SlashForge
Start here

Plan mode and /init

Claude Code ships both. Here is the line between them and SlashForge — and when you need neither.


These are the two things people reach for first, and the two objections worth answering before anything else. Neither is a competitor exactly. The honest framing is that SlashForge starts where both of them stop.

Plan mode

Plan mode is a pause before implementation: it drafts an approach, you approve it, and the agent then works freely to the end That first gate is genuinely valuable — and it is the one gate SlashForge shares.

Plan mode gates the plan. SlashForge gates the plan, the branch, the PR, and the cleanup.

The difference is everything after approval.

Plan modeSlashForge
GatesOne, at the planFour — plan, branch strategy, PR target, post-merge cleanup
After approvalStops enforcingVerification, code review, and PR phases still run
VerificationNot requiredPhase 6 must pass lint, tests, and build — a failure stops the run before a PR exists
Branch and PRLeft to the conversationEach is an explicit decision you answer
PersistenceEphemeral — the next session starts cold/slashforge:setup writes CLAUDE.md and .claude/, so the next session starts informed
RepeatabilityAs consistent as that day’s promptThe phases run identically every time, for everyone on the team
Not exclusive

You can use plan mode inside SlashForge. Phase 2 is exactly where it fits — the ten phases care that a plan was confirmed, not how you drafted it.

/init

Claude Code ships with a built-in /init, and it is good at what it does. The distinction is narrower than “which tool is better”:

/init writes a file. /slashforge:setup installs a workflow.

/init (built-in)/slashforge:setup
Writes CLAUDE.mdYesYes
Asks clarifying questionsNo — discovers and suggestsYes, in batches, before writing anything
CreatesCLAUDE.md only (or + skills/hooks with CLAUDE_CODE_NEW_INIT=1)Full .claude/ — rules, skills, agents, commands, hooks, plus CLAUDE.md
ApproachOpinion-lightOpinionated — enforces multi-agent layout, 200-line cap, global vs specialist split
AgentsNoneMandatory developer, code-reviewer, git, plus specialists
MonoreposSingle-repo focusedRoot plus a CLAUDE.md per app
Safe to re-runSuggests CLAUDE.md improvementsYes — generated_by markers decide what may be overwritten
Installs a workflowNoYes — /slashforge:code and its ten phases

The opinionation is the point. It produces the same structure every time, which is what makes the output reviewable across a team.

When you need neither

A small single-purpose repo. A script. A prototype you will delete next month.

If nobody is going to review the output and nothing ships from it, plan mode alone is the right amount of process — the gates cost more than they save, and /init is fast, unopinionated, and already installed.

Honest limit

SlashForge is deliberately heavy. A full /slashforge:code run costs 100–250k tokens and stops to ask you four questions. If what you want is an agent that turns a prompt into a patch as fast as possible, this is the wrong tool and it will annoy you.

Using both

They compose. Run /init first for a starter CLAUDE.md, then /slashforge:setupits Update flow reads what is already there and fills gaps rather than overwriting.

Files /slashforge:setup did not generate carry no generated_by marker, so they are treated as yours: edited to fill gaps, never overwritten. See /slashforge:setup for how the markers work.