Project initialization
Creates project state and an empty constitution so the workflow can start cleanly, and the other stages can initialize it automatically if needed.
FluencyLoop
BeanspruchenFluencyLoop is an AI-assisted development workflow for Claude Code and Codex that keeps decisions, docs, and review summaries with the code.
AI-assisted development workflow for branch-based feature work
FluencyLoop is an AI-assisted development workflow for Claude Code and Codex that keeps understanding attached to the code it produces. It turns a branch into a structured loop of initialization, feature work, teaching, documentation, and review so that the rationale for decisions is captured while the work is happening.
The product is built around a simple sequence: initialize once, then run one feature loop per branch, with an optional planning stage for larger initiatives. It generates a living constitution, design documents, session notes, and reviewer-facing summaries from the actual branch, and it also provides a backfill path for work that was merged without the loop.
Core capabilities
Initiative planning
Offers an optional planning path for initiatives that are too large for one branch, breaking them into feature-sized tasks and roadmap diagrams.
Build and teach in one loop
Builds each feature in slices while teaching the reasoning behind real decisions, including the chosen path and the rejected alternative.
Branch-local documentation
Writes plans, Mermaid design diagrams, feature sessions, and review summaries beside the code in docs/fluencyloop/ so the documentation comes from the branch itself.
Decision journaling
Tracks decision rationale, including what was chosen, where it applies, why it was chosen, what was rejected, and whether the rationale was verified.
Review assembly and backfill
Assembles branch sessions and decisions into a reviewer-facing pull request view, and supports backfill for merged work that skipped the loop.
Practical ways to use FluencyLoop
Feature work with decision history
Use the feature loop when you want each branch to produce both code and a record of the reasoning behind it, so later reviewers can see why decisions were made.
Breaking down a larger initiative
Use the planning stage for larger initiatives that need to be split into smaller branch-sized tasks before implementation begins.
Preparing a reviewer-facing PR view
Use the review stage to gather the branch’s sessions and decisions into a pull-request-oriented summary for collaborators.
Reconstructing skipped work
Use backfill when a change was merged without following the loop and you need to reconstruct the design and decision trail from available evidence.
Keeping documentation current with implementation
Use the workflow when you want documentation, diagrams, and session notes to be created during delivery rather than written later from memory.
Pros and Cons
Pros
- Captures design decisions, rationale, and rejected alternatives while the work is fresh.
- Produces documentation, diagrams, and review summaries alongside the code instead of after context is lost.
- Supports an optional planning stage for larger initiatives.
- Includes a backfill path for merged work that skipped the workflow.
- Maintains a per-developer knowledge profile so explanations can be calibrated to familiarity.
Cons
- The workflow depends on Claude Code or Codex rather than being a standalone general-purpose tool.
- The repository notes that some update and startup checks are best-effort and silent if a hook, CLI, or marketplace check fails.
- It is oriented around branch-based feature loops, so it may be less suited to work that does not fit that structure.
FAQ
What workflow does FluencyLoop use?
FluencyLoop is designed for AI-assisted development in Claude Code and Codex. It uses a workflow of init, feature, and review stages, with an optional plan step for larger initiatives.
Do I need to install FluencyLoop separately from the plugin?
The source describes it as a plugin that bundles its deterministic CLI, so there is no separate system-wide FluencyLoop installation required for the supported toolchains.
What does FluencyLoop require to run?
The repository states that FluencyLoop requires Claude Code or Codex, git, and either Bash on macOS/Linux/Git Bash/WSL or PowerShell on native Windows.
Can FluencyLoop handle work that skipped the workflow?
If work was merged without using the loop, FluencyLoop includes a backfill path in Claude Code or Codex to reconstruct and verify the design and decisions from evidence.
Where does FluencyLoop store its output?
The source says documentation, diagrams, feature sessions, and review summaries are generated beside the code under docs/fluencyloop/, while workflow state lives in .fluencyloop/ and per-developer calibration lives in ~/.fluencyloop/.
Quick Facts
- Category
- Developer Tool
- Primary users
- Developers using Claude Code or Codex
- Workflow
- init → feature → review, with optional plan and backfill
- Platform support
- macOS, Linux, Git Bash, WSL, and native Windows via PowerShell
- Source domain
- github.com
- Repository license
- Not stated in the provided source text