tap-skills

The operating system for human+agent dev teams. Enables agents to work autonomously and helps human developers support them effectively.

Two questions TAP answers:

1. How to enable agents to work autonomously in the most effective way
2. How can human developers support AI agents to actually complete the work

Install

# Add Team Brilliant marketplace (one-time)
/plugin marketplace add teambrilliant/marketplace

# Install
/plugin install tap-skills@teambrilliant

Or install directly from GitHub:

/plugin install --from github teambrilliant/tap-skills

Skills

Skill	What it does
/tap-skills:tap-audit	Assess how ready a repo is for autonomous agent work; seeds .tap/architecture.md (incl. feature-flag system)
/tap-skills:loop-check	Assess what's needed to make one workflow's feedback loop autonomous — focused sibling of tap-audit
/tap-skills:blast-radius	Impact analysis of PR changes before merging
/tap-skills:systems-health	Measure dev system health via stocks, flows, feedback loops
/tap-skills:retrospective	Just-in-time retro focused on improving agent autonomy
/tap-skills:tighten-loop	Harvest this session's course-corrections into durable, repo-portable fixes — in-session sibling of retrospective
/tap-skills:tech-roadmap	Build 12-month outcome-based tech roadmap for CEO/board
/tap-skills:curate-product-context	Install and maintain .tap/product.md — product vision, focus, bets, non-goals
/tap-skills:qa-smoke-catalog	Explore a web app and build/update the .tap/smoke-tests.md release smoke-test catalog
/tap-skills:qa-smoke-run	Execute the smoke-test catalog in a browser and report only what's broken
/tap-skills:alignment-atlas	Generate/maintain alignment-diagram atlases — flow-by-flow grids on one navigable surface (default .tap/diagrams/atlas/, or any target dir, e.g. per PARA area)
/tap-skills:render-doc	Deterministically render a markdown doc (or a whole docs tree with wiki-links) into self-contained styled HTML — the pre-publish step
/tap-skills:dossier-publish	Publish docs/bundles to the private Dossier platform (teambrilliant.dev): publish, pull source+comments on any machine, share externally, archive

How they work together

Agent enters repo
      │
      ▼
  /tap-audit ──────────► assess whole-repo readiness, identify gaps
  /loop-check ─────────► assess a single workflow's feedback loop (focused)
  /qa-smoke-catalog ───► build the release smoke-test catalog
      │
      ▼
  (implement + test using dev-skills)
      │
      ▼
  Agent opens PR
      │
      ▼
  /blast-radius ──► human reviews impact, merges or rejects
      │
      ▼
  /qa-smoke-run ──► execute the smoke catalog against the release
      │
      ▼
  /systems-health ──► measure how the system is performing
      │
      ▼
  /retrospective ──► what to improve so agents need less help next time (event-driven)
  /tighten-loop ───► same, harvested from the current session's steers (in-session)

The harness meta-skills

Four skills read a repo or session and prescribe autonomy improvements, in the same Context/Harness/Feedback/Scope vocabulary:

	Full repo	Single loop / session
Assess	tap-audit	loop-check
Learn	retrospective	tighten-loop

Project memory

Skills read and write to .tap/ in the target repo:

.tap/
  tap-audit.md      ← repo readiness assessment
  system-health.md  ← latest health metrics
  learnings.md      ← retrospective insights (append-only)
  architecture.md   ← discovered ADRs, design decisions, and feature-flag system
  product.md        ← durable product context: what we build, focus, bets, non-goals
  smoke-tests.md    ← release smoke-test catalog
  qa-runs/          ← smoke-run failure artifacts (only written when a run fails)
  diagrams/atlas/   ← alignment-diagram atlas (self-contained file:// viewer + per-flow map data; default location, can also live per-area, e.g. areas/<x>/diagrams/atlas/)

.tap/architecture.md and the feature-flag system

tap-audit always captures the repo's feature-flag system as a dedicated decision in .tap/architecture.md — provider (PostHog / LaunchDarkly / Unleash / Flagsmith / static / DB-driven / env-var / "None — direct deploy only"), where flags are defined and read, and naming conventions.

This is the contract that dev-skills's rollout-primitives reference reads when planning a change: discover from .tap/architecture.md first, fall back to grep, ask only if nothing is found. Capturing it once during audit means every downstream plan reads the same answer instead of re-deriving it on each invocation.

If a repo has no flag infra, .tap/architecture.md says so explicitly — planners then default to "direct deploy" without spending cycles re-confirming the absence.

Requirements

• Claude Code
• gh CLI (for blast-radius, systems-health, retrospective)
• Chrome DevTools MCP (for qa-smoke-catalog, qa-smoke-run)

Companion plugin

teambrilliant/dev-skills — generic development workflow skills (shaping, grooming, planning, implementing, testing, browser QA). tap-skills is the methodology layer that wraps around them.