Templates, slash commands, agents, and reference materials for running solo projects with structured state management, review gates, and decision tracking. Two versions live in this repo: the original file-based workflow (v1) and a phase-gated agent workflow (v2).
Start with v1 (workflow-package/). It's simpler, requires no agents or
advanced features, and covers the full dev cycle.
Use v2 (repo-update/v2-phase-gated-agent-workflow/) if you're doing
data science, analytics, or reporting projects and are comfortable with
Claude Code's subagent system. It adds automated review phases but is more
complex to run.
The original workflow for solo portfolio projects. Emphasizes:
- State in files, not in sessions
- Structured journaling at session end
- Explicit failure capture (not just success)
- Vertical slices over horizontal phases
- Measurable rules over vague guidance
Core commands (full set in workflow-package/slash-commands/, which also
includes /qa, /pre-ship, /office-hours, and plan-review commands):
/init— Scaffold a new project with all workflow files and a guided walkthrough/log— Save a checkpoint (what changed, what's next)/wrap— End-of-session protocol (captures everything for next session)/improve— Review and improve an existing project (audit + guided fixes + tracking)/add-workflow— Retrofit workflow files onto an existing project
Quick start: see workflow-package/README.md and
workflow-package/START-HERE.md (beginner walkthrough). Setup is copying the
slash-command Markdown files into ~/.claude/commands/ and the templates
into your project — no build or install step.
A phase-gated development workflow using Claude Code's subagent system. Specialized agents handle planning, code review, data/analysis validation, prose/narrative review, remediation tracking, and final audit. Built for data science, analytics, and reporting projects but works for any project type.
What it adds over v1:
- Automated phase prompting — Claude Code tells you where you are and what comes next at every phase boundary, including after session restarts and context compaction
- Dedicated review agents — separate code quality, data/analysis correctness, and prose/narrative reviews with structured findings
- Data science validation layer — checks aggregation logic, join integrity, metric definitions, chart-data alignment; traces summary numbers back to source data
- Remediation tracking — consolidates review findings into a single prioritized checklist with resolution status
- Scope change protocol — plan amendments are logged; reviews check against the current spec
- Cross-model scope review — optional independent plan validation by a second LLM (e.g., Gemini) before building
Phase flow:
PLAN → BUILD → CODE REVIEW → DATA REVIEW → PROSE REVIEW → REMEDIATE → AUDIT → COMMIT
↑ │
└───────────────────────────────────────────┘
(loop if needed)
Setup: see repo-update/v2-phase-gated-agent-workflow/README.md — you
drop its .claude/ directory (workflow definition + agent files) into your
project root.
Plain Markdown throughout: slash-command prompt files, agent definitions, and state-file templates for Claude Code. No code, no dependencies.
workflow-package/— v1:slash-commands/,templates/(CLAUDE.md, PLAN.md, HANDOFF.md, DECISIONS.md, FAILURES.md, src/tests variants),reference/(chat project instructions, confidence prompt), START-HERE and quick-start guidesrepo-update/— staged repo README update plusv2-phase-gated-agent-workflow/(v2 documentation)AUDIT.md,PLAN.md,DECISIONS.md,HANDOFF.md— this repo's own workflow state files
Built by Shawn Phillips for solo portfolio and consulting work — data analysis, reporting, visualization, data hygiene audits, and adjacent projects. Distilled from working in Claude Code across multiple projects.
MIT — see LICENSE.