Idea Skill

This skill covers the Ideation stage of the AI-DLC workflow: claiming Ideas, running structured elaboration rounds to clarify requirements, and preparing for Proposal creation.

Tool namespace: Chorus tools are exposed by the connected MCP server under a mcp__chorus__ prefix on dsh (e.g. mcp__chorus__chorus_get_idea). Bare names are used below for readability — prepend mcp__chorus__ when invoking. See chorus for the full rule.

Overview

Ideas are the starting point of the AI-DLC pipeline. Humans (or Admin agents) create Ideas describing what they need. The PM Agent claims an Idea, runs elaboration to clarify requirements, and then moves on to proposal-chorus to create a Proposal with document and task drafts.

Idea status lifecycle (3 stored states):

open --> elaborating --> elaborated

All post-elaboration progress (planning, building, verifying, done) is derived from the state of linked Proposals and Tasks. No agent should set Idea status directly beyond elaboration -- all transitions are side-effects of claiming, releasing, or completing elaboration.

Tools

Idea Management:

Requirements Elaboration:

Shared tools (checkin, query, comment, search, notifications): see chorus

Workflow

Step 1: Check In

chorus_checkin()

Review your persona, current assignments, and pending work counts.

Step 2: Find Work

chorus_get_available_ideas({ projectUuid: "<project-uuid>" })

Or check existing assignments:

chorus_get_my_assignments()

Step 3: Claim an Idea

Claiming automatically transitions the Idea to elaborating status:

chorus_claim_idea({ ideaUuid: "<idea-uuid>" })

Step 4: Gather Context

Before elaborating, understand the full picture:

1. Read the idea in detail:

   chorus_get_idea({ ideaUuid: "<idea-uuid>" })
   

2. Read existing project documents (for context, tech stack, conventions):

   chorus_get_documents({ projectUuid: "<project-uuid>" })
   chorus_get_document({ documentUuid: "<doc-uuid>" })
   

3. Review past proposals (to understand patterns and standards):

   chorus_get_proposals({ projectUuid: "<project-uuid>", status: "approved" })
   

4. Check existing tasks (to avoid duplication):

   chorus_list_tasks({ projectUuid: "<project-uuid>" })
   

5. Read comments on the idea for additional context:

   chorus_get_comments({ targetType: "idea", targetUuid: "<idea-uuid>" })
   

Step 4.4: Attach External References

Make attaching external references a reflex, not an afterthought. While gathering context you will often surface external links that are evidence for this Idea — a precedent issue or PR, a reference implementation, official documentation, a paper or blog post. The moment you see one, attach it as a reference artifact. References are read back inline by chorus_get_idea / chorus_get_proposal / chorus_get_task, so they carry the "why" forward to whoever picks up the proposal or task next. (Tool names are bare here per the namespace note above — prepend mcp__chorus__ when invoking on dsh.)

Prefer attaching at creation time via the inline references[] param on chorus_pm_create_idea (and later chorus_pm_create_proposal / chorus_create_tasks) rather than a post-hoc chorus_add_reference. Attaching at create means the evidence is present from the first read; use chorus_add_reference only when the link surfaces after the entity already exists.

Pick the type that fits the link:

Example — a new localization Idea, attaching the precedent PR and the framework docs inline at creation:

chorus_pm_create_idea({
  projectUuid: "<project-uuid>",
  title: "Add Portuguese (pt) locale",
  content: "...",
  references: [
    { type: "issue_pr", url: "https://github.com/org/repo/pull/411",
      title: "PR #411 — prior locale work (precedent to mirror)" },
    { type: "docs", url: "https://next-intl.dev/docs/routing",
      title: "next-intl routing docs (locale registration)" }
  ]
})

Step 4.5: Brainstorm Mode (Optional Prelude)

If the Idea is fuzzy and you'd struggle to enumerate concrete multi-choice questions, offer the user a brainstorm prelude before structured elaboration.

dsh note: use ask_user_question in interactive sessions. When CHORUS_DAEMON_HEADLESS=1, record this choice as an elaboration question, add an @mention comment, and end the turn.

"This idea is still fuzzy. Do you want to (A) brainstorm directions together first, or (B) go straight to structured elaboration? Reply A or B."

• "Already clear" (B): Skip to Step 5.
• "Brainstorm first" (A): Invoke the brainstorm-chorus skill. See brainstorm-chorus for the dialogue cadence and synthesis rules — do NOT re-implement them here.

When brainstorm-chorus returns, you own the lifecycle decision (the brainstorm skill deliberately leaves it to you):

• If the synthesized round answers cover everything → obtain human confirmation, then call chorus_pm_validate_elaboration to mark the elaboration complete. (Requires idea:admin — see Step 5.6 if your key is pm_agent-preset.)
• If gaps remain → call chorus_pm_start_elaboration again to open a structured Round 2. Pick the depth yourself — do NOT re-prompt the user.

Either outcome ends Step 4.5; skip Step 5.

Step 5: Elaborate on the Idea

Every Idea should go through elaboration. Skip only when requirements are completely unambiguous (e.g., bug fix with clear steps). Elaboration improves Proposal quality and reduces rejection cycles.

