Agent bundles / agent-core
Superpowers, skill management, and shell environment for pi.
Understanding-oriented
Why the bundle is built this way.
agent-core holds the skills that every other bundle depends on: plain-language writing rules, SOP authoring, skill authoring, shared context, and subagent spawning through muxer-subagents.
It is the only bundle with an extension (extensions/pi-shared-context) for OMP and Pi. Its SessionStart hooks add the shared-context route table and the <muxer-session> tag to each session.
Learning-oriented
Use the agent-core bundle for the first time.
mise run claude.bootstrap once, so that Claude Code installs the bundles. For OMP, Pi or Zot, apply the mise links instead./agent-core:explain git worktrees in Claude Code, or /skill:explain git worktrees in OMP.The skill is loaded when the reply names the explainer file it wrote.
Task-oriented
Steps for one task each.
Skill: muxer-subagents
The muxer-subagents skill decides how to open child agents from the muxer and agent that run the session.
muxer-subagents detect-muxer. It prints herdr, zellij, tmux, hrdx or unknown-muxer.muxer-subagents detect-agent. It prints omp, pi, zot, codex, claude or unknown-agent.mise links muxer-subagents into ~/.local/bin. Hooks use the same command, so a hook and the skill agree on the agent.
| Harness | Steps |
|---|---|
| Claude Code | Raise version in .claude-plugin/plugin.json and in the marketplace.json entry, then run: claude plugin update agent-core@zenobius-agent-bundles. Start a new session. |
| OMP | Start a new session. The symlink shows the change at once. |
| Pi | Start a new session. The symlink shows the change at once. |
| Zot | Start a new session. The symlink shows the change at once. |
Information-oriented
Read from the bundle files at build time.
| Part | Claude Code | OMP | Pi | Zot | Codex |
|---|---|---|---|---|---|
| Skillsskills/**/SKILL.md. Instructions that the agent or the user invokes. | supported | supported | supported | supported | not supported |
| Agentsagents/*.md. Subagent definitions with a role and a tool list. | supported | supported | supported | not supported | not supported |
| Hookshooks/hooks.json. Commands that run on harness events, such as SessionStart. | supported | supported | not supported | supported | not supported |
| Harness | Install | Reads |
|---|---|---|
| Claude Code | mise run claude.bootstrap (runs claude plugin install agent-core@zenobius-agent-bundles) | .claude-plugin/plugin.json, skills/, agents/*.md, hooks/hooks.json |
| OMP | mise symlinks ~/.omp/agent/bundles/agent-core to the bundle | package.json (pi key), skills/, agents/*.md, extensions/ |
| Pi | mise symlinks ~/.pi/agent/bundles/agent-core to the bundle | package.json (pi key), skills/, agents/*.md, extensions/ |
| Zot | mise symlinks ~/.local/state/zot/extensions/agent-core to the bundle | extension.json, skills/ |
26 skills
| Skill | Invoked by | Scripts | Description |
|---|---|---|---|
| artifact-shares | user and agent | yes | Routes artifact publishing to private GitHub repositories, with optional Pages sites — create, share, enable Pages later, list, sync, redact, or rebuild. Use when the user says artifact share, publish a report, share a session, or store an artifact in a private repository. Produces a checked repository change and, when Pages is enabled, a public site URL. |
| asd-ste100 | user and agent | — | Write or rewrite technical text with the rules of ASD-STE100 Simplified Technical English so it is clear, unambiguous, and free of AI slop. Use for documentation, READMEs, runbooks, procedures, error messages, release notes, incident reports, and API guides. Also use when the user says "STE", "Simplified Technical English", "ASD-STE100", "de-slop", "make this readable", "write for non-native readers", or asks for docs that translate well. Enforces the standard's 53 rules: 20/25-word sentence limits, one word one meaning, simple tenses, active voice, condition before command. |
| converting-and-parsing-documents | user and agent | — | Use Docling to understand the content of documents in any supported format — PDF (born-digital or scanned), DOCX, PPTX, XLSX, HTML, Markdown, AsciiDoc, CSV, images, audio, and XML — by converting them into a unified DoclingDocument (Markdown or structured JSON). Use this skill whenever you need to read, parse, convert, extract, or chunk a document you cannot read directly: "what's in this PDF", "convert this to markdown", "extract the tables", "chunk this for RAG", "read this scanned document", "parse this DOCX/PPTX". Covers the `docling` CLI, the Python SDK (DocumentConverter + PipelineOptions), the remote Service Client (self-hosted or managed docling-serve), and the docling-slim install extras for a minimal dependency footprint. |
| explain | user and agent | — | Research a subject and build a single-page Diátaxis explainer in minimalist HTML, when a user asks for a researched web explanation, resulting in a validated HTML page with sources |
| iso-24495 | user and agent | — | Applies plain-language rules based on the principles of ISO 24495-1 to all output for clarity and accessibility. |
| iso-24495-1 | user and agent | — | Core Plain Language standard (ISO 24495-1:2023). Governs all user-facing responses to ensure clear, structured, findable, understandable, and actionable output. |
| iso-24495-2 | user and agent | — | Sector-specific Plain Language standard for legal communication (ISO 24495-2:2025). Applied during contract drafting, license review, and legal/compliance writing. |
| iso-24495-3 | user and agent | — | Sector-specific Plain Language standard for science and technical writing (ISO 24495-3:2026). Applied during software documentation, architecture specs, and technical analysis. |
| iso-24495-4 | user and agent | yes | Provisional task skill for organisational plain language implementation (based on ISO/CD 24495-4, committee draft). Activates for plain language gap analysis, policy drafting, review workflow design, and organisational readiness for the future published standard. Does not activate for ordinary writing, rewriting, or reviewing of individual documents. |
