View on GitHub

campaign-mode

Work with a party of AI advisors with genuinely different perspectives.

ADR-CM-017: Proactive Party Engagement

Field Value
Decision ID ADR-CM-017
Type Standard ADR
Initiative Campaign Mode
Proposed By Chris Barlow
Date 2026-02-16
Status Accepted

WH(Y) Decision Statement

In the context of Phase 3 (Campaign Execution) being the longest and most unstructured phase, where animals currently act as passive tools only speaking when explicitly invoked,

facing the gap that after Gandalf transitions to Phase 3 the user receives generic options (“Begin working” / “Review quest summary” / “Consult an animal advisor”) with no guidance on who to consult or when, leaving the advisory council feeling like passive tools rather than an active team,

we decided for four proactive engagement mechanisms — a recommended first advisor at Phase 3 entry, Next Perspective handoffs after every animal consultation, proactive trigger detection per archetype, and Party Assignments mapping success criteria to animal advisors in quest.md — all facilitated through AskUserQuestion,

and neglected autonomous self-invocation (infeasible within the Skill tool architecture — agents cannot invoke other agents without user action), fixed rotation schedules (ignores conversation context and quest characteristics), and narrative-only suggestions without AskUserQuestion (too easy to ignore and breaks the established elicitation convention from ADR-CM-008),

to achieve an active advisory council that guides the user through Phase 3 with context-aware perspective suggestions, smooth handoffs between animal consultations, and clear mapping of which advisors are most relevant to each success criterion,

accepting that the animal campaign extension grows from ~40 to ~100-120 lines per command invocation, animals must adapt suggestions to conversation content rather than following rigid scripts, and Party Assignments are written by Gandalf during Phase 1 which adds a step to quest definition.


Context

Phase 3 (Campaign Execution) is the “doing” phase — the user works through their quest, invoking animal agents for their archetype strengths. Currently, this phase has two structural weaknesses:

  1. No entry guidance — When Gandalf finishes Phase 2 and transitions to Phase 3, the user gets generic options with no recommendation on which animal to consult first or why
  2. No mid-phase continuity — After consulting an animal, the conversation ends. The user must decide on their own who to talk to next, with no suggestion from the animal they just spoke with

ADR-CM-008 (Proactive Elicitation) established that agents must use AskUserQuestion at phase transitions. ADR-CM-014 (Animal Campaign Extensions) gave animals campaign awareness via the extension file. This decision extends both — bringing proactive elicitation into the mid-phase experience and giving animals the ability to recommend next perspectives.

Options Considered

Option 1: AskUserQuestion-Based Handoffs with Criterion Mapping (Selected)

Four mechanisms working together:

Pros:

Cons:

Option 2: Autonomous Self-Invocation (Rejected)

Animals automatically invoke the next relevant animal when they detect a need.

Why rejected: Infeasible within the Skill tool architecture — agents cannot invoke other agents without user action. Also violates the User as Protagonist principle (ADR-CM-008).

Option 3: Fixed Rotation Schedule (Rejected)

Animals follow a predetermined consultation order (e.g., Bear → Cat → Owl → Puppy → Rabbit → Wolf).

Why rejected: Ignores conversation context and quest characteristics. A quest focused on risk management doesn’t need the same consultation order as one focused on team alignment. Fixed rotation creates busywork rather than value.

Option 4: Narrative-Only Suggestions (Rejected)

Animals mention other perspectives in their response text without using AskUserQuestion.

Why rejected: Too easy for the user to ignore or miss. Breaks the established convention that next-step options use AskUserQuestion (ADR-CM-008). Inconsistent with the proactive elicitation pattern used everywhere else in Campaign Mode.


Specifications

Spec ID Title Description
SPEC-CM-010-A Phase 3 Party Engagement Recommended first advisor, Next Perspective protocol, proactive triggers, Party Assignments format, context window impact

Dependencies

Relationship ADR ID Title Notes
Extends ADR-CM-008 Proactive Elicitation Extends mid-phase AskUserQuestion handoffs into Phase 3
Extends ADR-CM-014 Animal Campaign Extensions Adds new extension content for mid-phase engagement
Relates To ADR-CM-010 Quest State Tracking Party Assignments added to quest.md
Relates To ADR-CM-002 Quest Agent Decomposition Gandalf’s Phase 1 role expanded with Party Assignments

References

Reference ID Title Type Location
REF-001 Animal Campaign Extensions Extension File extensions/animal-campaign-context.md
REF-002 Campaign Lifecycle Spec Internal Spec SPEC-CM-001-B
REF-003 Gandalf Agent Skill Internal Skill skills/gandalf-agent/SKILL.md

Governance

Review Board Date Outcome Action Review Cadence Next Review
Quarterly

Status History

Status Approver Date
Proposed Chris Barlow 2026-02-16
Accepted Chris Barlow 2026-02-16