → software-gap-analysis-agent → deep-repository-impact-analysis
Use this path before substantial changes to a repository you do not yet understand.
deep-repository-impact-analysis → implementation agent → enterprise-code-review → evidence-first-coding-agent-delivery
Use this path when a plausible patch is not enough and you need impact, independent review and verification evidence.
api-developer-builder → middleware-api-architect → mcp-builder → enterprise-code-review
Use this path for APIs, adapters, gateways, tool integrations and MCP-based agent capabilities.
scrum-product-owner-orchestrator → claude-goal-forge → evidence-first-coding-agent-delivery → gpt-self-reflect
Use this path to connect product value, bounded implementation goals, delivery control and learning from completed sessions.
Each skill is stored as a .zip package.
- Choose the skill by job, not by name alone.
- Download and inspect the package before use.
- Read its
SKILL.md/skill.md, references, scripts and declared tool requirements. - Install or expose it only in an environment that supports those requirements.
- Give the agent the real repository/files/context needed by the skill.
- Verify consequential claims with tests, repository state, CI, runtime evidence or human review as appropriate.
The repository contains reusable skill packages, but platform compatibility is not claimed globally. A package may depend on specific tools, connectors, filesystem access, coding agents or runtime capabilities. Treat the package itself as the source of truth.
A prompt collection usually gives an LLM something better to say.
A good engineering skill should give an agent a better way to work:
- trigger conditions;
- a repeatable workflow;
- evidence requirements;
- tools and validation steps;
- stop conditions and risk gates;
- reusable output contracts;
- explicit limits and uncertainty.
That distinction matters when the output can change real software.
A convincing answer is not the same thing as a verified result.
Repository analysis and impact assessment should precede broad code mutation.
More generated code is not automatically more useful software.
Important changes benefit from adversarial review rather than self-approval by the generating agent.
Agentic development becomes safer when scope, acceptance criteria and verification remain bounded.
This repository is currently a growing collection of packaged skills. The initial public structure is intentionally simple: downloadable bundles plus this catalog.
Known limitations of the current repository presentation:
- packages are distributed primarily as ZIP archives rather than browsable source directories;
- there is not yet a repository-wide compatibility matrix;
- there is not yet a repository-wide automated validation badge for every package;
- a repository-wide license has not been declared here, so do not infer one from this README.
Those are real next steps, not details to hide behind marketing.
If a skill is useful, breaks in your environment, needs clearer documentation or is missing an important software-engineering workflow, open an issue or pull request.
Useful contributions include:
- reproducible compatibility reports;
- installation improvements;
- eval scenarios and regression tests;
- documentation corrections;
- missing software-engineering workflows;
- evidence that a skill works — or does not work — at a real boundary.
See CONTRIBUTING.md.