Consult

Consult Mode

A one-to-one conversation with an expert persona, which becomes a panel only when you invite one. You lead by default and can hand the lead to an agent when you want them to drive.

Follow the shared conventions in dpm/shared/skill-conventions.md — read that file at startup. This skill uses Session Startup, Library Check, Conversational Output, Gate Presentation and Cross-References from it.

Startup

1. Session — follow the shared Session Startup procedure with skill dpm:consult. A consultation is the one thing here worth resuming: the session's state carries the topic, the active agents in order, who holds the lead, and the discussion highlights, and without it a resumed conversation is a different one with the same participants.
2. Roster — mcp__plugin_dpm_dpm__list_agent, then mcp__plugin_dpm_dpm__read_agent with include_body for each agent brought into the room. The traits are body columns, so the list gives you names and roles and the read gives you the voice; rendering a persona off the list alone is rendering it off nothing. The roster is rows, so a persona this project added and the plugin never shipped is consultable by name with no file edit anywhere.
3. Library — follow the shared Library Check procedure with scope keyword consult.

Input

$ARGUMENTS selects who and about what:

• An agent name or role — match it against mcp__plugin_dpm_dpm__list_agent, case-insensitively, against display_name and role. Start there and ask what they would like to discuss.
• A topic with no agent — infer the most relevant agent from the roster's role values, confirm the choice before starting — "I'd suggest {icon} {display_name} ({role}) for this — start with them?" — and take a different one where the user names one. An inferred agent is a guess about expertise, and starting on it unasked spends the first exchange correcting it.
• A phrase naming something in the corpus — mcp__plugin_dpm_dpm__search with it, and a limit above what the project plausibly holds. Hits come back as an entity and an entity_id; read each through the tool for its entity, passing include_body — mcp__plugin_dpm_dpm__read_document_section, mcp__plugin_dpm_dpm__read_requirement, mcp__plugin_dpm_dpm__read_acceptance_criterion, mcp__plugin_dpm_dpm__read_story_criterion, mcp__plugin_dpm_dpm__read_finding, mcp__plugin_dpm_dpm__read_observation or mcp__plugin_dpm_dpm__read_task — and use what they say as context. The index covers requirement, criterion, finding, observation and task text as well as section bodies, so a term used once inside a criterion is reachable, which is most of what a consultation is actually looking for.
• A URL — fetch it and use the content as context. External, so it is read rather than queried.
• Nothing — present the roster and ask who and what about.

The conversation loop

For each message: check for a command first, then respond.

Commands

• Invite — "bring in the architect". Resolve against the roster by display_name or role, mcp__plugin_dpm_dpm__read_agent with include_body for their traits — personality and communication_style are withheld by default, and they are the traits — append them to the active list, and have them introduce themselves and respond. Already present is a no-op, said once.
• Dismiss — "thanks Priya, you can go". Remove them. If they held the lead it returns to the user, said aloud. Dismissing the last agent does not end the consultation — ask who to bring in instead.
• Lead transfer — "Margot, take the lead". Invite them first if they are not in the room, then mark their turns (leading). A leading agent asks probing questions, proposes structure and challenges assumptions; the others defer and speak only when genuinely additive.
• Exit — "wrap up", "done", "end consultation".

Distinguish a command from conversation. "I'd like to dismiss that idea" is not a dismissal. Where it is ambiguous, treat it as conversation.

Use "wrap up" as the exit word. exit and quit are the CLI's and end the session outright, which would leave the consultation unsaved.

Responses

{icon} **{display_name}**: {response in character}

Render each voice from that agent's stored personality and communication_style, and from nothing else — a trait not on the row is a trait invented for them. One agent means a dialogue, not a panel: no multi-voice formatting, no roster header. Several means each responds in roster position order, the first-invited agent first, and each shorter than they would be alone. Agents answer the user rather than each other, and disagree where they genuinely differ — that tension is most of why a second agent was invited.

Research before asking. A question about the corpus is answered with mcp__plugin_dpm_dpm__search and the read tools; a question about code is answered by reading it. Ask the user about intent, priorities and decisions — the things only they hold.

After each round, offer the exit quietly: "Type wrap up to end the consultation, or carry on."

Saving the discussion

On exit, acknowledge in one sentence and write it down:

1. mcp__plugin_dpm_dpm__create_discussion with a title and a slug from the topic. The number is allocated; do not supply one.
2. One mcp__plugin_dpm_dpm__create_document_section per part of the record — the key points, the decisions with their reasoning, and the thread that was open when it ended — at the positions they should read in.
3. mcp__plugin_dpm_dpm__create_document_agent per agent who took part, with document_kind set to discussion and the agent name — everyone who was in the room, including any dismissed before the end.
4. mcp__plugin_dpm_dpm__update_session to close the run.

Write the substance, not a summary of it. The session state has been carrying decisions with their reasoning; a discussion record that compresses them to bullets throws away the thing it exists to keep. Length is the wrong economy here.

Who was in the room is a row, and naming them in the prose is not the same fact. A reader can see the names either way; nothing else can. A run asking which discussions an agent took part in reads rows, and a persona mentioned in a paragraph is invisible to it — which is also why a dismissal is not a deletion here. The prose still reads however it needs to; it is not the record.

Then offer, and do not run: /dpm:discover, /dpm:spec or /dpm:epics with this discussion as their starting context, or nothing.

Degradation

Output

A discussion document, its sections and one document_agent row per participant, plus the session row. Nothing else is written, and no file is composed by hand.

Next Action

Offer the pipeline handoff above. A consultation earns itself by what someone does next with it.