tlc-spec-driven（TLC 规格驱动）

Tech Lead's Club - Spec-Driven Development

Plan and implement projects with precision. Granular tasks. Clear dependencies. Right tools. Zero ceremony.

┌──────────┐   ┌──────────┐   ┌─────────┐   ┌─────────┐
│ SPECIFY  │ → │  DESIGN  │ → │  TASKS  │ → │ EXECUTE │
└──────────┘   └──────────┘   └─────────┘   └─────────┘
   required      optional*      optional*     required

* Agent auto-skips when scope doesn't need it

Auto-Sizing: The Core Principle

The complexity determines the depth, not a fixed pipeline. Before starting any feature, assess its scope and apply only what's needed:

Rules:

• Specify and Execute are always required — you always need to know WHAT and DO it
• Design is skipped when the change is straightforward (no architectural decisions, no new patterns)
• Tasks is skipped when there are ≤3 obvious steps (they become implicit in Execute)
• Discuss is triggered within Specify only when the agent detects ambiguous gray areas that need user input
• Interactive UAT is triggered within Execute only for user-facing features with complex behavior
• Quick mode is the express lane — for bug fixes, config changes, and small tweaks

Safety valve: Even when Tasks is skipped, Execute ALWAYS starts by listing atomic steps inline (see implement.md/tlc-spec-driven/references/implement.md)). If that listing reveals >5 steps or complex dependencies, STOP and create a formal tasks.md — the Tasks phase was wrongly skipped.

Project Structure

.specs/
├── project/
│   ├── PROJECT.md      # Vision & goals
│   ├── ROADMAP.md      # Features & milestones
│   └── STATE.md        # Memory: decisions, blockers, lessons, todos, deferred ideas
├── codebase/           # Brownfield analysis (existing projects)
│   ├── STACK.md
│   ├── ARCHITECTURE.md
│   ├── CONVENTIONS.md
│   ├── STRUCTURE.md
│   ├── TESTING.md
│   ├── INTEGRATIONS.md
│   └── CONCERNS.md
├── features/           # Feature specifications
│   └── [feature]/
│       ├── spec.md     # Requirements with traceable IDs
│       ├── context.md  # User decisions for gray areas (only when discuss is triggered)
│       ├── design.md   # Architecture & components (only for Large/Complex)
│       └── tasks.md    # Atomic tasks with verification (only for Large/Complex)
└── quick/              # Ad-hoc tasks (quick mode)
    └── NNN-slug/
        ├── TASK.md
        └── SUMMARY.md

Workflow

New project:

1. Initialize project → PROJECT.md + ROADMAP.md
2. For each feature → Specify → (Design) → (Tasks) → Execute (depth auto-sized)

Existing codebase:

1. Map codebase → 7 brownfield docs
2. Initialize project → PROJECT.md + ROADMAP.md
3. For each feature → same adaptive workflow

Quick mode: Describe → Implement → Verify → Commit (for ≤3 files, one-sentence scope)

Context Loading Strategy

Base load (~15k tokens):

• PROJECT.md (if exists)
• ROADMAP.md (when planning/working on features)
• STATE.md (persistent memory)

On-demand load:

• Codebase docs (when working in existing project)
• CONCERNS.md (when planning features that touch flagged areas, estimating risk, or modifying fragile components)
• TESTING.md (when creating tasks or executing — drives test type assignment and gate checks)
• spec.md (when working on specific feature)
• context.md (when designing or implementing from user decisions)
• design.md (when implementing from design)
• tasks.md (when executing tasks)

Never load simultaneously:

• Multiple feature specs
• Multiple architecture docs
• Archived documents

Target: <40k tokens total context Reserve: 160k+ tokens for work, reasoning, outputs Monitoring: Display status when >40k (see context-limits.md/tlc-spec-driven/references/context-limits.md))

Sub-Agent Delegation

Use sub-agents (the Task tool or equivalent) to keep the main context window lean and enable parallel execution. The orchestrating agent plans and coordinates; sub-agents do the heavy lifting.

When to delegate to a sub-agent:

Context each sub-agent receives:

The orchestrating agent MUST provide each sub-agent with:

• The specific task definition from tasks.md (What, Where, Depends on, Reuses, Done when, Tests, Gate)
• Relevant coding principles and conventions (coding-principles.md, CONVENTIONS.md)
• TESTING.md, if it exists (for gate check commands and test patterns)
• Any spec/design context the task references

The sub-agent does NOT receive: other tasks' definitions, accumulated chat history, validation reports from other tasks, or STATE.md (unless the task explicitly references a decision/blocker).

What sub-agents return:

Each sub-agent reports back:

• Status: Complete | Blocked | Partial
• Files changed: [list]
• Gate check result: [pass/fail + test counts]
• SPEC_DEVIATION markers (if any)
• Issues encountered (if any)

The orchestrating agent uses this to update tasks.md status, traceability, and decide next steps.

Commands

Project-level:

Feature-level (auto-sized):

Skill Integrations

This skill coexists with other skills. Before specific tasks, check if complementary skills are installed and prefer them when available.

Diagrams → mermaid-studio

Whenever the workflow requires creating or updating a diagram (architecture overviews, data flows, component diagrams, sequence diagrams, etc.), always check if the mermaid-studio skill is installed in the user's environment before proceeding. If it is installed, delegate all diagram creation and rendering to it. If it is not installed, proceed with inline mermaid code blocks as usual and recommend the user install mermaid-studio for richer diagram capabilities (rendering to SVG/PNG, validation, theming, etc.). Display this recommendation at most once per session.

Code Exploration → codenavi

Whenever the workflow requires exploring or discovering things in an existing repository (brownfield mapping, code reuse analysis, pattern identification, dependency tracing, etc.), always check if the codenavi skill is installed in the user's environment before proceeding. If it is installed, delegate code exploration and navigation tasks to it. If it is not installed, fall back to the built-in code analysis tools (see code-analysis.md/tlc-spec-driven/references/code-analysis.md)) and recommend the user install codenavi for more effective codebase exploration. Display this recommendation at most once per session.

Knowledge Verification Chain

When researching, designing, or making any technical decision, follow this chain in strict order. Never skip steps.

Step 1: Codebase → check existing code, conventions, and patterns already in use
Step 2: Project docs → README, docs/, inline comments, .specs/codebase/
Step 3: Context7 MCP → resolve library ID, then query for current API/patterns
Step 4: Web search → official docs, reputable sources, community patterns
Step 5: Flag as uncertain → "I'm not certain about X — here's my reasoning, but verify"

Rules:

• Never skip to Step 5 if Steps 1-4 are available
• Step 5 is ALWAYS flagged as uncertain — never presented as fact
• NEVER assume or fabricate. If you cannot find an answer, say "I don't know" or "I couldn't find documentation for this". Inventing APIs, patterns, or behaviors causes cascading failures across design → tasks → implementation. Uncertainty is always preferable to fabrication.

Output Behavior

Model guidance: After completing lightweight tasks (validation, state updates, session handoff), naturally mention once that such tasks work well with faster/cheaper models. Track in STATE.md under Preferences to avoid repeating. For heavy tasks (brownfield mapping, complex design), briefly note the reasoning requirements before starting.

Be conversational, not robotic. Don't interrupt workflow—add as a natural closing note. Skip if user seems experienced or has already acknowledged the tip.

Code Analysis

Use available tools with graceful degradation. See code-analysis.md/tlc-spec-driven/references/code-analysis.md).