Orion-app

Publishes an Orion release end-to-end — version bump, build, push release prep, wait for GitHub checks/install smoke (skippable when CI already passed on the pre-bump commit), tag, GitHub release, npm/PyPI publish via scripts/publish-release.sh (NPM_TOKEN + PyPI token in ~/.zprofile), sync version tags in Orion-api and Orion-docs, then post-release verification. Use when the user asks to publish a release, ship a version, bump and publish orion-notebook or orion-ui, or cut a new npm/PyPI release.

Builds Orion-native notebook UI outputs with the Python `orion_ui` package. Auto-loaded for notebook work; use when the user asks for interactive notebook UI features like plotly charts, sliders, selects, buttons, etc.

Operates Orion user and workspace settings safely. Use when reading, creating, validating, or changing `~/.orion/settings.json` or `<workspace>/.orion/settings.json`, settings overrides, model pins, provider config, workspace-level settings, or recovering user settings from automatic backups in `~/.orion/settings-backup`.

Creates or edits Orion notebook App View selections from notebook cells and outputs. Use when the user asks to make a notebook app, dashboard, report UI, or app view from notebook content.

Reference for Orion notebook metadata contract. Use when deciding which fields under notebook `metadata.orion` or cell `metadata.orion` are supported, what values they accept, and which keys are legacy/internal-only.

Runs a plain-language business analysis of uploaded data, producing key findings, data quality notes, charts, and suggested next questions for non-technical users.

Reviews inventory age, sell-through, stale stock, and stockout risk for retail or ecommerce operations teams.

Analyzes revenue, cost, gross profit, and margin by product, category, channel, store, customer, or period for retail and operations teams.

Builds a concise monthly business report from uploaded operational or finance data, with plain-language narrative, charts, and App View-ready outputs.

Runs tool-use test prompts while preserving question/answer separation. Use when executing prompt suites, agent tool tests, bash/background tests, sub-agent tests, notebook tool tests, or edge-case test prompts without seeing expected behavior.

Uses the EDA research profile for exhaustive exploratory data analysis inside Orion Research mode. Use only when the user explicitly asks for deep, exhaustive, or thorough EDA.

Writes or updates end-user help in Orion-docs with clear fix-first steps, plain language, and issue links when relevant; includes POSIX and Windows (NT) paths or commands when both apply. Use when the user asks for user-facing documentation, public help pages, troubleshooting guides for Orion users, or markdown under Orion-docs.

Authors AGENTS.md or CLAUDE.md workspace rule files that Orion injects into every chat. Use when the user wants to create, update, or scaffold project conventions, coding standards, AGENTS.md, CLAUDE.md, or persistent agent instructions — not when they only want the domain work done without authoring a rule file.

Teaches Orion agents how to inspect and reference past chats stored in the local SQLite database at `~/.orion/orion.db`. Use when the user asks about previous chats, prior conversations, chat history, remembered decisions, earlier instructions, or finding messages from past Orion sessions.

Authors new workspace SKILL.md files under .agents/skills by default; use .orion/skills for Orion-only overrides of the same skill name. Use when the user wants to add, create, or scaffold a skill, SKILL.md format, or skill best practices — not when they only want the domain work done without a new skill file.

Authors notebook-defined Orion sub-agents under `.agents/subagents/*.agent.ipynb` by default; use `.orion/subagents/` for Orion-only overrides of the same id. Use when the user wants to create, scaffold, name, structure, or update a reusable sub-agent definition — not when they only want that sub-agent's domain work done in this chat without authoring a notebook.

Adds new LLM models to the Orion model catalog or updates existing ones. Searches for model availability via provider APIs, looks up token pricing from official docs, and edits lib/agent/model-catalog.ts. Use when adding models, updating model pricing, labels, context limits, long-context pricing, client_avail, or pinned_by_default.

Creates prompt-based tool-use test suites under public/test-prompts with separated questions and answer keys. Use when adding agent tests, prompt suites, edge-case tests, or evaluator/executor-compatible test prompts.

Reviews code changes from a refactor or UI work and produces a prioritized manual UI test checklist. Use after large refactors, component extractions, editor or layout changes, or when the user asks for manual tests, UI QA, or a test plan before shipping.

Creates GitHub issues for the Orion repo following project conventions. Use when the user asks to create an issue, file a bug, add a feature request, or track work in GitHub issues.

Opens a GitHub pull request using the gh CLI with a clear title and body, optional issue links, and draft or ready state. Use when the user asks to open a PR, create a pull request, submit changes for review, or publish a branch to GitHub.

Fetches GitHub issue descriptions using the gh CLI and implements the fix, feature, optimization, or refactor described. Use when the user asks to fix a GitHub issue, implement an issue, work on an issue by number, or when given an issue URL or number.

Evaluates tool-use test executions against separated answer keys. Use when grading an agent run, comparing execution transcripts to expected behavior, scoring prompt-suite results, or auditing whether the executor avoided answer files.