openplan-skill

Orchestrate OpenPlan-style skills in auto or YOLO mode while preserving human decision boundaries. Use when Codex, Claude Code, or another agent should automatically inspect, align enough to proceed, research, test, review, garden, summarize, hand off, or generate temporary HTML without silently approving, changing durable goal/intent, bypassing risk decisions, or pretending soft guidance is hard enforcement. This skill coordinates openplan-core, openplan-operate, openplan-align, openplan-research, openplan-test, openplan-review, openplan-garden, openplan-record, openplan-handoff, openplan-html-brief, and openplan-docs-init.

Apply OpenPlan philosophy as a platform-neutral agent operating protocol. Use when Codex, Claude Code, or another coding agent must preserve OpenPlan-style principles without depending on project-specific enforcement tools: minimal effective context, human-owned decisions, dispatcher/subagent boundaries, evidence-backed state, separation of roles, authority flow from goal/intent and design to execution, reversible automation, and no silent approval.

Initialize and run an OpenPlan-style agent session on Codex, Claude Code, or another agent platform. Use after openplan-core when a task needs startup discipline, behavior constraints, subagent delegation rules, mode selection, dual-surface awareness, or safe auto/YOLO execution before alignment, research, design, implementation, testing, or review. This skill is platform-neutral and treats enforcement checks as optional evidence, not prerequisites.

Design, run, and report a proportional OpenPlan-style test evidence framework. Use when Codex, Claude Code, or another agent must choose test layers, prove regression tests are meaningful, separate fast checks from privileged or expensive checks, compare implementation behavior with fixtures/oracles/properties, or report tests run/skipped without treating passing tests as approval.

Review automated or YOLO-mode agent output using OpenPlan philosophy without requiring project-specific enforcement tools. Use for Codex or Claude Code automated review flows when changes, plans, generated docs, specs, code patches, skill/plugin artifacts, or multi-agent workflow results must be checked for minimal context, human decision boundaries, layered authority, role/mode separation, research-before-design, evidence-backed findings, reversible automation, and no silent approval. This is a read-only review skill, not an implementation or auto-fix skill.

Initialize OpenPlan-style documentation generation before durable docs, memory, specs, or project knowledge are created. Use when Codex or Claude Code must bootstrap a docs repo, add OpenPlan-style docs to an existing repo, create initial doc maps/templates/needs snapshots/decision records, or prepare a project for document generation while preserving human goal/intent, dual-surface authority, ADR discipline, and garden/review follow-up.

Inspect and improve documentation quality using OpenPlan-style document gardening. Use when Codex or Claude Code must check memory, docs, design notes, handoff snapshots, review reports, specs, or generated documentation for goal/intent drift, broken structure, missing rationale, stale claims, duplicated content, unclear authority, poor future-agent readability, zero-context comprehension gaps, or periodic documentation maintenance. This skill is platform-neutral and read-only by default; local link checkers and linters are optional evidence sources, not required gates.

Organize, update, and reconcile memory and documentation as human goal/intent records in an OpenPlan-style workflow. Use when Codex or Claude Code must classify or propose what belongs in memory, docs, handoff snapshots, review reports, or temporary briefs; when consolidating user preferences, project philosophy, design decisions, workflow constraints, or lessons learned; or when preventing agent inference from being recorded as the human goal/intent. This skill is for goal/intent preservation and documentation maintenance, not code implementation.

Clarify and structure ambiguous OpenPlan-style requests before research, design, implementation, or review. Use with OpenPlan philosophy when a user request is underspecified, strategically risky, solution-first, missing success criteria, or needs blind-spot analysis, advisor-style questions, route selection, or a needs-snapshot-style summary. This skill does not implement or review changes directly.

Create a minimal OpenPlan-style handoff snapshot for cross-agent or cross-session continuity. Use when work must move between Codex, Claude Code, subagents, reviewers, auto/YOLO runs, or compacted sessions while preserving minimal context, human decisions, dual-surface state, evidence, blockers, next safe action, and residual risk. This skill creates handoff artifacts; it does not approve or continue the work itself.

Create a temporary human-facing HTML frontend for OpenPlan-style agent work. Use when an agent needs to present what it did, what it found, what decisions remain, a handoff snapshot, an auto-review report, or a compact teaching/explanation artifact in a clearer visual format. Generate disposable, self-contained HTML for human reading; do not treat it as persistent project documentation or a committed artifact unless the user explicitly asks.

Run bounded OpenPlan-style research before design, implementation, review, or documentation generation. Use when Codex or Claude Code must reduce uncertainty with evidence, compare options, inspect source material, or produce a research report without turning findings into human decisions, design approval, or durable intent.