Claude Tools / MCP Servers / ai-development-framework
MCP Servers · joaoariedi/ai-assisted-development-framework
ai-development-framework
Spec-driven development for Claude Code: quality gates, custom agents, security guardrails, and a full specification pipeline. Ships its hooks — including a TaskCompleted gate that mechanically enforces the Verification Iron Law.
CLEARED_STATICmcp-portablev5.2.0
Pipeline results
Coverage
CLEARED_STATIC
Tests run
7 / 8
Portability
MCP-portable
Type
MCP Servers
Tests & results
What the pipeline checked
What's inside
Components found
agents/code-reviewer.md
componentmedium
agents/forensic-specialist.md
componentmedium
agents/quality-guardian.md
componentmedium
agents/repo-scout.md
componentmedium
agents/review-coordinator.md
componentmedium
agents/test-specialist.md
componentmedium
commands/adf.agent.md
componentlow
commands/adf.context.md
componentlow
commands/adf.pr-summary.md
componentlow
commands/adf.quality.md
componentlow
commands/adf.security-scan.md
componentlow
commands/adf.sync.md
componentmedium
commands/speckit.analyze.md
componentmedium
commands/speckit.baseline.md
componentmedium
commands/speckit.brainstorm.md
componentmedium
commands/speckit.checklist.md
componentmedium
commands/speckit.clarify.md
componentmedium
commands/speckit.constitution.md
componentmedium
commands/speckit.fix.md
componentmedium
commands/speckit.implement.md
componentmedium
Coverage
Not assessed
- behavioural execution not run
- egress patterns logged (FLAG-INFO): 1 lexical match(es)
README
From the repository
# 🤖 AI Development Framework v5.2 > A Claude Code plugin for **spec-driven development**: write the spec, then the plan, then the > code — with quality gates that are enforced by hooks rather than by good intentions. --- ## 🎯 What it is Claude Code will happily write code from a one-line prompt. That works until the change is big enough that "what were we building?" stops having an obvious answer — and then it fails quietly, by building the wrong thing well. This framework adds the missing middle. A feature goes **idea → spec → plan → tasks → code**, with a human gate at each seam, and the pipeline refuses to skip ahead. Around that sit six specialist agents, nine hooks, and a set of rules that load into every session. The parts that matter are the ones you cannot talk your way past: - ✅ **A `TaskCompleted` hook blocks a task from being marked done while the tests fail.** Not a reminder — a hook that exits non-zero and refuses. - 🔐 **A pre-commit hook runs secrets detection and linting**, and blocks the commit if they fail. - 🚧 **A plan-phase hook blocks edits outside `.specify/`** while a plan is being written, so the agent cannot start coding "just to check somethi…