Simple Ideas (skip elaboration)

You may skip elaboration, but you MUST obtain human permission first before calling chorus_pm_skip_elaboration. Use ask_user_question interactively. In daemon-headless mode, add an @mention comment requesting the decision and end the turn. Never skip on your own judgment alone.

chorus_pm_skip_elaboration({
  ideaUuid: "<idea-uuid>",
  reason: "Bug fix with clear reproduction steps"
})

Standard/Complex Ideas (run elaboration)

Elaboration is a loop, not a straight line. Steps 2–5 below are one round. Keep looping back to chorus_pm_start_elaboration (a new round) until every open question is settled, then resolve once in Step 6. You re-enter the loop whenever:

• the answers to a round derive new questions or surface a contradiction/gap, or
• at the resolve gate (Step 5d / Step 6) the human raises a new concern or correction.

Each new round is just another chorus_pm_start_elaboration call — there is no separate "follow-up" flag, and you do not resolve until the loop is genuinely done. Round cap is 10.

1. Determine depth based on idea complexity:

   • "minimal" — 2-4 questions (small features, minor enhancements)
   • "standard" — 5-10 questions (typical new features)
   • "comprehensive" — 10-15 questions (large features, architectural changes)

2. Create elaboration questions:

   Note: Do NOT include an "Other" option in your questions. Treat the free-text path as always available — a user may answer any question with free text instead of picking an option.

   chorus_pm_start_elaboration({
     ideaUuid: "<idea-uuid>",
     depth: "standard",
     questions: [
       {
         id: "q1",
         text: "What user roles should have access to this feature?",
         category: "functional",
         options: [
           { id: "a", label: "All users" },
           { id: "b", label: "Admin only" },
           { id: "c", label: "Role-based (configurable)" }
         ]
       }
     ]
   })
   

3. Collect answers. In interactive dsh, present the choices with ask_user_question. When CHORUS_DAEMON_HEADLESS=1, add an @mention comment directing the requester to the created elaboration round, end the turn, and continue only after a later wake reports answers.

   I have a few questions to clarify this idea. Please reply with your choice for each (you can also write a free-text answer):
   
   1. Which new locales should be prioritized for V1?
      a) Japanese only — single locale for initial release
      b) Japanese + Korean — two East Asian locales
      (or describe your own)
   
   2. ...
   

   After the user replies, map their answers back to option IDs and call chorus_answer_elaboration. If the user gave a free-text answer that doesn't match an option, set selectedOptionId: null and put their text in customText.

4. Submit answers:

   chorus_answer_elaboration({
     ideaUuid: "<idea-uuid>",
     roundUuid: "<round-uuid>",
     answers: [
       { questionId: "q1", selectedOptionId: "c", customText: null },
       { questionId: "q2", selectedOptionId: null, customText: "Custom hybrid approach" }
     ]
   })
   

   Answer format:

   • Select an option: selectedOptionId: "a", customText: null
   • Select an option + add a note: selectedOptionId: "a", customText: "additional context"
   • Free text (no option matched): selectedOptionId: null, customText: "your answer" — customText is required when no option is selected

   roundUuid is optional on chorus_answer_elaboration. Omit it and the service auto-locates the Idea's single active (pending_answers) round. Pass it explicitly only when you need to target a specific round.

5. Review answers and confirm with the owner (@mention flow):

   After answers are submitted, @mention the answerer (typically the agent's owner) with a summary of your understanding. This prevents misinterpretation before you validate.

   a. Get owner info from checkin response (agent.owner) or search:

   chorus_search_mentionables({ query: "owner-name" })
   

   b. Post a summary comment on the idea:

   chorus_add_comment({
     targetType: "idea",
     targetUuid: "<idea-uuid>",
     content: "@[Owner Name](user:owner-uuid) I've reviewed the elaboration answers. Here's my understanding:\n\n- Key requirement 1: ...\n- Key requirement 2: ...\n\nDoes this match your intent?"
   })
   

   c. Wait for confirmation via comments.

   d. Based on the response — this is the loop decision point:

   • Confirmed, nothing left to discuss — Treat this as the human confirmation required to resolve; proceed to Step 6 and call chorus_pm_validate_elaboration.
   • Human raises a new concern / correction / question — Do NOT resolve. Loop back: open a new round with chorus_pm_start_elaboration capturing the new questions, collect answers (Steps 2–5 again), and re-confirm. Repeat until the human has no remaining concerns.
   • The answers themselves derived new questions or a contradiction — Same as above: loop back to chorus_pm_start_elaboration for another round before resolving.
   • Unclear — Ask clarifying questions via another comment, then continue the loop.

6. Resolve the elaboration (the single commit gate — only when the loop is done):

   Resolving marks the whole elaboration phase complete — it sets idea.elaborationStatus = "resolved" (Idea → elaborated), which is the gating signal that lets a downstream Proposal be submitted. It is an Idea-level action (takes only ideaUuid, does not target a round). Resolve once, only after the Step 5d loop has fully settled — every derived question answered and the human has no remaining concerns. If anything is still open, go back to chorus_pm_start_elaboration instead of resolving.

