Skillify

Create a well-structured, reusable Amplifier SKILL.md skill file that captures a repeatable process from the current session so it can be invoked again later via /skill-name. The skill must conform to the Agent Skills specification with Amplifier extensions.

Inputs

• $ARGUMENTS: (Optional) Description of the process to capture as a skill.

Steps

1. Consult Skills-Assist

Before writing any skill content, load the authoritative skills reference:

load_skill("skills-assist")

Ask skills-assist about:

• Current frontmatter field reference and best practices
• Fork vs inline decision criteria
• Step annotation conventions
• Testing patterns for the type of skill being created

skills-assist is the source of truth for Amplifier skill conventions. The examples in this skill are illustrative samples — consult skills-assist for the complete and up-to-date specification.

Success criteria: You have loaded and consulted skills-assist for the latest skill authoring conventions.

2. Analyze the Session

Before asking the user anything, analyze the conversation history to identify:

• What repeatable process was performed
• What the inputs and parameters were
• The distinct steps in order
• The success criteria and artifacts for each step (not just "writing code" but "an open PR with CI passing")
• Where the user corrected or steered you — these are important design signals
• What tools were used (read_file, write_file, bash, glob, grep, delegate, etc.)
• What agents were delegated to
• What the goals and measurable outcomes were

Success criteria: You have a clear mental model of the process, its steps, inputs, outputs, and success criteria.

3. Interview the User

Output density rule: Group related decisions into natural clusters — not one question per turn (tedious) and not everything at once (overwhelming). Present your analysis first, let the user absorb it, then ask related questions together. For example, present identity/routing decisions as one cluster, execution model decisions as another.

Calibrate interview depth to the complexity of the process. A simple 2-step workflow needs 1-2 rounds. A complex multi-step workflow with parallel tasks and irreversible actions needs the full treatment.

Round 1: High-level confirmation

• Suggest a name and description for the skill based on your analysis. Ask the user to confirm or rename.
• Suggest high-level goals and specific success criteria.
• Present the high-level steps you identified as a numbered list.

Round 2: Details and arguments

• If the skill needs arguments, suggest them based on what you observed.
• Ask whether this skill should run inline (in the current conversation) or forked (context: fork) as an isolated subagent. Forked is better for self-contained tasks that don't need mid-process user input; inline is better when the user wants to steer mid-process.
• Ask where the skill should be saved:
  • This project (.amplifier/skills/<name>/SKILL.md) — project-specific
  • Personal (~/.amplifier/skills/<name>/SKILL.md) — follows you across projects
  • The skills bundle (amplifier-bundle-skills/skills/<name>/SKILL.md) — if contributing to the curated collection

Round 3: Per-step breakdown (complex processes only)

Skip this round for simple skills with obvious steps. For complex skills:

• What does each step produce that later steps need?
• What proves each step succeeded?
• Should the user confirm before irreversible actions?
• Are any steps independent and could run in parallel?
• How should each step be executed? (direct, delegate to an agent, user action)
• What are hard constraints or preferences?

Pay special attention to places where the user corrected you during the session. These corrections are the most valuable design signals.

Round 4: Final confirmation

• Confirm when this skill should be invoked and review trigger phrases for the description field.

…