askill
refine-prompt

refine-promptSafety 75Repository

Transforms rough, incomplete prompts into clear, actionable instructions with full codebase context awareness. Analyzes the repository to identify relevant files, patterns, and constraints. Assigns appropriate execution persona, defines negative constraints (what NOT to do), performs self-critique to catch gaps, and confirms intent with user before finalizing.

0 stars
1.2k downloads
Updated 3/22/2026

Package Files

Loading files...
SKILL.md

Refine Prompt

Transform rough prompts into clear, actionable instructions grounded in codebase context.

When to Use

Use this skill when:

  • A prompt is vague or incomplete
  • You need codebase-aware context added to a request
  • Complex tasks need decomposition into sub-tasks
  • You want persona assignment for task execution
  • You need negative constraints (what NOT to do) defined

Quick Start

/refine-prompt "add caching to the api"

Core Workflow

  1. Context Analysis - Analyze repo structure, find relevant files and patterns
  2. Intent Extraction - Parse the prompt, expand implicit goals, fix errors
  3. Persona Assignment - Assign appropriate expertise level and role
  4. Reference Injection - Add specific file paths, functions, patterns
  5. Negative Constraints - Define anti-patterns and forbidden approaches
  6. Decomposition - Break complex requests into ordered sub-tasks
  7. Validation - Check feasibility, identify risks and edge cases
  8. Self-Critique - Review for completeness, ambiguity, quality
  9. Intent Confirmation - Verify understanding with user
  10. Output - Save refined prompt and execution log

Parameters

Parameters are defined in .octon/framework/capabilities/runtime/skills/registry.yml (single source of truth).

This skill accepts one required parameter (raw_prompt) and three optional parameters for execution mode, context depth, and confirmation behavior.

Output Location

Output paths are defined in .octon/framework/capabilities/runtime/skills/registry.yml (single source of truth).

Outputs are written to .octon/framework/scaffolding/practices/prompts/ (refined prompt) and /.octon/state/evidence/runs/skills/refine-prompt/ (execution log).

Boundaries

  • Never change the core intent of the original prompt
  • Always preserve explicit user preferences
  • State all assumptions explicitly
  • Reference only files that actually exist
  • Write only to designated output paths
  • Always perform self-critique before finalizing
  • Confirm intent unless explicitly skipped

When to Escalate

  • Unresolvable contradictions in the prompt
  • Completely unclear intent
  • Referenced files don't exist
  • Request conflicts with project constraints
  • Scope exceeds 20 files
  • Self-critique reveals major issues

References

For detailed documentation:

Install

Download ZIP
Requires askill CLI v1.0+

AI Quality Score

52/100Analyzed 3/29/2026

Refine-prompt skill with solid framework structure and good workflow definition, but heavily dependent on external references for actual execution details. Located in deep internal Octon framework path (.octon), making it project-specific despite the broadly applicable concept. The 10-step workflow, boundaries, and escalation criteria are well-defined, but the actual phase-by-phase instructions, parameters, and output paths are all deferred to external files (registry.yml, references/*), limiting standalone actionability.

75
78
60
55
45

Metadata

Licenseunknown
Version-
Updated3/22/2026
Publisherjamesryancooper

Tags

apigithub-actionsprompting