← aidoc-flow.io

AIDoc Flow plugin for Claude Code

Pre-release v0.18.0 — breaking changes possible between minor versions. Your bug reports help.

The plugin drives Claude Code through the AIDoc Flow eight-layer spec-driven loop — BRD → PRD → EARS → BDD → ADR → SPEC → TDD → IPLAN — with multi-persona review crews and a deterministic structural-gate floor on every layer. Claude itself does the generation, validation, and scoring; there's no MCP backend and nothing to run separately.

Install

From a Claude Code session:

/plugin marketplace add vladm3105/aidoc-flow-framework
/plugin install aidoc-flow@aidoc-flow-framework

The plugin is self-contained — it vendors the framework spec it needs, so it installs and runs from a marketplace with no external checkout, no MCP server, no extra setup.

Quickstart

The five commands to go from empty repo to a green BRD:

/aidoc-flow:doc-flow            # "which skill do I need?" — start here
/aidoc-flow:project-init        # scaffold the docs/ layer tree
/aidoc-flow:doc-brd-autopilot   # draft the first layer (BRD) end-to-end
/aidoc-flow:doc-brd-audit       # score it against the layer's quality gate
/aidoc-flow:doc-validator       # validate cross-doc refs & traceability

Work down the layers (doc-prd, doc-ears, … doc-iplan), running each layer's -audit before promoting to the next. doc-flow is the orchestrator: describe your goal and it routes you to the right skill.

What's inside

Full skill catalog

Per-layer skills (32)

Each of the 8 SDD layers ships in 4 variants — the same shape every time:

• doc-<layer> — author the layer's artifact (template-driven)
• doc-<layer>-autopilot — draft + audit + fix in one cascade
• doc-<layer>-audit — score against the layer's quality gate (default 90)
• doc-<layer>-fixer — apply remediation findings

Where <layer> is one of brd, prd, ears, bdd, adr, spec, tdd, iplan — e.g. doc-brd-autopilot, doc-iplan-audit.

CHG governance overlay (4)

Same 4-variant shape, but governs edits to existing artifacts rather than authoring a new layer:

• doc-chg · doc-chg-autopilot · doc-chg-audit · doc-chg-fixer

Utility skills (14 active + 2 deprecated stubs)

Review crews — the differentiator

Every layer has its own weighted review crew. In team mode (default at gates), the layer's audit skill fans out one Claude Code Task subagent per lens, in parallel, each writing its persona-output record to a slot on a per-artifact blackboard. The synthesizer agent then reduces the slots deterministically and produces a single readiness score against the gate floor (default 90).

Crews + weights are declared in framework/governance/REVIEW_CREWS.yaml; the model contract is in framework/governance/REVIEW_TEAM.md. BRD is currently fully wired for team mode (BRD-RT-001); the other layers fall back to single_pass while per-layer wiring lands (PRD-RT, EARS-RT, … coming).

Per-layer playbooks (51)

Each lens, on each layer, carries a playbook: layer-specific finding categories the lens looks for, with worked examples. The audit skill injects the relevant playbook into each lens subagent's brief, so reviews stay grounded in layer-shaped checks rather than generic prose. 45 of 45 lens-specific finding categories covered across the SDD layers; 6 additional playbooks for the CHG overlay.

Playbooks live under framework/playbooks/<NN>_<LAYER>/<lens>.md. Each ships a stable schema (finding categories, severity, example detect+fix pairs) consumed by the per-layer audit skills.

Project overrides — tune the gate

Drop a .aidoc/profile.yaml in your project to override review behaviour:

• review_mode: team | single_pass — team fans out per-lens subagents; single_pass runs one model context applying every lens. Defaults to team at gates.
• audit_threshold: <int> — raise the gate score floor (raise-only; can't go below the framework default of 90).
• section_toggles — toggle optional template sections (e.g. BRD §2 Executive Summary).
• active_layers — restrict the cascade to a subset of the 8 layers.

The closed adaptation surface is defined in framework/governance/ADAPTATION_SURFACE.yaml — these are the only knobs a project may turn.

Platform info

Status — pre-release, in flight

The framework spec is at v0.21.1; the plugin at v0.18.0. We use it on ourselves daily. The install command is stable; the SKILL surface is still settling. Expect breaking changes between minor versions until v1.0.

Found a bug? Surprised by a behaviour? Have a half-formed idea? Drop it in GitHub Issues — the messier, the better. This is the dev site; the polished release notes live at aidoc-flow.ai when it ships.

Links

• GitHub repository
• Plugin README (full reference)
• Releases & changelog
• Issues — feedback welcome
• REVIEW_TEAM.md — review-crew architecture
• ADAPTATION_SURFACE.yaml — what a project may tune