Skip to content
v0.2.1MIT

Stack-neutral skills for a specification-first workflow with use cases: requirements, entity model, use case specifications, review, implementation, tests, change handling, coverage audit, recovery of existing systems and scaling to several teams.

upspec

Agent skills for specification-first development with use cases, for any stack.

You write down what the system should do. A person reviews and approves it. Only then does the agent write or change code, and scripts check the code and the tests against what was approved. The skills cover the whole path: vision, requirements, entity model, use cases, review, implementation, tests, change and coverage. They also bring an existing codebase under specifications.

Version 0.2.1. Fourteen skills. Their scripts are tested against fixture projects, and each skill is evaluated against sessions without it.

Install

Claude Code:

/plugin marketplace add ChristianOrrala/plugins
/plugin install upspec@christian-orrala

Codex:

codex plugin marketplace add ChristianOrrala/plugins
codex plugin add upspec@christian-orrala

Other agents: copy the folders under skills/ into the folder where your agent finds skills, for example .agents/skills/ in the project. Each skill folder carries its own references, templates and scripts.

Requirements: Python 3.9 or later (the scripts use the standard library only) and git.

Quick start

Ask your agent in plain words. The right skill loads on its own.

You haveAsk
A new projectSet this repo up for spec-driven development.
An existing codebaseReverse-engineer this system into specifications from its code.
Work under wayWhere are we with the specs, and what comes next?

The last one runs the router. It reads the repository's state and names the one next step and the skill that does it, so you never need to remember the order.

In Claude Code you can also call a skill by name, for example /upspec:up-5-writing-use-case-specs.

What you get

  • Invention lists. The agent lists every decision its inputs did not settle, instead of quietly choosing.
  • Status needs evidence. A use case becomes Implemented or Tested only when a check or a report shows it.
  • Checks that hold. Lint, the finish check, the guard check and coverage are scripts, so they work even when a model skips an instruction.
  • Reviewers that don't edit. Review and audit report findings and leave the files alone.
  • A rigor dial. Throwaway, solo or team: the ceremony fits the project.
  • People decide. The vision is written by a person, and only a person approves a specification.

The skills

Type up- to list them all. The number is the skill's place in the chain.

SkillWhat it does
up-0-navigating-spec-driven-workSays where the project stands and what comes next
up-1-bootstrapping-spec-driven-projectsSets the repo up: rigor profile, docs tree, agent rules, team checks
up-2-cataloging-requirementsBuilds the requirements catalog and records conflicts
up-3-modeling-domain-entitiesModels the entities as a shared vocabulary, with their lifecycles
up-4-mapping-use-casesFinds the use cases, at the right level, in delivery order
up-5-writing-use-case-specsWrites one use case an agent can implement without inventing
up-6-reviewing-specificationsLints and reviews specifications; reports only
up-7-implementing-use-casesImplements one approved use case, or brings the code back in line
up-8-deriving-use-case-testsDerives tests from the specification and proves they guard its rules
up-9-auditing-spec-coverageShows coverage from requirement to test; reports only

Four side entries, without a number:

SkillWhat it does
up-writing-journey-test-casesWrites end-to-end journeys across several use cases
up-triaging-change-requestsTurns a bug, a change or a hotfix into a change that starts at the specification
up-recovering-specs-from-codeRecovers the specifications of an existing system
up-splitting-bounded-contextsSplits a specification set that has grown too large for one review

How they fit together

bootstrapping ─► vision (a person) ─► requirements ─► entities ─► use case map
                                                                        │
      ┌─────────────────────────────────────────────────────────────────┘
      ▼
  writing ─► reviewing ─► approval (a person) ─► implementing ─► tests ─► audit
                                                                   │
                                               journey test cases ◄┘

Triage turns every later request into a change that starts at the specification.
Recovery brings an existing system to the same starting point.
Splitting cuts a set that has outgrown review.
The router tells you where you are in this picture.

One skill set at a time

Seven of these skills answer the same requests as other spec-driven skill sets: requirements, the entity model, use case diagrams and specifications, their review, test cases and recovery. With two such sets installed, both respond. Keep one set installed for those requests.

Documentation

MANUAL.md is the full guide: concepts, the project's files, workflows for each kind of work, every check, rigor profiles, working with different agents, safety and troubleshooting.

Nothing leaves your machine: no script calls a network service, and no skill pushes, publishes or deploys.

Credits

These skills are inspired by Simon Martinelli's work on the Unified Process. They are an independent work, not affiliated with or endorsed by him.

License

MIT. Copyright (c) 2026 Christian Orrala. See LICENSE.