6 dev workflows for Claude Code + create your own. Compiled pipelines, not prompts.
gstack and superpowers tell Claude what to do. Claude decides if it listens.
fwstack compiles the process into a pipeline. Claude gets control at waitForAgent pauses. Everything else runs without it. Real eslint runs. Real git diff runs. Real secret scans run. The AI can't skip what the compiler enforces.
/fwstack:review
codeReview
1. ✓ diff (2.3s), ✓ lint (2.3s) (parallel)
2. ⏸ reviewer (agent)
3. · extract
4. · rank
5. · report
⏸ waiting for agent
Real eslint runs. Real tsc runs. Claude gets the diff AND the linter output. It can't skip steps because it doesn't control the order.
/plugin marketplace add synergenius-fw/claude-plugins
/plugin install fwstack
That's it. Plugin bundles its own MCP server, workflows, and rendering.
Gathers git diff and runs linters in parallel. Pauses for Claude to analyze. Extracts findings as structured JSON, deduplicates by file:line, sorts by severity.
codeReview
1. ✓ diff (2.3s), ✓ lint (2.3s) (parallel)
2. ✓ reviewer (6.1s) (agent)
3. ✓ extract (<1ms)
4. ✓ rank (<1ms)
5. ✓ report (<1ms)
✓ complete (6.0s)
Reads your codebase (file tree, package.json, git log). Pauses for Claude to generate a plan with acceptance criteria. Every task needs a file path and shell commands that prove it's done. Validator rejects vague language and weak commands (echo, true, exit 0, ls).
In implement mode, the workflow iterates over each task: pauses for Claude to implement, then runs the acceptance checks. No exit 0, no next task.
Two pauses. First: Claude writes a failing test. Workflow runs npm test, checks exit code != 0. Second: Claude implements. Workflow runs tests again, checks exit code == 0. Can't skip to implementation because the RED gate blocks.
tddCycle
1. ✓ writeTest (10.7s) (agent)
2. ✓ runRed (175ms)
3. ✓ assertRed (<1ms)
4. ✓ implement (14.5s) (agent)
5. ✓ runGreen (156ms)
6. ✓ assertGreen (<1ms)
✓ complete (26.0s)
Gathers source code and runs SAST tools (npm audit, semgrep) in parallel. Pauses for OWASP + STRIDE analysis. Drops findings below 8/10 confidence. Ranks by severity.
Checks you're on a feature branch. Runs full test suite. If tests fail, stops. No override. If tests pass, gathers the diff and pauses for changelog generation.
shipRelease
1. ✓ precheck (48ms)
2. ✓ tests (115ms)
3. ✓ gate (<1ms)
4. ✓ changes (166ms)
5. ✓ changelogAgent (7.9s) (agent)
✓ complete (8.1s)
Describe what you want, the workflow builds itself. Gathers project context, pauses for Claude to design a structured spec, validates the spec (rejects vague requirements and missing gates), loads Flow Weaver authoring docs, pauses again to build the actual workflow file, validates it with fw validate, installs dependencies, and copies to .fwstack/workflows/ for auto-discovery.
createWorkflow
1. ✓ ctx (70ms)
2. ✓ specAgent (17.5s) (agent)
3. ✓ validate (1ms)
4. ✓ docs (13ms)
5. ✓ buildAgent (58.5s) (agent)
6. ✓ check (1.1s)
7. ✓ deps (<1ms)
8. ✓ install (1ms)
✓ complete (77.6s)
Each command calls a compiled Flow Weaver workflow via an MCP server.
- Deterministic nodes run real tools (git diff, eslint, npm test)
waitForAgentpauses the workflow and returns context + prompt- Claude responds
fwstack_resumefeeds the response back, workflow continues- More deterministic nodes extract, validate, format the result
The workflow stays in memory between calls. No re-execution on resume.
Live progress bar during execution. Colored pipeline tree after each tool call. Per-node timing. Parallel branches visible.
All rendered by the plugin, not by Claude.
Drop a .ts workflow file in .fwstack/workflows/ in your project. fwstack discovers it automatically on the next tool call and registers it as a new command.
Or run any workflow directly:
/fwstack:run
License: MIT