Propose Feature Concept Skill

Create a feature concept document that pitches the feature idea and explains the problem it solves. This is the first step in the feature proposal process.

Roles

You (reading this file) are the orchestrator.

⚠️ You do NOT read files in phases/ — pass them to subagents via context_files. Subagents read their phase file and execute it.

When to Use

• User wants to propose a new feature for crumbly, forester, or other forest crates
• Starting to document a feature idea
• Need to explain "why" and "what" before diving into requirements

Prerequisites

• User has described the feature idea

Optional: Offer Idea Honing First

Before starting the orchestrator loop, ask the user:

Would you like to use an idea honing process to bring more clarity to the concept? This involves working through questions one at a time to explore ambiguities and design considerations.

If yes, use the idea-honing skill following the protocol in skills/README.md. Return to this skill after idea honing is complete.

Orchestrator Loop

import json

workspace = f"planning/{feature_slug}"
bash(f"mkdir -p {workspace}", on_error="raise")

while True:
    result = bash(f"python3 skills/propose-feature-concept/next-step.py {workspace}", on_error="raise")
    action = json.loads(result)
    
    if action["type"] == "done":
        setup = json.loads(fs_read("Line", f"{workspace}/setup.json", 1, -1))
        review = fs_read("Line", f"{workspace}/review.md", 1, -1)
        log(f"Feature concept created at {setup['feature_dir']}/concept.md")
        log(review)
        break
    
    if action["type"] == "gate_failed":
        log(f"Gate failed: {action['reason']}")
        break
    
    if action["type"] == "spawn":
        r = spawn(
            action["prompt"],
            context_files=action["context_files"],
            context_data=action.get("context_data"),
            allow_tools=True
        )
        write("create", f"{workspace}/{action['output_file']}", file_text=r.response)

Handling Exceptions

The state machine handles the happy path. When things go wrong, exercise judgment:

Don't silently advance past failures. Either retry, fail explicitly, or document gaps.

Anti-Patterns

Phases

1. SETUP: Determine feature number, create feature name, check for idea honing document, create feature directory
2. DRAFT: Copy template and work with user to fill in concept document as narrative
3. REVIEW: Review for narrative flow, validate creation, provide recommendations

Inputs

• Feature idea description from user
• Optional: idea-honing.md from prior idea honing session

Outputs

• docs/features/NNNN-feature-name/concept.md - The feature concept document
• planning/<feature-slug>/review.md - Review and next steps

Next Steps

After creating the concept:

1. Review and refine the narrative
2. Get feedback on whether the feature is valuable
3. Once concept is solid, move to requirements using propose-feature-requirements skill