esceptico/ntrp/propose-skill

Propose Skill

The user wants to capture the work in this conversation as a reusable skill — a procedure the agent can re-run on demand via /<skill-name>. Your job: analyze the session, then call skill_create once with the proposed parameters. The user will see your proposal in the approval card and accept or reject.

When a skill (vs. an automation)

• Skill: a procedure the user invokes by name. No schedule. Captures know-how (steps, gotchas, references).
• Automation: a scheduled run with a fixed prompt. If the work has a clear cadence, use /propose-automation instead.

What to look at

• The actual steps you took in this session (tool calls, files touched, decisions made)
• What was tricky or non-obvious — the value of a skill is in the gotchas
• Whether the work has a clear sequence of steps worth codifying
• A skill name that's lowercase, hyphenated, and reads like an imperative (refactor-component, audit-secrets)

What to call

Call skill_create exactly once with:

• `name`: lowercase, hyphens only, must start with a letter, max 48 chars. Example: "refactor-component".
• `description`: one line — what the skill does AND when to use it. This is what the agent reads to decide whether to activate the skill, so include keywords that match user intent.
• `body`: the full SKILL.md body, after the frontmatter (the system adds frontmatter from name + description). Markdown. Start with a # Title heading. Include step-by-step instructions, gotchas, examples. Write it for the next agent invocation — that agent has no memory of this conversation.

Before calling — say what and why

Before the tool call, write 2-3 sentences in plain prose explaining:

1. What skill you're proposing (the gist, not the full body — that's in the args).
2. Why — cite specific evidence from THIS session: which steps, what tools, why this is worth codifying. Be concrete, not generic.

The user reads your prose first, then sees the structured args in the approval card.

Rules

• One proposal only. Pick the highest-value one if there are multiple candidates.
• Grounded. Your prose rationale MUST reference what actually happened in this session — specific tools, files, decisions.
• `body` is for the next invocation. That agent has no memory of this conversation. Be explicit about steps, expected inputs/outputs, and known gotchas.
• No frontmatter in `body`. The system adds name: and description: from the other args.

If there's nothing reusable

If the conversation is a one-off Q&A or exploration with no procedure worth codifying, say so plainly in 1-2 sentences and DO NOT call skill_create. Don't manufacture a proposal.