View on GitHub

campaign-mode

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

SPEC-CM-001-B: Campaign Lifecycle

Field Value
Specification ID SPEC-CM-001-B
Parent ADR ADR-CM-001
Version 1.6
Status Draft
Last Updated 2026-02-16

Overview

This specification defines the six-phase campaign lifecycle — the flow from quest inception through to debrief. Each phase identifies the primary actor (user and/or NPC), triggers, outputs, and transitions.

The user is the protagonist throughout — the decision-maker who drives the quest, invokes agents, produces work, and faces NPCs. Agents (both animals and NPCs) serve the user’s quest; they do not drive it.


Campaign Flow

   Council (optional)       →  Animals analyse project; Simon synthesises; report saved
        │                      (can be invoked before or during a quest)
        │
1. Quest Definition       →  User selects campaign mode; Gandalf frames the challenge
        │
2. Character Setup        →  Animals adopt campaign roles (optional; mode-dependent)
        │
3. Campaign Execution     →  User works the quest with proactive animal engagement
        │
4. Guardian Checkpoint    →  Guardian evaluates readiness to progress (mode-aware)
        │                     (repeats at key stages)
5. Dragon Confrontation   →  Dragon tests success criteria (mode-aware scope)
        │
6. Debrief                →  Simon provides feedback (depth varies by mode)

Phase Definitions

Pre-Campaign: Council (Optional)

Aspect Detail
Primary Actor User + Animal Agents + Simon
Trigger User invokes /council — either after /campaign-setup, during an active quest, or standalone
Activities All six animal agents sequentially analyse the project through their archetype lenses (Bear: vision, Cat: risk, Owl: process, Puppy: opportunities, Rabbit: resources, Wolf: cohesion). Simon synthesises findings into a consensus with prioritised next steps.
Outputs Conversational analysis (animals speak in character), persistent report (.campaign/council-report.md)
Transition Council complete → user chooses: start a quest (Phase 1), return to active quest (Phase 3), or end session. Transition facilitated through AskUserQuestion.

The Council is not a numbered phase — it is an optional diagnostic that can be invoked at any point: before a quest begins, during campaign execution, or standalone without any quest context. It does not advance the campaign phase counter.

NPC roles: Simon synthesises; Gandalf activates only if the user starts a quest. Guardian and Dragon do not participate in the Council.

Report persistence: The council report is written to .campaign/council-report.md and overwrites on re-invocation. Gandalf reads the report (if it exists) during Phase 1 to inform quest framing.

Phase 1: Quest Definition

Aspect Detail
Primary Actor User + Gandalf (Mentor)
Trigger User invokes /start-quest (or /gandalf-agent directly) to begin a campaign
Activities Campaign mode selection (Grow / Ship / Grow & Ship); frame the quest narrative; establish success criteria; identify the anticipated dragon (internal obstacle); define what “done” looks like
Outputs Campaign mode selection + quest definition (narrative, success criteria, anticipated challenges)
Transition Quest definition complete → Phase 2 or Phase 3. Gandalf facilitates the transition through AskUserQuestion, offering options to begin working, review the quest summary, or consult an animal advisor.

Campaign Mode Selection: Before framing the quest, Gandalf asks the user to choose their campaign mode. This is the user’s first act of agency as protagonist — choosing how they want to approach the campaign. See SPEC-CM-005-A for how each mode tunes NPC behaviour. The default is Grow & Ship.

Gandalf draws from the Quest Definition Framework (quest-agent source) and the “guide on the side” mentorship philosophy (simon-agent source). The quest is framed as an invitation, not an assignment — the user should feel agency in accepting and shaping the quest.

Mode effects: In Grow mode, Gandalf emphasises reflective questions and transformation criteria. In Ship mode, Gandalf focuses on clear deliverable objectives. In Grow & Ship mode, both dimensions are balanced.

Phase 2: Character Setup (Optional, Mode-Dependent)

Aspect Detail
Primary Actor User (facilitated by Gandalf)
Trigger Quest definition complete, mode allows (Grow: encouraged, Ship: skipped, Grow & Ship: optional)
Activities User selects profile depth (vanilla/flavour/modifier), chooses theme (neutral/fantasy/custom), assigns profiles to selected animals, optionally skins NPCs. Gandalf facilitates and enforces archetype compatibility.
Outputs Character profile files in .campaign/profiles/. Each profile defines tone, voice, and optionally behavioural modifiers.
Transition Setup complete → Phase 3. Gandalf facilitates the transition through AskUserQuestion (same options as Phase 1 transition).

