Changelog

All notable changes to Campaign Mode are documented here.

v0.4.8 — Current Release

• Profile emoji overrides — Profiles can now include an optional emoji field in frontmatter. When present, speaker tags use the profile emoji instead of the archetype’s hardcoded default. For example, Eeyore (Cat archetype in Hundred Acre Wood pack) now shows **🫏 Eeyore:** instead of **🐱 Eeyore:**. Fall back to archetype emoji when no profile or no emoji field exists.
  • All 27 built-in profile pack files updated with character-appropriate emoji
  • SPEC-CM-006-A v1.2: added emoji field to frontmatter table, updated examples, added emoji column to theme tables
  • SPEC-CM-009-A: added emoji as optional frontmatter in Required Frontmatter section
  • CLAUDE.md: speaker identification and profile name override paragraphs updated with emoji override logic
  • council.md: Steps 1, 4, and 5 updated to read and use profile emoji
  • All three NPC SKILL.md files: speaker identification sections updated with emoji override instructions
  • animal-campaign-context.md: transcript format example updated to note profile emoji usage
  • SPEC-CM-012-A v1.1: speaker tag examples updated to show profile emoji; filename convention updated to append profile name
• Transcript filename includes profile name — When a profile exists, transcript filenames now append the profile’s skin-name in round brackets: {YYYY-MM-DD}-{HH-MM}-{agent}({profile-name}).md (lowercase, hyphens for spaces). For example, 2026-02-18-14-32-cat(eeyore).md. Makes it easy to identify which character was consulted when scanning conversation logs. No brackets when no profile is assigned.
  • Updated: SPEC-CM-012-A, CLAUDE.md, animal-campaign-context.md, all NPC SKILL.md files, council.md, SPEC-CM-006-B, ADR-CM-019

v0.4.7

• Fix: Transcripts sometimes summarised instead of verbatim — Instructions said “full verbatim” but never explicitly warned against summarising. LLMs naturally compress long content unless told not to, and the template placeholder {...full exchange...} was ambiguous. Added explicit anti-summarisation language (“Do not summarise, condense, paraphrase, or omit”) to every transcript recording instruction: CLAUDE.md, animal-campaign-context.md, Gandalf SKILL.md, Guardian SKILL.md, Dragon SKILL.md. Expanded template examples to show multiple message turns instead of an ambiguous ellipsis. Updated SPEC-CM-012-A body sections table, write protocol, and template format to reinforce the requirement.

v0.4.6

• Fix: Transcript still not auto-saving — timing was impossible — The v0.4.5 instruction said “record transcript before presenting Next Perspective options”, but Next Perspective options are inline text (not a tool call). The model generates all text first, so it can’t interleave tool calls mid-response. Changed timing to: present response text (including options) first, then execute transcript tool calls in the same turn before the user responds. Applied consistently across animal extensions, Gandalf, Guardian, Dragon SKILL.md, and CLAUDE.md.

v0.4.5

• Fix: Agents not auto-saving conversation transcripts before transition options — The transcript recording instruction existed in its own section but was not cross-referenced at the point where agents actually present transition/Next Perspective options. Added explicit “ensure transcript is recorded” reminders inside the Next Perspective section (animal-campaign-context.md), Guardian Gate Decisions, and Dragon Interaction Patterns — right where the model reads when it’s about to present options.

v0.4.4

• Fix: Gandalf skipping Phase 2 (Character Profile) offer — Core Skill #6 (Conversation Transcript Recording) was incorrectly placed before Core Skill #5 (Character Profile Facilitation) in Gandalf’s SKILL.md, causing the model to skip the profile offer and jump straight to the Phase 3 transition. Moved transcript recording to after profile facilitation, restoring correct document order (1→2→3→4→5→6).

v0.4.3

• Conversation Transcript Recording — Every agent consultation now produces a full verbatim transcript saved to .campaign/conversations/. One file per session, written silently at the end of each consultation. Users can review past advice, trace reasoning evolution, and pick up context from previous sessions.
  • Transcripts use date-first filenames ({YYYY-MM-DD}-{HH-MM}-{agent}.md) for chronological sorting
  • YAML frontmatter captures agent, profile name, phase, campaign mode, and date
  • Full verbatim exchange with speaker tags, plus outcome section for verdicts and recommendations
  • Context isolation preserved: Guardian and Dragon write their own transcripts but must NOT read any — party reasoning remains off-limits to independent evaluators
  • All agents (animals, Gandalf, Guardian, Dragon, Council) record transcripts at their respective consultation endpoints
• ADR-CM-019 documents the conversation transcript recording decision
• SPEC-CM-012-A specifies the full transcript format, write protocol, isolation rules, and per-agent timing
• /continue-quest progress summary now includes 3 most recent conversation transcripts
• /continue-quest and /start-quest allowed-tools expanded to include Write and Bash for transcript recording
• /council now records a council session transcript (Step 6b)
• Updated: CLAUDE.md, animal-campaign-context.md, Gandalf SKILL.md, Guardian SKILL.md, Dragon SKILL.md, continue-quest.md, start-quest.md, council.md, SPEC-CM-006-B (v1.3), SPEC-CM-003-A (v1.1)

v0.4.2

• AskUserQuestion UX conventions — Hybrid approach to address Claude Desktop focus-stealing on AskUserQuestion. Phase transition questions keep AskUserQuestion with mandatory context summaries; mid-flow advisory questions switch to plain-text numbered choices.
  • Phase transitions (keep AskUserQuestion + context summary): campaign mode selection, transition to execution, transition to Dragon, Guardian approve/conditional, Dragon slain, continue-quest menu, start-quest menu
  • Mid-flow advisory (switch to plain text): animal Next Perspective handoffs, Guardian block recovery, Dragon prevails recovery
  • Every AskUserQuestion now requires a 1-2 sentence summary in the question field so users can decide without scrolling up
  • Plain-text questions use numbered options; agents parse the user’s response (number, partial label, or free-form description)
• ADR-CM-018 documents the AskUserQuestion UX conventions decision
• SPEC-CM-011-A specifies the full classification, context summary requirements, and plain-text format
• Updated: CLAUDE.md, animal-campaign-context.md, Guardian SKILL.md, Dragon SKILL.md, Gandalf SKILL.md, continue-quest.md, start-quest.md

v0.4.1

• Fix: AskUserQuestion exceeding 4-option limit — Consolidated option lists in continue-quest (7→4), start-quest (6→3), animal-campaign-context (5→4), and gandalf-agent animal menu (6→4). Related options are now grouped with follow-up questions where needed.

v0.4.0

• Proactive animal engagement in Phase 3 — Animals are no longer passive tools. Four new mechanisms make the advisory council an active team during Campaign Execution:
  • Recommended first advisor — Gandalf analyses quest characteristics and recommends which animal to consult first at Phase 3 entry (risk → Cat, timeline → Owl, vision → Bear, motivation → Puppy, resources → Rabbit, collaboration → Wolf)
  • Next Perspective handoffs — After every Phase 3 consultation, animals use AskUserQuestion to suggest the next advisor based on what was discussed, with an archetype complement fallback table
  • Proactive trigger detection — Each archetype detects trigger signals for other archetypes (e.g., Cat detects risk mentions, Owl detects timeline concerns) and prioritises them in suggestions
  • Party Assignments — quest.md now includes a table mapping each success criterion to primary and secondary animal advisors, written by Gandalf during Phase 1
• ADR-CM-017 documents the proactive party engagement decision
• SPEC-CM-010-A specifies all four mechanisms in detail
• Animal campaign extensions (extensions/animal-campaign-context.md) expanded with three new sections: Next Perspective, Proactive Engagement, Criterion Progress Awareness
• Gandalf Phase 3 transition now offers a recommended first advisor instead of generic options
• quest.md template includes Party Assignments table between Anticipated Dragon and Progress Log
• SPEC-CM-001-B updated with Phase 3 animal engagement paragraph
• README revision — Rewrote README for accessibility and engagement: new elevator pitch and hook, consolidated Quick Start (removed duplicate Installation section), new “What a Quest Looks Like” walkthrough example, second-person voice throughout. Moved changelog to CHANGELOG.md, architecture details to docs/ARCHITECTURE.md. Reduced from 535 to ~200 lines.

v0.3.2

• Family & Parenting pack — New 9-profile theme: The Elder, The Teenager, The Family Therapist, The Neighbour Kid, The Co-Parent, The Older Sibling + The Paediatrician, The Social Worker, Great Aunt Betty
• Third built-in pack added to Gandalf Phase 2 dialogue, SPEC-CM-006-A theme tables, and SPEC-CM-009-A built-in packs

v0.3.1

• Custom profile theme selection now offers Suggested (Gandalf designs a set fitting the quest’s setting and tone) instead of Fantasy, which is now covered by theme packs
• Retained Neutral (professional roles) and Custom (describe your own vibe) options

v0.3.0

• Profile Packs — Pre-built theme templates selectable during Gandalf’s Phase 2 flow. Pick a pack and all 9 profiles (6 animals + 3 NPCs) are installed instantly.
• Fantasy pack — Complete 9-profile D&D-inspired set: Paladin, Rogue, Sage, Bard, Artificer, Warden + Archmage, Sentinel, Ancient Wyrm
• Hundred Acre Wood pack — Complete 9-profile Pooh-inspired set: Pooh, Eeyore, Owl, Tigger, Rabbit, Piglet + Christopher Robin, Kanga, Heffalump
• Unified profile format (SPEC-CM-006-A v1.1) — All agent types now use the same frontmatter: archetype/skin-name/theme. Dropped explicit depth field (inferred from content). Relaxed body section naming. Migration guide included for v1.0 profiles.
• ADR-CM-016 documents the profile packs decision
• SPEC-CM-009-A defines the profile pack directory structure, file requirements, and contribution process
• Gandalf Phase 2 dialogue now offers “Install a theme pack” alongside vanilla and custom options
• CONTRIBUTING.md expanded with profile pack submission guidelines
• Added .gitignore to campaign-mode repo to properly exclude .campaign/ state directory

