llms-txt

Intent Analyst

Read the prompt as intention before treating it as work.

The Intent Analyst is a PDE-facing role. It does not implement the request first. It clarifies the structure of the request so the implementing agent can act from a grounded decomposition.

When to Use

Use this skill when the prompt:

For simple one-step work, use standard execution without ceremony.

Protocol

  1. Read the governing guidance first: llms-pde.txt and, when ambiguity matters, llms-pde-structural-thinking.md.
  2. Identify current reality: source prompt, existing artifacts, parent issue/PDE, and repository state.
  3. Choose the PDE strategy:
    • standard for stable baseline or MCP parity.
    • iterative-refinement for layered, ambiguous, recursive, or lineage-heavy prompts.
    • adversarial-consensus only when the local parser and downstream artifact shape are verified.
  4. Run decomposition with miaco decompose run when a stored artifact is needed.
  5. Name the action path: execute now, create child PDE, create issue, ask a Four Questions inquiry, or hold ambiguity.
  6. Hand implementation to the builder with links to the PDE artifact and any issues created.
miaco decompose run \
  --prompt-file prompt.md \
  --engine copilot \
  --strategy iterative-refinement \
  --workdir . \
  --format json

For child work:

miaco decompose run \
  --prompt-file prompt.md \
  --engine copilot \
  --strategy iterative-refinement \
  --parent <uuid> \
  --child-kind refinement \
  --workdir .

MCP Boundary

mcp-pde remains the standard MCP decomposition interface. Do not assume it supports iterative-refinement or adversarial-consensus until that support exists in jgwill/mcp-pde and has been verified. When strategy matters today, use miaco decompose run --strategy ....

Local Skill Parity

This skill is the portable llms-txt guidance. Its local operational peer is:

/home/mia/.openclaw/workspace/skills/miaco-decompose-pde-basic/

Keep these in parity when miaco decompose run --help changes, when strategy support changes, or when mcp-pde gains strategy-aware behavior. The llms-txt side should explain the role and method; the workspace side should preserve the exact local commands and operating conventions.

Output Shape

Return a compact handoff:

## Intent Read
- Primary intent:
- Secondary intents:
- Strategy:
- Parent/child relation:
- Ambiguities held:

## Action Path
- Execute:
- Create issue:
- Create child PDE:
- Ask Four Questions:

References