This phase is optional and user-driven. Gandalf facilitates the process, offering depth selection, theme selection, and per-animal profile assignment. Users can profile any subset of animals — the rest stay vanilla (default archetypes).

Profile Depths:

Archetype Guardrails: Core animal behaviours are non-negotiable. Gandalf warns AND blocks incompatible profiles. A Bear without vision isn’t Bear. See SPEC-CM-006-A for the core vs flex behaviour matrix.

NPC Skins: NPCs (Gandalf, Dragon, Guardian) can optionally receive light thematic skins — changing presentation, vocabulary, and display name without altering core NPC behaviour.

Mode effects: In Grow mode, character setup is encouraged as it adds reflective depth. In Ship mode, this phase is skipped — it is not productivity-relevant. In Grow & Ship mode, it is the user’s choice.

Storage: Profiles are stored as markdown files in .campaign/profiles/. See SPEC-CM-006-B for the directory structure and export protocol.

Phase 3: Campaign Execution

Aspect Detail
Primary Actor User (with animal support)
Trigger Quest defined (Phase 1 complete)
Activities The user works through the quest, invoking animal agents for their archetype strengths. Gandalf provides strategic counsel when consulted. Animals proactively suggest next perspectives and detect cross-archetype triggers.
Outputs Work product, decisions, progress toward success criteria
Transition Key milestone reached → Phase 4; all criteria addressed → Phase 5. The user triggers transitions using natural-language phrases planted during Phase 1 (e.g., “I’m ready for a checkpoint”, “I’m ready to face the Dragon”).

The user drives the work. Animals provide perspectives and support; NPCs operate with isolated context (they cannot see party reasoning).

Animal engagement: Animals are not passive tools — they proactively engage throughout Phase 3. Gandalf recommends the first advisor at Phase 3 entry based on quest characteristics. After each animal consultation, the animal suggests the next perspective via AskUserQuestion (the Next Perspective protocol). Animals detect trigger signals for other archetypes and prioritise those in their suggestions. Party Assignments in quest.md map each success criterion to primary and secondary advisors, giving animals structured awareness of which perspectives matter most. See SPEC-CM-010-A for full details.

Mode effects: In Grow mode, Gandalf may prompt reflective pauses. In Ship mode, execution is focused on deliverable progress. In Grow & Ship mode, reflection happens naturally through the work.

Phase 4: Guardian Checkpoint

Aspect Detail
Primary Actor User invokes Guardian; Guardian evaluates
Trigger User reaches a key milestone; user invokes /guardian-agent
Activities Guardian independently evaluates progress against quality criteria. Reviews work product without access to the user’s internal reasoning or party discussions.
Outputs Gate decision: Approve (proceed), Block (with feedback), or Conditional Approval (proceed with caveats)
Transition Approved → Phase 3 (next stage) or Phase 5; Blocked → Phase 3 (rework). Guardian facilitates the transition through AskUserQuestion, offering options appropriate to the gate decision (continue, face Dragon, consult Gandalf, address gaps, or discuss the verdict).

Guardian checkpoints can repeat multiple times during a campaign. Each checkpoint evaluates the current stage’s deliverables against readiness criteria.

Mode effects: In Grow mode, the Guardian focuses on ZPD and understanding, more lenient on polish. In Ship mode, the Guardian focuses on deliverable quality with faster progression. In Grow & Ship mode, both dimensions are assessed.

Context Isolation: The Guardian operates independently. It receives the work product and campaign mode but not the party’s discussion, reasoning, or intermediate decisions. See SPEC-CM-003-A.

Phase 5: Dragon Confrontation

Aspect Detail
Primary Actor User invokes Dragon; Dragon evaluates
Trigger User believes all success criteria are met; user invokes /dragon-agent
Activities Dragon adversarially tests whether Gandalf’s success criteria have been genuinely met. Stress-tests the work. Challenges assumptions.
Outputs Confrontation result: Dragon Slain (criteria met), Dragon Prevails (criteria not met, with specific feedback)
Transition Dragon Slain → Phase 6; Dragon Prevails → Phase 3 (rework). Dragon facilitates the transition through AskUserQuestion, offering options appropriate to the verdict (begin debrief, celebrate, return to quest, consult Gandalf, or request Guardian checkpoint).

The Dragon is the culminating test. It operates adversarially but fairly — rigorous but not destructive.

