Create, implement, review, and update self-contained execution plan documents for coding agents.
npx skills add https://github.com/shinzui/keiro --skill exec-plan-shinzui
Please help me install this Agent Skill. Skill: exec-plan Source: https://github.com/shinzui/keiro/tree/main/agents/skills/exec-plan Command: npx skills add https://github.com/shinzui/keiro --skill exec-plan-shinzui
docs/plans/ that carry everything a novice or coding agent needs to implement a change end-to-end, including progress tracking, decision logs, and provenance of which model wrote or reviewed the plan. ## Core Features & Use Cases - Plan lifecycle modes: create, implement, discuss, review, update, and status modes cover the full lifecycle of an execution plan, from initial research through completion and ADR distillation. - Scripted scaffolding and provenance: bundled Bun/TypeScript scripts (init-plan.ts, record-provenance.ts) generate numbered plan files with YAML frontmatter and append tamper-safe authorship and review entries, with runtime model discovery for Codex and Claude Code sessions. - ADR integration: a shared workflow discovers, cites, creates, and validates Architecture Decision Records, distilling durable project decisions out of completed plans. - Use Case: Ask the agent to plan a new feature; it researches the repository, generates docs/plans/4-add-queue-support.md with milestones and observable acceptance criteria, then implements it milestone by milestone while keeping progress, decisions, and git trailers in sync. ## Quick Start Ask the agent to create an execution plan for the feature you want to build, for example: create an ExecPlan for adding a health-check endpoint to the server.High-intent search queries and answers about installing and using this skill.
Run the skill in create mode with a title describing the change. The bundled init-plan.ts script researches nothing itself but generates a numbered Markdown file in docs/plans/ with YAML frontmatter and a skeleton of sections, which the agent then fleshes out from repository research.
An ExecPlan is a self-contained living document that a novice or coding agent can implement from without any prior context. Unlike a static design doc, it maintains Progress, Surprises & Discoveries, Decision Log, and Outcomes & Retrospective sections that are updated at every stopping point during implementation.
The record-provenance.ts script appends entries to a provenance block in the plan's YAML frontmatter, recording the exact runtime model, harness, timestamp, and verdict or mode. It discovers the model from Codex or Claude Code session files and never rewrites existing entries.
Yes. Each model appends its own review entry with a verdict of approved, changes-requested, or comments using the record-provenance script. Entries are append-only, so several models can review the same plan without overwriting each other's records.
ExecPlans are active execution memory, not durable project memory. Durable architectural decisions must be distilled into Architecture Decision Records under docs/adr/ before a plan is marked complete, following the ADR.md workflow.
mandor1428Create and maintain Architecture Decision Records with implementation plans and verification criteria.
lanegridCreate and maintain executable specifications in PLANS.md for feature delivery.
MaelwalserCapture architectural decisions as formatted ADR documents during coding sessions.
burgebjExecute small software changes with atomic commits and state tracking.