| iso-24495-5 | user and agent | — | Provisional sector-specific Plain Language standard for document design (based on ISO/WD 24495-5, under development). Applied when structuring complex documents so readers can find and navigate content through layout, visual hierarchy, and navigation aids. |
| iso-24495-code | user and agent | — | Plain language applied to source code (ISO 24495-1:2023 principles). Governs the parts of code a person reads: the order units appear in, their names, comments, and error messages. Applied when writing or restructuring code, not when explaining it. |
| iso-24495-text-audit | user only | yes | Audit user-selected Markdown or text files for deterministic plain-language findings. Use only when the user explicitly invokes this skill. |
| lynx-web-search | user and agent | — | Use when terminal-only internet research is needed and you must search the web or read pages without browser automation, especially when relying on lynx CLI to query search engines and save readable page dumps to /tmp. |
| mq-query | user and agent | — | Use the mise-installed harehare/mq command to query, read, transform, and create Markdown or structured content, with the harehare/okf.mq extension always loaded for OKF documents and bundles. Use when an agent needs mq syntax, CLI flags, cookbook patterns, or OKF-aware file work. Routes each task to a focused procedure and the bundled upstream mq documentation. |
| muxer-subagents | user and agent | yes | Spawn a child agent in a terminal multiplexer pane, tab, or workspace. Detects the active muxer (herdr, hrdx, zellij, tmux) and agent CLI (claude, omp, pi, zot), then resolves the contract that says how to open, launch, wait, read, and close a session. |
| result-delivery | user and agent | yes | Delivers validated skill results through explicit caller handoffs, machine JSON or short human summaries. Use for operation output, child result artifacts, acknowledgements and interrupted delivery. Keeps domain receipts separate from transport and presentation. |
| review-documentation | user and agent | — | Reviews supplied documentation for structure, coverage, clarity, accuracy, and usability; use for README and technical documentation audits; produces prioritised findings and exact edit recommendations. |
| searxng-web-search | user and agent | — | Searches the web through the local SearXNG instance using curl and renders readable results with lynx. Use for private terminal-based web searches through search.gameserver.lan. |
| shared-context | user and agent | yes | The shared-context skill resolves the alignment root, the shared-context store, and the store for tickets, evidence and worktree logs. It reports the route table and alignment root, and it resolves a record to its store, path, and stable reference. |
| skill-hunter | user and agent | — | Find and download skills. Use when you need to discover existing skills from GitHub repositories and store them in the correct local skills category. Results in discovered skills being downloaded into the users dotfile repo. |
| sop-authoring | agent only | — | Use when writing or creating new Standard Operating Procedures (SOPs) for AI agents. Covers effective SOP writing, clarity principles, and actionable instruction design. |
| sop-maintenance | user and agent | — | Use when updating, versioning, deprecating, or maintaining existing Standard Operating Procedures. Covers keeping SOPs accurate, relevant, and synchronized with implementation changes. |
| sop-rfc2119 | user and agent | — | Use when writing SOPs that require precise requirement levels using RFC 2119 constraint keywords (MUST, SHOULD, MAY). Covers proper usage of requirement keywords for deterministic agent behavior. |
| sop-structure | user and agent | — | Use when structuring Standard Operating Procedures with proper sections, organization, and markdown formatting. Covers SOP anatomy and section organization. |
| structured-research | user and agent | — | Delegates structured, multi-source research with verification and confidence levels; use for evidence-based investigations and decision support; produces traceable research files. This workflow is explicit-only because it writes research artifacts. |
| writing-skills | user and agent | yes | Creates and validates reusable agent skills with test-driven authoring and progressive disclosure, when creating, editing, restructuring, or reviewing skills, resulting in focused skills with verified agent behaviour |
| Agent | Tools | Description |
|---|---|---|
| context-builder | read, grep, find, ls, bash, web_search | Analyzes requirements and codebase, generates context and meta-prompt |
| delegate | all tools | Lightweight subagent that inherits the parent model with no default reads |
| generalist | all tools | General-purpose agent for documentation, cleanup, stale references, code organization, and miscellaneous maintenance tasks. |
| planner | all tools | Planning specialist that explores requirements and code context, weighs architecture trade-offs, and produces ordered implementation plans or plan files. May write planning artifacts; does not edit production code. |
| scout | all tools | Fast reconnaissance agent for exploring codebases, finding files by pattern, searching keywords, answering architecture questions, and returning compressed cited context or report artifacts. Specify thoroughness: quick, medium, or very thorough. |
| worker | all tools | General-purpose subagent with full capabilities, isolated context |
| Event | Command |
|---|---|
| SessionStart | bash -c 'set -o pipefail; bun run "${CLAUDE_PLUGIN_ROOT}/skills/shared-context/scripts/shared-context/cli.ts" inject | jq -c "{continue: true, hookSpecificOutput: {hookEventName: \"SessionStart\", additionalContext: (. // \"\")}}"' |
| SessionStart | "${CLAUDE_PLUGIN_ROOT}/skills/muxer-subagents/scripts/router.ts" session-context | jq -c '{continue: true, hookSpecificOutput: {hookEventName: "SessionStart", additionalContext: (. // "")}}' |