robertsfeir

robertsfeir/atelier-pipeline

Plugin Claude Code

Multi-agent orchestration for AI-powered IDEs — quality gates, wave-based QA, and persistent institutional memory.

robertsfeir/atelier-pipeline

Plugin Claude Code

Multi-agent orchestration for AI-powered IDEs — quality gates, wave-based QA, and persistent institutional memory.

robertsfeir/atelier-pipeline

Agent Claude Code

Documentation specialist. Invoke when documentation needs writing, updating, or restructuring. Handles user guides, API docs, architecture overviews, tutorials, troubleshooting guides, release notes.

robertsfeir/atelier-pipeline

Agent Claude Code

Senior Software Engineer. Invoke when Sarah has an ADR (or the scope is clear without one). Implements code, exercises what she shipped, and documents contract shapes produced. Runs the feedback loop before calling a change done.

robertsfeir/atelier-pipeline

Agent Claude Code

Lossless document compression engine. Strips formatting overhead while preserving every fact, decision, constraint, and relationship. Compression, not summarization. Subagent only -- never a skill.

robertsfeir/atelier-pipeline

Agent Claude Code

Commit and Changelog agent. Invoke when code has passed QA and is ready to be committed and pushed. Writes narrative commit messages and executes commit/push.

robertsfeir/atelier-pipeline

Agent Claude Code

Default post-build verifier. Blind code investigator. Invoke with raw git diff output -- no spec, no ADR, no context. Evaluates the diff on its own terms and exercises the code where practical. 1-3 findings typical; zero with confidence is acceptable.

robertsfeir/atelier-pipeline

Agent Claude Code

Product spec producer. Invoke to discover features, write product specs, and define acceptance criteria. Writes to docs/product/. Dual mode with robert (reviewer).

robertsfeir/atelier-pipeline

Agent Claude Code

Product acceptance reviewer. Invoke after final Poirot review to verify implementation against product spec. ADR-blind — reads only the spec and implemented code. Flags spec drift and doc accuracy. Read-only — no Write/Edit access.

robertsfeir/atelier-pipeline

Agent Claude Code

UX design producer. Invoke to create user experience documents, design user flows, interaction patterns, and accessibility guidelines. Writes to docs/ux/. Dual mode with sable (reviewer).

robertsfeir/atelier-pipeline

Agent Claude Code

UX acceptance reviewer. Invoke after Colby mockup (pre-UAT) and after final Poirot review (Large) to verify implementation against UX design doc. ADR-blind — reads only the UX doc and implemented code. Read-only — no Write/Edit access.

robertsfeir/atelier-pipeline

Agent Claude Code

Senior Software Architect. Invoke when a feature needs an ADR — Sarah explores the codebase just enough to pick a credible decision, then writes a short 1-2 page decision record (Context / Options / Decision / Rationale / Falsifiability). She does not produce implementation manuals or test specs.

robertsfeir/atelier-pipeline

Agent Claude Code

Read-only file/grep/read scout. Returns raw evidence with no synthesis. File-reading scouts emit the '=== FILE: {path} ===' delimiter format required by the brain-hydrate Sonnet extractor and the enforce-scout-swarm.sh hook. Subagent only -- never a skill.

robertsfeir/atelier-pipeline

Agent Claude Code

Security audit agent backed by Semgrep MCP static analysis. Runs at review juncture to identify vulnerabilities, injection risks, and security misconfigurations in changed code. Opt-in via pipeline-config.json.

robertsfeir/atelier-pipeline

Agent Claude Code

Relentless-detective bug investigator. Invoke when the user reports a bug (UAT failure, error message, "this is broken") AFTER Eva has conducted the 6-question intake. Diagnose-only -- no fixes applied. Subagent only -- never a skill. Runs in its own context with no session inheritance; the case brief from intake is…

robertsfeir/atelier-pipeline

Agent Claude Code

Filter/rank/trim of scout outputs into the compact named block consumed by a primary agent (research-brief for Sarah, colby-context for Colby, qa-evidence for Poirot). No opinions, no design proposals. Subagent only -- never a skill.

robertsfeir/atelier-pipeline

Command Claude Code

Invoke Sarah for architectural clarification -- conversational Q&A to gather context before ADR production.

robertsfeir/atelier-pipeline

Command Claude Code

Invoke Eva (DevOps) for infrastructure, deployment, CI/CD, monitoring, and operations questions outside the normal pipeline flow.

robertsfeir/atelier-pipeline

Command Claude Code

Invoke Agatha (Documentation Specialist) to plan documentation for a feature -- what docs need writing, updating, or restructuring -- before architecture begins.

robertsfeir/atelier-pipeline

Command Claude Code

Run the full Robert -> Sable -> Sarah -> Colby -> Ellis pipeline from the current starting point.

robertsfeir/atelier-pipeline

Command Claude Code

Invoke Robert (CPO) to run feature discovery, produce specs, and make product decisions.

robertsfeir/atelier-pipeline

Command Claude Code

Invoke Eva (Release) to cut a versioned GitHub Release from main, or backfill historical releases. Routes the version-bump commit through a PR (branch protection-friendly) and uses full 40-char SHAs with gh release create to avoid the silent short-SHA rejection.

robertsfeir/atelier-pipeline

Command Claude Code

Invoke Sable (UI/UX Designer) to design user experiences, interaction patterns, and interface flows from a feature spec before architecture begins.

robertsfeir/atelier-pipeline

Hook Claude Code

Runs before the agent uses a tool for Write, Edit, MultiEdit, Agent and Bash tool calls, executing prompt-eva-path-reminder.sh, enforce-eva-paths.sh, prompt-brain-capture-reminder.sh, enforce-brain-capture-gate.sh, enforce-sequencing.sh, enforce-pipeline-activation.sh, enforce-scout-swarm.sh, prompt-brain-prefetch.sh…