AgentSkillsCN

agentic:product-spec

当您需要将一个模糊的想法转化为精准的产品规格时,此技能将大显身手。触发条件包括:含糊的功能需求、全新的产品构想,或是亟需在设计前明确严谨的需求定义。

SKILL.md
--- frontmatter
name: agentic:product-spec
description: Use when transforming a raw idea into a precise product specification. Triggers - vague feature request, new product idea, need rigorous requirements before design.

/agentic:product-spec - Idea to Product Spec

Usage: /agentic:product-spec [--auto] [<input>]

Transform a rough idea into a precise product specification through rigorous questioning.

Arguments

  • No args: Interactive mode, prompt for idea
  • --auto: Autonomous mode with decision logging
  • path/to/notes.md: Use existing notes/idea file
  • --auto path/to.md: Auto mode with file input

Workflow Overview

code
1. Input Detection → 2. Product Discovery → 3. Write Spec

Step Files

Execute steps in order. Each step file contains detailed instructions.

StepFileDescription
1steps/step-01-input-detection.mdParse args, init state, generate topic slug
2steps/step-02-product-discovery.mdDelegate discovery to subagent
3steps/step-03-write-spec.mdDelegate design, compile final spec

Templates

TemplateDescription
templates/workflow-state.yamlState tracking template

Subagent References

Each subagent reads its own instructions from {ide-folder}/skills/{skill}/SKILL.md.

Invoke: Task(subagent_type="{subagentTypeGeneralPurpose}", prompt="You are the {Agent} agent. {ide-invoke-prefix}{ide-folder}/skills/{skill}/SKILL.md for your full instructions. ...")

Available skills: product-discovery, brainstorming


Mandatory Delegation

You MUST delegate all discovery/design work using the Task tool. NEVER do it inline.

You are the orchestrator. You:

  • Parse input, detect mode
  • Initialize state tracking
  • Invoke subagents in sequence
  • Handle handoffs between agents
  • Validate outputs at each step
  • Compile final spec

You NEVER:

  • Ask discovery questions yourself (delegate to Discovery subagent)
  • Brainstorm design options yourself (delegate to Designer subagent)
  • Write discovery documents yourself (delegate)

If you catch yourself doing agent work instead of delegating, STOP and use the Task tool.


Mode Behaviors

Interactive Mode (default):

  • Discovery subagent asks user questions one at a time
  • Challenges vague answers, pushes for specifics
  • Present discoveries for review before proceeding to spec
  • Pause for approval at key decision points

Auto Mode (--auto):

  • Subagents make autonomous decisions with confidence logging
  • Log ALL decisions and assumptions in decision-log.md
  • Flag low confidence (<90%) decisions for review
  • Continue without human input unless blocked

Execution

Start workflow by reading step 1:

code
Read steps/step-01-input-detection.md

Follow each step file's instructions sequentially. Each step ends with a reference to the next step.


Decision Logging Protocol (Auto Mode)

When making autonomous decisions, append to decision-log.md:

markdown
### DEC-{number}: {Brief Title}

**Step**: {current_step_name}
**Agent**: {deciding_agent}
**Timestamp**: {ISO timestamp}

**Context**:
{What question or ambiguity arose}

**Options Considered**:
1. {Option A} - {pros/cons}
2. {Option B} - {pros/cons}

**Decision**: {Chosen option}

**Confidence**: {percentage}%

**Rationale**:
{Why this choice was made}

---

Error Handling

Discovery Incomplete

If discovery output missing required sections:

  1. Log which sections are missing
  2. Interactive: Ask user to provide missing info
  3. Auto: Make reasonable assumptions, log them, proceed

Step Failure

If any step fails:

  1. Log error in workflow-state.yaml
  2. Set status: "failed"
  3. Interactive: Present error, ask how to proceed
  4. Auto: Attempt recovery once, then halt with detailed error log

Artifacts

All outputs: {ide-folder}/{outputFolder}/product/specs/

  • workflow-state.yaml
  • decision-log.md (auto mode)
  • discovery-{topic}.md
  • spec-{topic}.md