idea

idea

First stage of the five-stage sprint workflow (idea → brainstorm/plan → work → review → compound). Captures a new idea as an atomic, per-file markdown artifact with structured YAML frontmatter, or updates an existing one. Keeps docs/ideas/README.md as a lightweight index grouped by priority.

This skill does not brainstorm or plan. It captures. Ambiguity and requirements work belongs in /plan (aliased /brainstorm), which is the next stage.

When to use

TRIGGER when:

• user says "new idea", "let's capture an idea", "add IDEA-...", "note this idea down", "log a backlog item"
• user wants to update an existing idea's status, priority, or body (e.g. "mark IDEA-042 complete", "bump IDEA-088 to high priority", "add IDEA-110 as related to IDEA-111")
• agent proactively surfaces an improvement candidate during another workflow (e.g. after /compound identifies a project-specific learning worth tracking)

SKIP when:

• the user wants to plan or brainstorm an existing idea — route to /plan <slug> instead
• the idea is clearly a one-off trivial fix that does not warrant a backlog entry — just do the work
• the project has no docs/ideas/ tree yet AND a legacy monolithic backlog file exists (IDEAS.md, BACKLOG.md, ROADMAP.md) — route to /ingest-backlog first to establish the per-idea layout

Pattern

1. Creating a new idea

Auto-increment, two-phase capture.

Phase A — establish the record.

1. Determine the next IDEA number by scanning both IDEA-file locations — docs/ideas/IDEA-*.md and docs/archive/*/IDEA-*.md — for the greatest existing three-digit number, and adding 1. Default to 001 if no files exist. Users can override with an explicit number argument (/idea 200 → forces IDEA-200). Scanning only docs/ideas/ would collide with any IDEA currently in in-progress, complete, superseded, or rejected state, all of which live in the archive tree per RULE_ideas-location-status.
2. Ask the user for title, priority (high / medium / low), and an optional depends_on / related list referencing existing IDEA ids.
3. Use the platform's blocking question tool when available (AskUserQuestion in Claude Code, request_user_input in Codex) for the priority choice. Ask one question at a time.
4. Derive the slug from the title: lowercase, kebab-case, strip stopwords (a, the, for, into), truncate to ~40 chars. Confirm with the user if the slug is ambiguous.
5. Evaluate sprint-auto eligibility and set the two gate fields with explicit reasoning. Default both to false at capture — the question isn't "can we eventually automate this?" but "can sprint-auto run this tonight, unattended, with no human in the loop?" A /plan pass later can upgrade false → true once the unknowns are resolved. Rubric: