ic-15u: PR#5 revision: Rewrite Spellkave docs into canonical PRD
Snapshot: 2026-03-28T14:48:09Z
| Field | Value |
|---|---|
| Status | in_progress |
| Assignee | main-agent |
| Priority | 2 |
| Labels | priya |
| Created by | github-bridge |
| Created | 2026-03-28T09:20:31Z |
| Updated | 2026-03-28T10:33:40Z |
Description
PR revision needed: b4arena/spellkave#5 URL: https://github.com/b4arena/spellkave/pull/5 Title: Rewrite Spellkave docs into canonical PRD
Review feedback:
Review
The direction is right — separating product vision from implementation detail is good spec hygiene. The new PRD is clean, focused, and properly tech-stack agnostic. Three requests before merge, plus design observations for the PRD content itself.
Requests
1. Archive deleted content — don't discard it
The five deleted files contain ~950 lines of research, decisions, and structured thinking (risk matrix with mitigations, user personas, 40+ Phase 0 tasks with agent ownership, resource links to Mnehmos/ElizaOS/SpacetimeDB/Project Sid, etc.). This is institutional knowledge that took effort to produce.
Please move them to docs/archive/ rather than deleting outright. They'll serve as input for future Tier-2 implementation specs.
2. Rename "AI Dungeon" reference
docs/ai-dungeon-deep-research.md still carries the old project name. The README labels it "AI Dungeon Deep Research." Now that the project is Spellkave, rename the file (e.g., docs/spellkave-deep-research.md) or at minimum update the README label.
3. Clarify where implementation specs live next
The PRD's Boundaries section explicitly excludes tech stack, phases, acceptance criteria, and roadmap sequencing. Good — but the PR should state where these will live. A sentence in the PR description or a follow-up issue would suffice: "Implementation detail will be captured in Tier-2 specs / ADRs / beads."
Without this, the team has a vision document but no pointer to the next layer.
Design observations for the PRD
These aren't blockers, but they surface tensions the PRD should eventually address — either inline or as named open questions.
A. Death Stranding and asynchronous social design
The PRD's "Human and AI Coexistence" (§4.2) and "Observer Appeal" (§4.5) sections describe a world where presence is asynchronous and consequences outlast sessions. This is exactly the design space Death Stranding explored:
- Strand mechanics: Players never see each other directly, but they reshape the world for everyone. Ladders, bridges, roads, shelters — all persist. In Spellkave terms: a party that clears a bandit camp, builds a waystation, or establishes a trade route leaves infrastructure that other players and agents inherit.
- Shared infrastructure without coordination: DS proved that meaningful social play doesn't require synchronous presence. Players cooperated at civilizational scale (road-building) without ever meeting. Spellkave's AI agents could serve this role continuously — maintaining, degrading, and contesting shared structures.
- Entropy as a design force: DS's Timefall ages everything. Without maintenance, the world degrades. This maps directly to Spellkave — if nobody patrols a road, monsters reclaim it. If a shop isn't resupplied, it closes. Persistence isn't just "things stay" — it's "things decay without effort."
- Emotional bonds through traces: DS creates deep connection between players who never meet, through "likes," signs, and shared cargo. Spellkave could surface agent chronicles, graffiti, journals, or reputation echoes that make absent players feel present.
Death Stranding 2: On the Beach pushes this further with persistent shared bases and community-driven world evolution — worth tracking as a reference.
Suggestion: Add Death Stranding to the domain context (§5) as a design reference for asynchronous social persistence. It's the strongest non-MMO precedent for what Spellkave is attempting.
B. The time problem — always-on world vs. human-paced interaction
The PRD says the world "continue[s] evolving even when any single player is absent" (§4.1). This creates a fundamental temporal tension that the document doesn't yet name:
- AI agents act at machine speed. An agent can take 50 actions in the time a human reads a room description.
- D&D is turn-based. The source material assumes synchronous, sequential play. An always-on world can't pause for a human to deliberate while 30 agents blaze through combat rounds.
- Absence has consequences. If the world is truly persistent, logging off for 8 hours means the world moved on without you — possibly dramatically. Your shop was raided. Your faction lost territory. Your party disbanded.
This isn't a bug — it's a design lever. But it needs to be named as an open question. Possible approaches range from time dilation (the world slows near human players) to phasing (humans interact with a temporally buffered view) to full acceptance (the world moves on, deal with it — the Death Stranding approach).
Suggestion: Add an open question in §4.1 or §5 acknowledging the temporal pacing tension between always-on AI and session-based humans.
C. "Things happen only once" — world uniqueness as product principle
The PRD's "Persistent Consequences" section (§4.4) says "actions should leave traces." This understates what might be the single most important design principle: world events are unique and irreversible.
In most MMOs, every player kills the same dragon. The dragon respawns. The quest resets. In Spellkave, if the premise holds: the dragon is dead. The quest is gone. The loot was claimed by whoever got there first.
This creates:
- Real scarcity — resources, quests, and encounters are finite. Strategy matters.
- Real stakes — failure isn't "try again," it's "that opportunity is gone."
- Real history — "the day the dragon died" becomes shared lore, not a repeatable instance.
- A content consumption problem — unique events get permanently consumed. The world needs regenerative systems (new threats emerge, factions shift, consequences create new conflicts) or it runs out of things to do.
This is the product's sharpest differentiator and its hardest design challenge. The PRD should name it explicitly — something like: "World events are singular. The same quest cannot be completed twice. This creates genuine stakes but requires the world to continuously generate new situations from the consequences of resolved ones."
Suggestion: Elevate "world uniqueness / singular events" from an implication of §4.4 to a named principle, and flag the regenerative content question as an open design challenge.
Summary
Three concrete requests (archive, rename, clarify next tier), plus three design observations (Death Stranding precedent, temporal pacing tension, world uniqueness principle) that would strengthen the PRD's framing of its hardest problems.
Steps:
- Read the review feedback above and on the PR
- Make the requested changes
- Push to the PR branch
- Request re-review when done