SlashForge

Workflow slash commands for AI coding agents.

Plan. Gate. Verify. Ship.

Your agent
Install
$ npx slashforge

One install for Claude Code, Cursor and Codex.

Node 24+MITv5.3.1

Two runs, start to finish

Shipping a feature

feat/rate-limitingcomplete
$ /slashforge-code
> add rate limiting to the public API
PHASE 2 · PLAN
rateLimit.ts new — token bucket, Redis-backed
server.ts mount before route handlers
RISKS a Redis outage would fail open
TESTS refill · boundary 100 · per-key isolation
PHASE 3 · CONFIRM — stops and waits for you
> proceed
PHASE 6 · VERIFY
lint ✓ tests ✓ build ✓
PHASE 7 · REVIEW
code-reviewer · matches the plan · 0 blocking
PHASE 8 · PR opened against main
Nothing shipped that was not planned, gated, verified.

Reviewing someone else's

reviewing #42complete
$ /slashforge-review-pr
AWAITING YOUR REVIEW
1. #42 Add retry to upload queue +180 −24 (6 files)
2. #47 Bump astro to 7.2 +12 −12 (2 files)
> 1
CONTEXT
CI green · 2 existing comments · author is not you
FINDINGS · against the repo entry file and rules
blocking retry.ts:38 no cap, retries forever
should fix retry.ts:52 swallows the abort signal
nit retry.ts:12 name says ms, value is s
GATE · nothing is posted yet
approve · comment · request-changes · edit · cancel
> request-changes
POSTED · 3 line comments on #42
You saw the exact text before it went out.

Eight commands

Each command triggers one distinct workflow
/slashforge-setup

One-time repo setup. Explores the codebase, asks clarifying questions, then generates an entry file plus rules, skills, subagents, commands, and hooks tailored to it. Runs on every agent and writes each one’s own layout.

/slashforge-code

End-to-end development. Ten phases from plan to merged PR, with mandatory stop-and-confirm gates. Add -quick for lean mode on small changes.

/slashforge-investigate

Read-only research. Reproduces and root-causes a bug, then writes a findings report. Never touches your code.

/slashforge-fix

Patches a bug straight from an investigation. Reads the finding, locks context to the implicated files, and writes a regression test before the patch. No full planning phase.

/slashforge-resume

Resumes an interrupted run from its last checkpoint. Verifies the git HEAD still matches, then re-enters the workflow at the next phase instead of starting over.

/slashforge-review-pr

Reviews a pull request against your repo's own rules and conventions plus a security audit, then posts line-level comments or approves. Never posts without showing you the exact text.

/slashforge-test

Generates spec-based test coverage for existing files. A light, standalone flow: discovers your framework, writes tests, runs them, and reports — surfacing bugs without touching production code. No branch, no PR.

/slashforge-refactor

Refactors with a zero-functional-change guarantee, gated by your test suite passing with identical results before and after. Auto-builds a test net first when coverage is thin.

Names follow the agent picked in the header. Codex invokes them with $ rather than /. All eight run on every agent — setup writes each host's own layout, so on Cursor you get .cursor/rules/*.mdc and on Codex nested AGENTS.md plus .codex/agents/*.toml, never .claude/.

What it costs

Tokens per run
Repo setup
50–120k

Once per repo, before anything else. The largest fixed load of any command, then it writes your agent’s setup.

Full run
100–250k

Ten phases, four gates, agent code review. A multi-layer feature lands near the top.

With Graphify
75–225k

The same run, indexed. The graph shaves roughly 4–10%; it does not change the order of magnitude.

Lean mode (-quick)
40–70k

Skips brainstorming and the review agent. Every user gate and all verification stay.

Investigating a bug
15–60k

No code is written, so this is reading — however far the trail runs before the cause.

Reviewing a PR
15–70k

Almost entirely the diff. Three files is about 2k of it; fourteen is about 42k.

Covering a file (test)
15–50k

Discovers your framework, writes and runs spec-based tests for the targeted files, then reports. No branch, no PR.

Refactoring
40–120k

Zero functional change, gated by the suite passing with identical results before and after.

The range is driven by the size of the work, not by the tooling. SlashForge is deliberately heavy — if you want a prompt turned into a patch as fast as possible, this is the wrong tool.

Steering Committee

Review and feedback

Practitioners who run SlashForge on real work and shape it with their feedback.

Built to make disciplined engineering the default, not an afterthought.

Read the introduction