Agents Ship.Humans Decide.
Project memory loads your codebase context into every skill. The agent audits, implements, and tests in a real browser. You review and merge.
Claude Code has no memory of your repo.
It doesn't know your architecture decisions. It doesn't know what the team agreed on last sprint. It can't verify that what it built actually works in a real browser.
So every agent session starts blind. Agents repeat mistakes, make architectural choices that contradict past decisions, and open PRs with unknown blast radius.
TAP OS is the missing layer. Context, verification, and visibility — before anything merges.
Two workflows. First PR in minutes.
Workflow A
Human plans → Agent builds
You write the tickets and set direction. The agent implements, tests in a real browser, and opens a PR. You run /blast-radius and decide whether to merge.
Workflow B
Agent does everything
Point an agent at a repo and a task. It audits the codebase, builds, tests, and opens a PR. You only show up to verify and merge.
10 skills across two plugins.
dev-skills — Engineering Workflow
Shape → Plan → Implement → QA. The full build loop.
Converts rough ideas — feature requests, bug reports, customer feedback, Slack threads — into clear, actionable work definitions with acceptance criteria and risks.
Takes a groomed ticket and produces a detailed technical plan: specific file paths, complete code snippets, verification steps. Any developer or agent can follow it.
Executes implementation from a ticket or plan, phase by phase. Verifies after each phase, fixes issues before moving on, and adapts when reality differs from the plan.
Opens Chrome via DevTools MCP, navigates to the feature, runs through every acceptance criterion, and returns a structured pass/fail report. Unit tests are not enough — this checks the real thing.
Analyzes product decisions from multiple angles: user journey, business impact, technical feasibility, competitive landscape, and risk. Grounds recommendations in reality, not assumptions.
Breaks down complex products, features, or systems into fundamental primitives and building blocks. Outputs human-readable lists with ASCII diagrams showing relationships.
tap-skills — Repo Readiness & Visibility
Make your repo ready for autonomous work. Audit, verify, monitor, learn.
Scans the repo and produces a full readiness assessment: documentation gaps, missing MCP servers, test coverage blind spots. Populates .tap/ project memory so every agent session starts informed.
Reads the PR diff and maps the full impact surface: what changed, what depends on it, what isn't covered by tests, and what to verify manually. Risk-rated low → medium → high.
Pulls from git log and GitHub API to measure deploy frequency, PR merge rate, review turnaround, and cycle time. Identifies which feedback loops are working and which are broken.
Captures what worked, what didn't, maps findings to system dynamics, and appends insights to .tap/learnings.md. Creates improvement tickets. The skill that makes all other skills better over time.
Not a skill you run — the infrastructure that makes every skill better. Stores architecture decisions, retrospective learnings, system health, and codebase assessments across sessions. Skills read it automatically.
Add the marketplace, then install the plugins:
# Add the marketplace /plugin marketplace add teambrilliant/marketplace # Install plugins /plugin install tap-skills@teambrilliant /plugin install dev-skills@teambrilliant
For team-wide auto-install, add to your project's .claude/settings.json:
"enabledPlugins": {
"dev-skills@teambrilliant-marketplace": true,
"tap-skills@teambrilliant-marketplace": true
}A delivery system for
teams that ship with agents.
THIS IS FOR YOU IF:
- You use Claude Code and want agents doing real engineering work
- You want a structured workflow: shape → plan → implement → QA
- You want browser-verified QA on every PR — not just unit tests
- You want to know the blast radius before merging any agent PR
- You want project context that survives between agent sessions
THIS IS NOT:
- A replacement for engineering judgment — humans still decide what ships
- A no-code platform — this is for teams that write and review real code
- An autonomous system that merges its own PRs — agents propose, humans dispose
- A one-size-fits-all framework — skills adapt to your repo and workflow
- Magic — agents still need well-structured repos to do their best work
Common questions.
First, add the marketplace: /plugin marketplace add teambrilliant/marketplace. Then browse with /plugin and install what you need: /plugin install tap-skills@teambrilliant or /plugin install dev-skills@teambrilliant. For team-wide auto-install, add the plugins to your .claude/settings.json — new team members get the right skills automatically.
dev-skills is the engineering workflow: shape work, plan implementation, write code, QA in a browser. tap-skills makes your repo ready for autonomous work: audit the codebase, analyze blast radius before merging, monitor delivery health, and run retrospectives.
A set of files in your repo that store architecture decisions, retrospective learnings, system health metrics, and codebase assessments. Skills read these automatically. The more you use the skills, the richer the context, and the better the output.
Never. Agents open PRs, they never merge them. Every merge goes through /blast-radius first, then a human decides. That's not a limitation — it's the design.
Yes, if you're starting on an unfamiliar repo. /tap-audit builds the project memory that every other skill depends on. You can run individual skills without it, but you'll get better results once .tap/ is populated.
In Human plans → Agent builds, you write tickets and plans and the agent implements. In Agent does everything, you point an agent at a task and it handles the full loop. Both end the same way: agent opens a PR, human reviews with /blast-radius, human merges.
Start with one command.
One session with a properly-context-aware agent pays for this. One PR that doesn't break production pays for it 10×.
2 min install · works on any repo