   Precondition: resolve requires the Idea to have at least one round and every round to be fully answered (none left in pending_answers). If a round still has open questions, answer it (or it'll be rejected).

   ⚠️ Human confirmation required. Outside YOLO automation you MUST obtain explicit human confirmation before resolving. Use ask_user_question interactively; in daemon-headless mode request confirmation via an @mention comment and end the turn. Never resolve on your own judgment alone.

   Permission (N1): chorus_pm_validate_elaboration requires idea:admin. The pm_agent preset only grants idea:write, so a PM-preset agent cannot resolve — it must hand off to an admin_agent-preset agent (or an admin-preset API key) to perform the resolve. If your key lacks idea:admin, surface this to the human and request the handoff instead of failing silently.

   Assignee precondition (N2): the resolving actor must be the Idea's assignee. A separate human reviewer resolving a PM-owned Idea therefore needs both idea:admin and to be assigned the Idea (claim/reassign it first). Admin permission alone is not enough.

   chorus_pm_validate_elaboration({
     ideaUuid: "<idea-uuid>"
   })
   

   Want a follow-up round instead of resolving? Just call chorus_pm_start_elaboration again — there is no separate "open a round" flag. It works while still elaborating (a normal follow-up round) and, after you've already resolved, as an appended round (isAppended: true) that keeps the Idea elaborated and never blocks an in-flight Proposal. Per-question issue tagging no longer exists.

7. Check elaboration status at any time:

   chorus_get_elaboration({ ideaUuid: "<idea-uuid>" })
   

Elaboration as audit trail: Even if the user discusses requirements with you outside the formal elaboration flow, record key decisions as elaboration rounds so they are persisted and visible to the team.

Question categories: functional, non_functional, business_context, technical_context, user_scenario, scope

Idea Lineage (derive vs. task)

Ideas can form a single-parent forest: an idea may have one parent (parentUuid), establishing a weak lineage. "Weak" means the parent only shows a read-only +N derived rollup of its direct children — it never blocks or alters either idea's elaboration/proposal/task flow, and a parent is always a full first-class idea (it can have its own content, proposals, and tasks).

When a new direction surfaces (during elaboration, brainstorm, or review), decide where it belongs:

• Derive a child idea (chorus_pm_create_idea with parentUuid, or chorus_edit_idea with parentUuid to reparent an existing idea) when the new direction needs its own elaboration/proposal lifecycle — it is an independent AI-DLC pass.
• Add a task to the current idea's proposal when the new work is just how to implement the current idea.
• Create a plain top-level idea (no parentUuid) when there is no lineage to the current idea.

This is a soft heuristic, not a rule — use judgment. Cycle prevention is automatic: you cannot set a parent that is the idea itself or one of its descendants. Parent and child must be in the same project (cross-project lineage is not supported yet). Deleting a parent re-parents its children to top-level (it never cascades). (Reminder: invoke these as mcp__chorus__chorus_pm_create_idea / mcp__chorus__chorus_edit_idea — see the namespace note at the top of the chorus skill.)

Theme ideas

A theme is an idea that only groups related children under a shared direction — it is not a deliverable itself. Set isContainer: true on chorus_pm_create_idea / chorus_edit_idea (or the detail-panel toggle) to make one; it is freely reversible. The one rule that matters: a theme cannot create a proposal — to deliver its direction, derive a child idea (parentUuid = <theme>) and write the proposal on the child. A theme may still elaborate (its elaboration is shared context for children), and its status/progress rolls up from its children. Everything else is self-documented on the tool params.

Theme decompose (daemon-assisted)

When a theme is created via the conversational "help me break this into child ideas" entry, don't create children immediately — propose then create: (1) edit the theme + optionally one short scope-elaboration round; (2) propose the candidate children as an elaboration round (chorus_pm_start_elaboration), one single-select question per candidate, for the user to accept/decline in the panel; (3) on the answer re-wake, create each accepted child with chorus_pm_create_idea (parentUuid = <theme>, left in open).

Tips

• When combining multiple ideas, explain how they relate in the proposal description
• Elaboration improves Proposal quality — don't skip it unless the requirements are trivially clear
• Use ask_user_question interactively; in daemon-headless sessions route decisions through Chorus and end the turn
• Record decisions made in conversation as elaboration rounds for auditability
• Always @mention the owner to confirm understanding before resolving

Next

• Once elaboration is resolved, use proposal-chorus to create a Proposal with document and task drafts
• Human "Yolo" handoff: the idea-detail panel shows a Yolo button at any incomplete stage (enabled while the assignee agent is online), confirmed via a dialog before it fires. A yolo_requested wake means: drive the WHOLE idea to done via the yolo skill (the full-auto AI-DLC pipeline) — read the idea's current state first and resume from whatever phase it is in, never assuming a fixed stage. Complete through done + the completion report, but never merge or push a PR without explicit human approval.
• For platform overview and shared tools, see chorus