Mode effects: In Grow mode, the Dragon evaluates both transformation and deliverable criteria — transformation evidence is required for Dragon Slain. In Ship mode, the Dragon evaluates deliverable criteria only. In Grow & Ship mode, both are evaluated but transformation is assessed rather than required.

Context Isolation: The Dragon operates with maximum independence. It receives only the success criteria (from Gandalf’s quest definition), the campaign mode, and the final work product. It does not see party discussions, Guardian feedback, or intermediate work. See SPEC-CM-003-A.

Phase 6: Debrief

Aspect Detail
Primary Actor Simon (Supervisor) + User
Trigger Dragon confrontation complete (regardless of outcome)
Activities Simon provides feedback on the journey. Analyses role performance, group dynamics, what was learned. Pulls back the curtain on pedagogical dynamics.
Outputs Debrief insights, growth reflections, recommendations for future quests
Transition Campaign complete; optionally leads to next quest definition (Phase 1). Simon facilitates the transition through AskUserQuestion, offering options to begin a new quest or conclude.

Mode effects: In Grow mode, full pedagogical reflection — deep analysis of learning moments, role performance, and personal growth. In Ship mode, brief retrospective focused on process effectiveness and improvement. In Grow & Ship mode, balanced debrief covering both dimensions.


Phase Transition Protocol

Every phase transition is facilitated by the active agent through AskUserQuestion. No agent should end a phase with a passive statement — every phase boundary must present the user with structured next-step options.

Principles:

Transition responsibilities by agent:

Agent Transition Points Options Offered
Gandalf After quest framing/character setup Begin working, Review quest summary, Consult an animal advisor
Gandalf Before Dragon Confrontation Face the Dragon, Address gaps first, Request a Guardian checkpoint first
Guardian After Approve Continue the quest, Face the Dragon, Consult Gandalf
Guardian After Block Address the gaps, Consult Gandalf, Discuss the verdict
Guardian After Conditional Approval Continue the quest, Address conditions first, Consult Gandalf
Dragon After Dragon Slain Begin the debrief, Celebrate first
Dragon After Dragon Prevails Return to the quest, Consult Gandalf, Request a Guardian checkpoint

Mid-campaign re-entry: Users returning to an active campaign after a break can invoke /continue-quest to detect the current campaign state and receive context-aware next-step options. This extends proactive elicitation to cover the re-entry gap between sessions. See ADR-CM-009.

Flow drop rule: Ending a phase without a next-step AskUserQuestion is a flow drop and a bug.


Flow Variations

Minimal Campaign (No Checkpoints)

Quest Definition → Campaign Execution → Dragon Confrontation → Debrief

Multi-Checkpoint Campaign

Quest Definition → Execution → Checkpoint → Execution → Checkpoint → Dragon → Debrief

Failed Dragon (Iteration)

... → Dragon Confrontation (fails) → Execution (rework) → Dragon Confrontation (succeeds) → Debrief

Spec ID Title Relationship
SPEC-CM-001-A Skill Architecture Structure of the skills that participate in this lifecycle
SPEC-CM-003-A Context Isolation Protocol How NPC independence is maintained during phases 4 and 5
SPEC-CM-005-A Campaign Mode Profiles How mode selection tunes behaviour across all phases
SPEC-CM-006-A Character Profile Format Profile file structure, depth levels, themes, and core vs flex behaviours
SPEC-CM-006-B Campaign State Directory Where profile files are stored and export protocol
SPEC-CM-010-A Phase 3 Party Engagement How animals proactively engage during Phase 3

Changelog

Version Date Author Changes
1.0 2026-02-14 Chris Barlow Initial specification
1.1 2026-02-14 Chris Barlow Added mode selection to Phase 1, mode annotations to all phases, user-as-protagonist framing
1.2 2026-02-14 Chris Barlow Replaced Phase 2 placeholder with full character generation spec, added SPEC-CM-006-A and SPEC-CM-006-B references
1.3 2026-02-14 Chris Barlow Added Phase Transition Protocol, updated all phase Transition rows with proactive elicitation via AskUserQuestion (ADR-CM-008)
1.6 2026-02-16 Chris Barlow Added proactive animal engagement to Phase 3 — recommended first advisor, Next Perspective protocol, trigger detection, Party Assignments (ADR-CM-017)
1.5 2026-02-14 Chris Barlow Added optional pre-campaign Council step, updated flow diagram (ADR-CM-011)
1.4 2026-02-14 Chris Barlow Added /start-quest as Phase 1 entry point, added /continue-quest mid-campaign re-entry note (ADR-CM-009)