Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/8BitTacoSupreme/flowstate/llms.txt

Use this file to discover all available pages before exploring further.

flowstate init is the main entrypoint for FlowState. It walks you through a short intake interview, then runs the full five-step pipeline: deterministic context file generation, split-topic LLM research, a strategic pressure-test call, GSD management context setup, and a pure-Python discipline audit of your project’s git and test configuration. State is persisted to flowstate.json after every step so interrupted runs can resume cleanly.

Synopsis

flowstate init [OPTIONS]

Options

--dry-run
flag
Simulate tool execution without calling real CLIs. All pipeline steps run and produce output, but no claude subprocess is spawned. Useful for testing setup or CI.
--root PATH
string
default:"."
Project root directory. All context files and state are written relative to this path.
--skip-interview
flag
Skip the intake interview and use existing answers from flowstate.json. Use this to re-run the pipeline after manually editing your state file.
--model TEXT
string
Claude model to use for bridge calls. Accepts short names passed directly to the claude CLI (e.g., sonnet, opus, haiku).
--budget FLOAT
number
Maximum spend per bridge call in USD. Passed to the claude CLI as --max-budget-usd. Example: 0.50.
--effort TEXT
string
Effort level hint for the claude CLI. Accepted values depend on your Claude Code version (e.g., low, medium, high).

Interview prompts

When --skip-interview is not set, flowstate init opens an interactive intake interview divided into four sections:
SectionFieldPrompt
Intelligenceresearch_focusWhat specific libraries, APIs, or edge cases should we research before starting?
Strategycore_problemWhat is the core user problem we are solving?
Strategyten_x_visionWhat does the 10x version of this look like?
ManagementmilestonesWhat are the three most critical milestones for this phase? (comma-separated)
Disciplinetest_coverageRequired test coverage percentage?
Disciplinearchitecture_patternArchitectural pattern to follow? (e.g., hexagonal, event-driven, layered)
Projectproject_nameProject name
Press Enter on any prompt to accept the default shown in brackets. All answers are saved to flowstate.json immediately after the interview completes.

Pipeline steps

After the interview, flowstate init runs five steps in sequence:
StepLabelWhat happens
1/5Context GenerationWrites 5 context files deterministically from interview answers. No LLM, runs in under 1 second.
2/5ResearchSplits research_focus into individual topics and runs one focused claude --print call per topic (~30s each). Produces research/report.md.
3/5StrategyRuns a single pressure-test claude --print call (~75s). Produces research/strategy.md.
4/5ManagementSets up context files for GSD skills (already written in step 1, this step enriches them). GSD phases themselves run natively via flowstate launch gsd <N>.
5/5DisciplinePure-Python audit of git repo, test configuration, and hooks. No LLM required.
If the claude CLI is not found and --dry-run is not set, the pipeline automatically falls back to dry-run mode with a warning.

Context files written

After a successful run, five context files exist in your project:
FileConsumerContent
.planning/PROJECT.mdGSDVision, problem statement, architecture pattern, constraints
.planning/ROADMAP.mdGSDPhase breakdown from milestones with acceptance criteria
.planning/config.jsonGSDWorkflow preferences (mode, granularity, auto-commit)
.claude/CLAUDE.mdAll toolsProject context, active tools, current phase summary
research/brief.mdResearch adapterStructured research questions derived from interview answers

Examples

Run the full pipeline with the interactive interview:
flowstate init
Simulate the pipeline without calling real CLIs:
flowstate init --dry-run
Re-run the pipeline using answers already saved in flowstate.json:
flowstate init --skip-interview
Use a specific model with a spend cap:
flowstate init --model sonnet --budget 0.50
Run at high effort against a custom project root:
flowstate init --root ./my-project --effort high

Notes

  • If claude CLI is not installed and --dry-run is not passed, FlowState logs a warning and falls back to dry-run mode automatically.
  • State is written to flowstate.json after every pipeline step. If a step fails, it is marked blocked and the pipeline continues to the next step.
  • Run flowstate check before flowstate init to confirm your Claude Code bridge is configured correctly.

Build docs developers (and LLMs) love