/slashforge:setup
One-time repo setup — generates CLAUDE.md plus tailored rules, skills, agents, commands, and hooks.
/slashforge:setupOne-time setup for a repository. Explores the codebase, asks clarifying
questions, then generates a .claude/ configuration tailored to what it found.
Handles both fresh repos and partial setups — if .claude/ already exists, it
fills gaps rather than overwriting.
What it creates
| Output | Purpose |
|---|---|
CLAUDE.md | Root instructions — architecture, conventions, intent routing |
.claude/rules/ | Conventions the agent must follow |
.claude/skills/ | Repo-specific procedures |
.claude/agents/ | Specialist agents, plus shared git and code-reviewer |
.claude/commands/ | Repo-specific commands |
.claude/hooks/ | Automated behaviours |
In a monorepo it creates shared rules and root CLAUDE.md, then app-specific
rules, skills, agents, and per-app CLAUDE.md files.
Safe re-runs
Every generated file carries a generated_by marker — YAML frontmatter for
.claude/ files, an HTML comment for CLAUDE.md. On re-run the marker decides
what happens:
| Marker state | Behaviour |
|---|---|
| Present, version current | Safe to refresh |
| Present, version older | Stale — proposes a refresh, asks before overwriting |
| Missing or edited | Treated as yours — gaps filled, never overwritten |
Remove or edit the marker on any file you want left alone permanently.
Monorepos
In a monorepo it creates a root CLAUDE.md with shared rules and global agents,
then a separate CLAUDE.md per app with app-specific rules, skills, and
specialist agents.
Graphify
During exploration, /slashforge:setup checks language fit for
Graphify. If at least 70% of
non-trivial source files are in a supported language, it offers to install and
index. On YAML, shell, or config-only repos it skips silently.
This is a setup-time offer, not a per-command check. It always shows you the exact commands before asking — nothing is installed without your say-so.
If you accept, it also synthesises graphify-out/SUMMARY.html, a
browser-readable interpretation of the machine-formatted graph report.
What it costs
~50–120k tokens — once per repo. Unlike the other commands, this is not a per-run cost. You pay it when you adopt SlashForge in a codebase and then not again unless you re-run it.
It carries the largest fixed instruction load of any command: ~20k tokens, before it has read a single line of your code. That is the command plus twelve guide files, and it is the price of the output being tailored rather than templated — the guides are what let it write rules that match your conventions instead of generic ones.
The rest divides in two, and the second half is the expensive direction:
| Exploration | Reading your repo — structure, conventions, existing config. Scales with repo size. |
| Generation | Writing CLAUDE.md, rules, skills, agents, commands and hooks. Output tokens, a dozen or more files. |
Accepting the Graphify offer adds roughly 5–15k for
synthesising SUMMARY.html.
A small, conventional repo lands near 50k. A large polyglot one with several distinct subsystems — more to read, and more rules worth writing — approaches the top.
What it does not do
It does not write application code, create branches, or open PRs. It only
produces configuration. Use /slashforge:code
for development work.