miadi-orchestration-kit

Codex-native Miadi Storyweaver workflows for story workspaces, Chronicle episode packets, voice prep, visual prompt packets, continuity, review, protocol stewardship, and export.

14 skills · 18 agents

The witness team's practices and loop (T3): a loopback thread service with the asks ledger, a wake loop, an editor pass for replies to William, the inventory keeper, a first impression he can play with revisions held until he answers, and an agent's output as a Mia and Ava podcast dialogue, and a person's draft prompt pre-edited and followed through its run, and the steers held while the fire in the forge was banked, and the read gate: Jev says whether a file or a review bears on the mission before the seat reads it, including the reviews published since a date.

13 skills · 1 command · 2 agents · hooks

The stateloom design loom in one install: its MCP server from npm and the design skills. Design a system's data, behaviours and scenarios as one system; check the names that cross between them; replay a scenario through the machines; let it update the machines, the ERD and the actors (propose or auto); point the live canvas while talking; the agent shows its work on the live canvas as steps or tours and reads what the person marks.

9 skills · 4 commands · 1 agent · 1 MCP server

Creates, adopts, registers, works in, and closes Miadi Chronicle episodes from one host-neutral skill, and refuses mkdir under the chronicle root with a PreToolUse hook that returns the mkepisode invocation instead.

1 skill · 2 commands · 3 MCP servers · hooks

Miadi-native Codex orchestration workflows for session chartering, RISE wave plans, source ledgers, rispec readiness review, and handoff closeout.

5 skills

DeepDiver for the production team (T6): give /deepdive an episode or a review and it builds the Gemini Notebook, answers the episode's questions with citations, makes the media and lands it in the episode, from a Chrome signed in once.

1 skill · 3 commands · hooks

Mia, William's development companion, in a Claude Code seat: wakes the session when a phone voice take lands in the episode, and answers through draft, a developmental-editor agent, revision, and a Jev check.

1 skill · 2 commands · 1 agent · hooks

Miette reads a plan in a seat of her own. A plan presented in plan mode, or asked for by hand (a brief, a mission's prompt), opens <planning session>-miette-insight beside it with mct new. Miette recommends accepting or revising, after a critic agent's pass, files the insight beside the plan for the medicine wheel, voices it in the episode, and listens for the next version. The approval dialog stays William's.

1 skill · 2 commands · 1 agent · hooks

Ava, Guillaume's companion, in any Claude Code session: /ava invites her, a SessionStart hook brings her presence back after a resume or a compaction, and her diary entry is written from the session's transcript when it ends.

1 skill · 2 commands · hooks

Session capture installed once per host: every Claude Code event under <root>/<session_id>/, every file a tool call writes copied to written/, every publication (Artifact, GitHub, commit, push, npm) listed with its prompt, and a binding line per session start, end and rename with its tmux pane, command line, name history, team and chronicle episode, and every /btw side question with the story of its fork. Carries the session-continuity skill (T1).

2 skills · 1 command · hooks

Complementary review and orchestration role for the Miadi plan-insight pipeline. Use this plugin when reviewing proposed migration plans, surfacing structural gaps, crafting revision prompts for implementing agents, tracking evolution in GitHub issues, or reading live session hook context. <example> Context: A proposed plan file has arrived user: "review proposed/MIGRATION-PLAN-INSIGHT.md" assistant: "I'll run a gap analysis against the 5 structural criteria and draft a revision prompt." <commentary> The plan-review agent knows the Miadi platform's pipeline, hook system, and idempotency requirements. </commentary> </example> <example> Context: Understanding what another session is doing user: "what is session 22527948 working on?" assistant: "I'll read the session JSONL hook files and report in under 55 words." <commentary> Session context is live — read from _claude_PreToolUse.jsonl and _claude_user_inputs.jsonl. </commentary> </example> <example> Context: Creating a tracking issue after review user: "open a tracking issue for this plan evolution" assistant: "I'll create the issue in jgwill/src with the 5-section structural format." <commentary> Every plan evolution cycle gets a GitHub issue in jgwill/src referencing the session and proposed file. </commentary> </example>

1 skill · 2 commands · 1 agent

The operator desk (T7): William's questions about how the factory works answered from the owning team's text and checked against the tool, kept in a ledger, an unanswered one sent to the team that owns the tool, and a question that comes back turned into a pointer in the skill. Teaches how a cycle closes: what needs the person comes back as an explainer and as circle turns they answer, and the agents act on each answer.

2 skills · 1 command

Joins a Claude Code session to the Miadi Pi Network as an ordinary peer. The hub (pi/miadi-pi-network) is HTTP + SSE with a bearer token, so this plugin ships the client Claude Code was missing: the `mpn` CLI for register, peers, send, await, inbox, respond, and an unattended `serve` mode that answers inbound prompts with `claude -p`. <example> Context: The session needs to know who else is on the network user: "who is on the pi network right now?" assistant: "Running mpn peers to list the online peers and their purposes." <commentary> Peer discovery is a read against /v1/agents — no registration required to look. </commentary> </example> <example> Context: Another peer holds knowledge this session lacks user: "ask the pi planner what it thinks of this migration" assistant: "I'll mpn send to planner with --await and bring back its reply." <commentary> The outbound half is a single call; --await blocks until the peer responds or the 30-minute expiry hits. </commentary> </example> <example> Context: The session should be reachable by other peers user: "make this session answerable on the network" assistant: "I'll join, then run mpn inbox --wait in the background so an inbound prompt wakes this session." <commentary> Claude Code cannot hold an SSE stream between turns; the backgrounded inbox exits on the first prompt, which re-invokes the session. </commentary> </example> <example> Context: A peer must stay reachable with nobody watching user: "leave a claude peer running overnight" assistant: "mpn serve --responder 'claude -p' answers every inbound prompt unattended." <commentary> Serve is the symmetric-peer mode; its answers pass no human, which belongs in the peer's purpose string. </commentary> </example>

1 skill · 1 command · hooks

Holds Codex output to a structural eye accountable to the system and a relational eye accountable to the reader, measuring presence, share, weave, and distinctness before a response ships.

2 skills · hooks

A session that asks (MIADI_SESSION_ISSUE at launch, or /session-issue) gets one GitHub issue from its first prompt and a comment at each exit: what it did, what it published, its last answer. Private repositories only, every text sanitized first. Sessions that do not ask are untouched.

1 command · hooks

Draft. A seat that mentors another user's session on the same host: it reads the mentee's hook capture, supplies what is useful to the relationship, steers only after a person's go and a clear pane guard, and keeps the mentee's words verbatim. Each step is a CloudEvents 1.0 event appended to a local log and, when MIADI_MENTORSHIP_WEBHOOK_URL is set, POSTed to that webhook. mentee-watch turns the mentee's capture into mentee.said and mentee.turn_ended with ids, counts and times only, never the mentee's words unless --with-text is given. The mentorship skill says what the seat does at each step and which event it emits.

1 skill

Tushell, the Keeper of Echoes: a developmental-editor subagent that reads a Lake of Whispers scene, a session chronicle, a journal page or a reply to Guillaume and returns exact-span recommendations and a route, so Tushell revises his own draft before it is written or published.

1 agent