v0.2.8

• Added hero image to README — D&D concept art tavern scene featuring all Campaign Mode agents
• Critical fix: Plugin injection now works for all installation methods — replaced !`cat` with !`echo` + Read tool to bypass Claude Code’s sandbox restriction (GitHub #9354). Affects all 12 injection points across 4 command files.
• /campaign-setup now begins with a Gandalf welcome greeting explaining Campaign Mode, all agent roles (NPCs, animals, Simon), and what to expect
• Simon is now offered as a mid-campaign option in /continue-quest and /start-quest for meta-analysis and reflection
• Expanded “The Supervisor (Simon)” section in README to clarify the Gandalf/Simon distinction and Simon’s three campaign touchpoints
• ADR-CM-015 documents the plugin injection sandbox fix
• Updated SPEC-CM-007-C (v1.2), SPEC-CM-007-D (v1.1), SPEC-CM-008-A (v1.1) with new injection syntax
• See ROADMAP-CM-003 for full details

v0.2.7

• Gandalf SKILL.md now has mechanical Phase 3 progress tracking instructions (Read/Write tool steps) matching the detail level of Phase 1/2 — fixes Gandalf not updating quest.md during campaign execution
• Animal agents (Bear, Cat, Owl, Puppy, Rabbit, Wolf) now receive campaign extensions when invoked through quest commands — enabling quest-aware advice, progress tracking, and phase transition awareness
• New extensions/animal-campaign-context.md defines campaign behaviour extensions for animal agents, injected into /continue-quest, /start-quest, and /council via dynamic context injection
• Six Animals SKILL.md files gain a generic “Context Extensions” stub, priming animals to respect supplementary instructions from any orchestration context
• ADR-CM-014 and SPEC-CM-008-A document the animal campaign extensions approach and cross-project dependency
• See ROADMAP-CM-002 for full details

v0.2.6

• CLAUDE.md campaign conventions now injected into all command invocation paths (/start-quest, /continue-quest, /council) via dynamic context injection
• NPC SKILL.md files now include a condensed “Campaign Conventions” section covering cross-cutting rules (identity, lifecycle, elicitation, debrief) for direct skill invocations
• Corrected SPEC-CM-007-B: plugin CLAUDE.md is NOT auto-loaded by the plugin infrastructure
• ADR-CM-013 and SPEC-CM-007-D document the hybrid injection + condensed subset approach
• See ROADMAP-CM-001 for full details

v0.2.5

• Claude Desktop installation instructions in Quick Start and Installation sections
• start-quest and continue-quest now invoke animal agents via the Skill tool instead of attempting to read skill files from the filesystem — animal skills belong to the Six Animals plugin, not Campaign Mode

v0.2.4

• Plugin commands now work in Claude Desktop via dynamic context injection (!`command` syntax with ${CLAUDE_PLUGIN_ROOT})
• Commands inject skill definitions at invocation time instead of referencing file paths at runtime
• ADR-CM-012 and SPEC-CM-007-C document the approach

v0.2

• /council command — multi-perspective project diagnostic from all six animal agents with Simon synthesis, persistent report in .campaign/council-report.md
• /campaign-setup now offers structured next steps (start quest, convene council, just exploring) instead of slash command references
• /continue-quest detects council reports and offers reconvene option in both no-quest and mid-campaign paths
• Quick Start guide in README

v0.1

• Gandalf, Dragon, and Guardian skill definitions with mode-aware behaviour
• Campaign mode selection: Grow, Ship, Grow & Ship
• Character generation system — two-depth profiles (flavour + behavioural modifiers), theme-agnostic, user-assigned, stored in .campaign/profiles/
• Proactive elicitation — agents offer structured next-step options at every phase transition via AskUserQuestion; no slash command references in user-facing text
• Campaign Debrief Protocol — Dragon-to-Simon handoff with mode-aware context
• Dragon Voice and Tone — imposing, terse, authoritative confrontation presence
• Plugin packaging for Claude Code marketplace distribution
• CLAUDE.md campaign guidelines loaded per session
• /campaign-setup command for project onboarding
• /start-quest and /continue-quest commands for ergonomic quest entry and mid-campaign re-entry
• Quest state persistence — campaign mode, success criteria, and progress log stored in .campaign/quest.md for durable state across sessions
• Speaker identification — every agent response begins with an emoji + bold name tag; profile name overrides apply when character profiles are assigned
• User-as-protagonist framing throughout
• Architecture Decision Records documenting key design choices (ADR-CM-001 through ADR-CM-011)
• Design specifications for each agent, lifecycle, mode profiles, context isolation, character profiles, and plugin structure
• North-star vision document with open questions

Future

• Simon’s campaign-mode extensions (GM responsibilities, expanded debrief)
• Alternative cultural and thematic mappings
• Multi-quest campaign arcs
• Extended NPC roster

See docs/north-star.md for the full vision and open questions.