From project-mapper
Generate the complete set of agent-context files for a codebase so any coding agent has full project context the moment it enters the repo. Use when onboarding a project for agentic coding — "map this project", "set up .claude", "create the CLAUDE.md / rules / subagents for this repo", "what docs does an agent need here", or bringing a repo up to a fully-mapped standard. Produces a lean CLAUDE.md, path-scoped .claude/rules, authored .claude/agents subagents, a domain glossary, and a data-contract / key-module reference — all grounded in the real code.
How this skill is triggered — by the user, by Claude, or both
Slash command
/project-mapper:map-projectThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Produce the durable files an agent needs to be oriented in a repo — built **from
Produce the durable files an agent needs to be oriented in a repo — built from the real code, lean, layered, and non-duplicating. The goal: a fresh agent opens the repo and already knows the architecture, the boundaries, where things live, how to run/test, the domain vocabulary, and the central data model.
| File | Purpose | Loads |
|---|---|---|
CLAUDE.md (repo root) | Always-on orientation: what the app is, architecture, where things live, how to add a feature, run/test commands, roadmap, standing rules | every session |
.claude/rules/<area>.md (path-scoped) | The boundary + do/don'ts for one layer/package; one per major area | only when matching files are touched |
.claude/rules/docs-sync.md (global) | "If you change X → also update Y" so docs never drift | every session |
.claude/agents/<name>.md | Authored subagents (reviewers/workers), memory: project | when spawned |
.claude/skills/<name>/SKILL.md | Domain workflows the agent can invoke | on demand |
.claude/commands/<name>.md | Saved multi-step prompts (optional) | when invoked |
.context/GLOSSARY.md | Domain + project vocabulary a newcomer won't know | reference |
.context/<DATA_CONTRACT>.md | The central object/schema: fields, lifecycle, invariants | reference |
(Use .context/ or docs/ — match what the repo already uses.)
scripts/scan.py (prints
the layer map, each package's classes + __all__, and run/test hints), then
read: the entry points, the package/layer layout, each layer's public
surface, the central data model, the run/test commands, and any existing
README / CLAUDE.md / docs and their house style. Every claim you write
must trace to a real file:line.CLAUDE.md: < 200 lines. If one exists, improve it, don't overwrite.
Cover: what the app is · architecture (and how it's enforced, e.g. a gate
test) · where things live · how to add a feature · run/test commands ·
roadmap links · standing rules.paths: frontmatter, the layer boundary, concrete
do/don'ts, and real file pointers. They complement CLAUDE.md — don't
restate it wholesale and never contradict it.tools; memory: project; read-only unless they
must edit.GLOSSARY.md: the domain terms (and project-specific terms) an outsider
needs to read the code.docs-sync.md: a table mapping each kind of change to the doc(s) that must
update in the same commit.docs-sync cover every doc you created.paths globs match real directories; agents
have name + description; commands/run-lines actually work; nothing
duplicates or contradicts across files.See references/templates.md for ready skeletons of every file above.
CLAUDE.md < 200 lines, accurate tree + run/test commandsmemory: projectdocs-sync rule keeps it all from driftingdocs/ vs .context/ choice, match it.Salvus_MASW/salvus_sw — a fully-mapped repo
(lean CLAUDE.md + per-layer rules + subagents + .context/{GLOSSARY, SIMULATION_JOB}). Mirror its shape for sibling repos.npx claudepluginhub mersadfathizadeh1995/programming_tools --plugin project-mapperProvides CDSS development patterns for drug interaction checking, dose validation, clinical scoring (NEWS2, qSOFA), and alert classification integrated into EMR workflows.