返回 Skill 列表
extension
分类: 开发与工程无需 API Key

pal-thinkdeep

使用PAL MCP进行多阶段深度调查和推理以解决复杂问题。适用于架构决策、复杂分析、性能挑战,或当你需要详尽的推理时。在遇到需要深入思考、假设检验或专家分析的复杂问题时触发。

person作者: jakexiaohubgithub

PAL ThinkDeep - Deep Investigation

Systematic multi-stage investigation for complex problem analysis.

When to Use

  • Complex architectural decisions
  • Performance challenges requiring analysis
  • Security concerns needing investigation
  • Problems requiring hypothesis testing
  • When surface-level analysis isn't enough
  • Strategic technical planning

Quick Start

result = mcp__pal__thinkdeep(
    step="Investigating intermittent performance degradation under load",
    step_number=1,
    total_steps=3,
    next_step_required=True,
    findings="Beginning systematic investigation",
    problem_context="API response times spike from 50ms to 5s randomly",
    focus_areas=["performance", "database", "caching"],
    hypothesis="Unknown - needs investigation",
    confidence="exploring"
)

Required Parameters

| Parameter | Type | Description | |-----------|------|-------------| | step | string | Current investigation narrative | | step_number | int | Current step | | total_steps | int | Estimated total | | next_step_required | bool | More investigation needed? | | findings | string | Evidence and insights |

Optional Parameters

| Parameter | Type | Description | |-----------|------|-------------| | hypothesis | string | Current theory | | confidence | enum | exploring → certain | | focus_areas | list | ["architecture", "performance", "security"] | | problem_context | string | Background and constraints | | relevant_files | list | Key files (absolute paths) | | files_checked | list | All files examined | | issues_found | list | Problems with severity | | continuation_id | string | Continue session | | thinking_mode | enum | minimal/low/medium/high/max |

Investigation Process

Step 1: Define problem and scope
        ↓
Step 2: Gather evidence, form hypotheses
        ↓
Step 3: Test hypotheses with data
        ↓
Step 4: Refine or pivot based on evidence
        ↓
Step N: Reach conclusion with confidence

Confidence Progression

| Level | Evidence | Action | |-------|----------|--------| | exploring | None yet | Gather data | | low | Initial clues | Form hypothesis | | medium | Supporting evidence | Test hypothesis | | high | Strong evidence | Verify conclusion | | very_high | Confirmed | Document findings | | certain | Proven | Skip validation |

Example: Architecture Decision

# Step 1: Frame the problem
mcp__pal__thinkdeep(
    step="Should we adopt event sourcing for the order system?",
    step_number=1,
    total_steps=4,
    next_step_required=True,
    findings="""
    Current state:
    - CRUD-based order management
    - 50K orders/day, growing 20%/month
    - Audit trail requirements from compliance
    - Complex order lifecycle (10 states)

    Questions to investigate:
    - Performance implications at scale
    - Team learning curve
    - Migration complexity
    - Operational overhead
    """,
    problem_context="Evaluating event sourcing vs enhanced CRUD",
    focus_areas=["architecture", "scalability", "complexity"],
    hypothesis="Event sourcing may be over-engineering for current scale",
    confidence="exploring"
)

# Step 2: Analyze trade-offs
mcp__pal__thinkdeep(
    step="Analyzing event sourcing trade-offs for our context",
    step_number=2,
    total_steps=4,
    next_step_required=True,
    findings="""
    Evidence gathered:

    FOR event sourcing:
    - Built-in audit trail (compliance requirement)
    - Temporal queries for analytics
    - Replay capability for debugging

    AGAINST event sourcing:
    - Team has no ES experience (3-6 month learning)
    - Infrastructure complexity (event store, projections)
    - 50K orders/day is manageable with CRUD + audit table

    Middle ground discovered:
    - Audit table with change tracking gives 80% of benefits
    - Can evolve to ES later if needed
    """,
    problem_context="Weighing complexity vs benefits",
    hypothesis="Audit table + CDC is sufficient for current needs",
    confidence="medium",
    continuation_id=result["continuation_id"]
)

Focus Area Guidelines

Architecture

  • System boundaries and responsibilities
  • Data flow and dependencies
  • Coupling and cohesion
  • Scalability patterns

Performance

  • Bottleneck identification
  • Resource utilization
  • Query optimization
  • Caching strategies

Security

  • Attack surface analysis
  • Data protection
  • Authentication/authorization
  • Compliance requirements

Best Practices

  1. Start with questions - What do we need to learn?
  2. Gather evidence first - Don't jump to conclusions
  3. Consider alternatives - Challenge your assumptions
  4. Document reasoning - Future you will thank you
  5. Update confidence honestly - Uncertainty is information
  6. Use continuation_id - Preserve context across steps