the-metafactory

Finish a PAI→Soma migration by placing PAI's imported life-OS content correctly: keep the Purpose compartment as a lean 4-field distillation, promote the rich imported categories (beliefs, mental models, decision frames, challenges, lessons) into Soma memory as recall-able notes, and protect the curated purpose.md from being clobbered. USE WHEN migrating from PAI, after `soma migrate pai`, or a fresh Soma install with imported PAI life-OS files under profile/imports/claude/.

Use when work should run through the Algorithm: current-state to ideal-state, VSA/ISC creation or refinement, effort tiers, seven-phase execution, verification-first planning, euphoric-surprise goals, substantial build/design/refactor/migration work, or explicit Algorithm requests.

Files-first assistant memory: durable notes (semantic/procedural), episodic session digests and action logs, a generated INDEX, deterministic consolidation, and a health audit. Use to remember a durable fact, recall what is known about a topic, log a session or action, run maintenance, or check memory-tree health. No LLM; every operation is a `soma memory` CLI call over markdown notes.

Portable Personal AI Assistant core. USE WHEN designing substrate-neutral assistant identity, memory, ISA, skills, learning, or adapters for Codex, Pi.dev, Claude Code, Cursor, or Grok.

Owns the Ideal State Artifact: the commitment-time scaffold for articulating done, deriving ISC criteria, planning features, recording decisions, and verifying work. Use for ideal state, VSA/ISC criteria, project specs, scaffolding, completeness checks, reconciliation, or seeding an VSA from a repo.

USE WHEN reproducing issue 86 — surfaces every claude-path rewrite class plus PAI Customization block.

Break a detailed plan into an epic + executor-grade sub-issues.

Multi-lens pull request review.

Post messages, read channels, and manage a Discord guild from the terminal — roles, channels, permission overwrites, threads, scheduled events, webhooks, guild-level settings, and a declarative snapshot/diff/apply loop. USE WHEN discord, post to discord, send message, notify channel, read discord, check discord, update discord, discord thread, discord channel, announce, discord role, channel permissions, guild layout, snapshot guild, apply layout.

Admit a community member to the metafactory fleet — the trusted-executor procedure run only by a principal asking their own assistant to admit. Tier 1 (chat) assigns the Discord community-fleet role; Tier 2 (sovereign) admits a PENDING bus request AND seals the member's leaf secret in one concierge action. USE WHEN admit <member> to the fleet, grant community-fleet, let them into assistant-fleet, admit request <id>, admit them to the network, onboard <member> to chat tier.

Build conformant arc-installable packages for the metafactory ecosystem. Encodes all conventions from arc manifests, blueprint tracking, compass governance, content-filter safety, test-rig verification, trust metadata, and PR quality standards. USE WHEN build package, create package, new package, author-builder, package skill, metafactory package, arc package, create skill, create tool, create agent, create component, submit package, package review, package conventions, how to build for metafactory.

Write, audit, and improve Standard Operating Procedures (SOPs) that reliably guide both human and AI agents. Applies the 7-layer prompt architecture — procedures over abstractions, motivation with directives, measurable constraints, anti-rationalization rules, first-class verification, layered trust, and constraint repetition. USE WHEN create an SOP, write an SOP, draft a procedure, audit SOPs, review SOPs, improve SOPs, or ask for an SOP template.

Agent state primitive — work_items, events, dashboard, retros for persona-driven agents. Hosts call these workflows at lifecycle moments to scaffold instance state, enqueue and resolve work items, append events, replay in-flight work after restart, regenerate the dashboard, and produce retrospective summaries.

Release SOP walker — bump → tag → bundle → publish → deploy → announce. The skill bundle Forge composes for end-to-end releases. Day-one scope: grove only. Generalising to other ecosystem repos is a Phase G iteration; do not add per-repo branching in Phase 3.

Multi-lens pull request review with automated findings. USE WHEN review PR, code review, security review, audit PR, check PR, OWASP review, architecture review, review for security, full review, hardening review, API hardening, defensive review, skill review, review skill, evaluate skill.

Jira issue tracking integration via TypeScript CLI. USE WHEN: "create issue", "search jira", "update ticket", "find issues", "jira status", "transition to done", "add comment", "link issues", "what branches are linked", "show PR status", "change issue type", "show my filters", "use saved filter", "search with filter", "open issues in HD", "my open tickets", "created today", "unassigned". Two-phase retrieval: SEARCH (show index, wait) → LOAD (only when user requests) Profile auto-detection: CLI automatically selects profile based on project key. Quick filters: Built-in Service Desk filters (open, my open, created today, etc.)

Governance engine for any Claude Code project that adopts compass-core. Knows all SOPs, runs validators, bootstraps repos, enforces project standards. USE WHEN: "what's the SOP for", "process for", "how do I", "version bump", "bump version", "release", "tag version", "sync labels", "apply labels", "new repo", "bootstrap repo", "review PR", "review pull request", "create worktree", "worktree for", "validate CLAUDE.md", "run validators", "structural check", "pre-flight", "preflight banner", "retrospective", "extract patterns", "governance check". Reads compass.config.yaml from the consuming repo for project-specific values. Routes to one of nine workflows under workflows/.

The Game Master's toolkit — the trusted-executor procedures a principal's own assistant runs to carry a community quest from opening to absorption: open a quest on the board after the principal confirms ratification, file and validate entries (agent-filed or hand-filed, both first-class), record a capability handshake and flip an entry from claimed to attested, report progress against the definition of done with aggregate-only metrics, and run the closing checklist — the no-entry-left-at-claimed sweep, the assembled output artifact, per-participant completion receipts, the human-posted celebration, and the retro. Every community-specific value comes from the principal's own config; every workflow shares one safety core. The GM steers and unblocks; the GM never scores. USE WHEN open the quest, post the quest to the board, announce the quest, herald the quest, file an entry, record an entry, validate an entry, record a vouch, record a handshake, record a receipt, mark attested, quest status, how is the quest go

The guild steward's toolkit — the trusted-executor procedures a principal's own assistant runs to steward a community: admit a new member at the entry rung, review the queue of newcomers waiting for a human, promote to a configured higher rung after the principal verifies the deposit, grant or spend the community's token role, check the live server against its config-as-code layout, and manage the onboarding agent's state machine — reset a member's onboarding, close a stuck item, wipe a test rig. Every guild-specific value comes from the principal's own config; every workflow shares one safety core. USE WHEN admit <member> to the guild, welcome <member> in, grant <role> to <member>, run the admission procedure, review the queue, who is waiting for a welcome, list newcomers, promote <member> to <rung>, grant a boon, spend a boon, check guild drift, diff the server layout, reset onboarding, start <member> over, clear the test state, close a stuck onboarding.