fieldwork

Fieldwork

Claude Code for more than code.

You're using Claude Code for consulting, research, writing, or analysis — and every session starts from scratch. You re-explain the project, re-read the status, re-orient Claude to where you left off. Fieldwork fixes that.

It gives your knowledge work the same continuity that developers get from codebases: session briefings that load instantly, deliverable tracking that persists across sessions, and structured wrap-ups that prepare the next session before you close the current one.

Install

git clone https://github.com/bhpascal/fieldwork.git ~/.claude/plugins/fieldwork

Quick Start

/fieldwork:init          # Set up a new project
                         # ... do your work ...
/fieldwork:wrap-up       # End of session — commit, update status, generate briefing
                         # ... next session starts with instant context ...

That's the core loop. init scaffolds a project with deliverable tracking. wrap-up commits your work, updates the status table, and pre-computes a briefing. Next time you open Claude Code, the briefing loads instantly — no AI needed, no re-explaining.

Commands

Command	What It Does
/fieldwork:init	Initialize a new project from a template or conversational setup
/fieldwork:wrap-up	End-of-session ritual: sync status, commit, generate briefing
/fieldwork:status	View deliverable/milestone status table
/fieldwork:briefing	View or regenerate the session briefing
/fieldwork:todo	Generate a printable, interactive HTML checklist

Templates

Start with a template or go custom:

Template	Best For
Consulting	Client engagements with contracts, SOWs, defined deliverables
Research	Academic, policy, or market research with sources and methodology
Writing	Long-form writing with drafts, editorial process, audience awareness
General	Anything else — events, curricula, grants, planning
Custom	Conversational setup — 3-6 adaptive questions, then scaffold from answers

How It Works

Three layers of persistence, zero cloud dependency:

Layer	What It Captures
Status table (CLAUDE.md)	Current state of each deliverable — the cross-session source of truth
Commit log (local git)	What was done, when, in what order — invisible infrastructure
Project memory (.claude/memory/)	Decisions, lessons, non-obvious context

The pre-computed briefing is the trick: it's generated at session end (when Claude has full context), then loaded at session start (as a plain file read — zero AI cost, instant load).

Fieldwork also embeds AI fluency patterns from Anthropic's research: iteration prompts in briefings, honest status assessment in the deliverable table, and rules that discourage over-polishing early work.

Compatibility

Works alongside other Claude Code plugins (personality-roulette, superpowers, etc.). See CLAUDE.md for architecture details.

License

MIT