miniplan

Plugin miniplan from marketplace miniplan-skills in biuayi/miniplan-skills. Commands read from the repo's manifest — 19 days ago.

Skills (5)

• agents

  Base layer of miniplan, auto-loaded at session start. Defines the agent roles, model routing, the per-request subagent budget and how an approved run may raise it, read-only fan-out, and the Task Contract / Compact Result formats used by miniplan:brainstorming, miniplan:plans, and miniplan:working.

• brainstorming

  Stage one of miniplan, entered from miniplan:miniplan for a new topic, or when the user explicitly asks to brainstorm or design something before building it. Splits the problem into decisions, gets each decision explicitly from the user, writes a design draft, has a critic falsify it, and stops for

• miniplan

  Main entry point of the miniplan pipeline. Invoke as /miniplan <request>, or /miniplan alone to resume unfinished work, for any broad, vague, multi-module, or investigation-type request (整体/全面/整理/梳理/盘点/调研/优化/重构/接入/把X做起来/有哪些能力). Detects the current stage from docs/miniplan/ and runs or resumes brains

• plans

  Use after a design has been approved through miniplan:brainstorming, or when the user hands over an approved design file, to turn it into a task manual — tasks with dependencies, deliverables, executable acceptance checks, per-agent prompts, agent assignments, and a spawn budget. Stops for approval

• working

  Use when an approved task manual from miniplan:plans exists and the user says to start, continue, or resume it (开工/继续/接着跑). Dispatches each task's agents with the manual's prompts, gates every task on independent evidence, keeps a progress ledger so work survives context compaction, repairs once, an

Subagents (8)

• builder

  Implements one approved, clearly bounded change after scope and acceptance criteria are known. Use for code-writing work, not open-ended exploration or final approval.

• critic

  Falsifies a design draft, task plan, or high-impact claim before work starts. Every objection must cite evidence (path:line or quoted text) and end with a fix or a decision for the owner; also lists what it tried to refute and could not. Read-only. Use after a design draft exists, not on code diffs

• designer

  Designs complex architecture or implementation plans on the main conversation's model tier. Use when an isolated design pass will reduce main-context load; do not use for trivial tasks.

• judge

  Resolves material conflicts between review or verification results and makes high-risk acceptance decisions. Use only when the user asks.

• reviewer

  Reviews a stable change for correctness, security, behavioral regressions, and missing tests. Use after implementation; does not edit code.

• runner

  Executes a bounded command or data-collection plan and returns concise raw evidence. Use for deterministic runs and large outputs, not design or interpretation.

• scout

  Performs read-only codebase or documentation exploration and returns a compact evidence map. Use to locate relevant facts and execution paths before design or implementation.

• verifier

  Independently verifies acceptance criteria, tests, and high-impact claims after a change or investigation. May run checks but does not edit source files.

Hooks (6)

• PreToolUse

  bash "${CLAUDE_PLUGIN_ROOT}/hooks/agent-guard.sh" check

• PreToolUse

  bash "${CLAUDE_PLUGIN_ROOT}/hooks/run-guard.sh" workflow

• PreToolUse

  bash "${CLAUDE_PLUGIN_ROOT}/hooks/test-gate.sh"

• UserPromptSubmit

  bash "${CLAUDE_PLUGIN_ROOT}/hooks/agent-guard.sh" reset

• Stop

  bash "${CLAUDE_PLUGIN_ROOT}/hooks/run-guard.sh" stop

• SessionStart

  jq -Rs --arg root "${CLAUDE_PLUGIN_ROOT}" '{hookSpecificOutput:{hookEventName:"SessionStart",additionalContext:("# miniplan:agents (auto-loaded at session start)\n\n" + (gsub("<!-- host.codex[^\n]*\n(

Similar plugins