evidence-driven
Evidence-Driven
A discipline overlay for the execution layer. Where goal-driven owns why and design-driven owns what shape, evidence-driven owns prove it works.
The thesis is one line: no progress claim survives without falsifiable evidence. Every State update, every Verify check-off, every "this is done" must carry an observation that could in principle have shown the opposite. A test that can't fail isn't evidence; a checklist run with no captured output isn't evidence; "I tried it and it seemed fine" isn't evidence.
Position in the layered stack
concept: reframe (what shape in a new paradigm; only when no precedent)
strategy: goal-driven (why; success criteria; when to STOP)
architecture: design-driven (what shape; module boundaries; mechanisms)
More from lidessen/skills
memory
Manages cross-session knowledge persistence. Triggers on "remember", "recall", "what did we", "save this decision", "todo", or session handoff.
82design-driven
Design-driven development methodology — the design/ directory is the single source of architectural truth; read it before coding, stay within its boundaries, and update it first when the system's shape changes. Use whenever starting development on this project, or when the user asks to create/update architecture docs, add a feature that may cross existing boundaries, refactor system structure, or understand the codebase architecture. Triggers on "design first", "update the design", "does this change the architecture", "write a design for", "what's the current design", or onboarding to a codebase's shape. Args — `/design-driven init` to configure a project, `bootstrap` to generate design from existing code, `audit` to reconcile design/ against current code.
24harness
Agent harness architecture — structure a project's agent context across layers for effective AI-assisted development. Covers CLAUDE.md, skills, design docs, hooks, and all artifacts that shape how an agent understands and operates in a codebase. Use when setting up or improving agent configuration, when agent context feels bloated or disorganized, when onboarding a project for AI-assisted development, or when the agent keeps losing architectural awareness mid-task. Triggers on "set up claude", "improve CLAUDE.md", "agent keeps forgetting", "context is too long", "harness setup", "organize agent context", "how should I structure my prompts". Args — `/harness audit` to evaluate an existing project's context architecture, `init` to set up harness from scratch.
20housekeeping
Manages project housekeeping including documentation organization, dependency management, directory structure, code cleanup, technical debt tracking, and infrastructure configuration. Use when organizing documentation, cleaning up dependencies, reorganizing folders, removing dead code, addressing tech debt, or maintaining project structure.
18validation
Unified validation orchestration for code quality, consistency, and project health. Auto-triggers on code changes, PR creation, or explicit validation requests. Coordinates refining, housekeeping, and custom validators into cohesive pipelines. Use for "validate", "check", "verify", "验证", "检查", or when quality assurance is needed.
17orientation
Orients agents in new projects by scanning entry documents and discovering capabilities. Use at session start, when entering unfamiliar territory, or when asking "what can you do" or "where do I start".
16