Generate fix prompts from rejected review groups

Show review approval summary and merge readiness

Create, revise, or audit technical specifications following RFC/IETF conventions. Delegates to the spec-writer agent.

Deep codebase quality audit with automated fixes. Runs the code-steward agent for line-by-line review, produces a findings document, then orchestrates parallel subagents to implement all fixes.

Iterative code quality audit that loops until clean. Runs steward in a bounded fix-audit cycle with human approval on first pass only.

Session-scoped code quality review of current branch changes. Lighter version of /prove:steward for active work.

Clean up all task artifacts (plans, reports, branches) with optional archiving

List, install, or remove prove tools

Validate configs, detect schema drift, and apply safe migrations

Run the orchestrator in autopilot mode to autonomously implement a planned task

Explore options and record decisions via structured brainstorming

Structured debugging protocol — sequential hypothesis testing with backtracking

Group changes into logical units and create semantic conventional commits

Complete a task by merging its branch to main and running cleanup

Socratic quiz on recent diffs to build deep comprehension of agent-generated code

Create a new Claude Code subagent

Create a new Claude Code slash command

Create a new Claude Code skill with best practices for descriptions, triggers, and resource bundling.

Generate human-readable documentation for projects, APIs, modules, or scripts

Generate LLM-optimized documentation for agents, APIs, or modules

Generate human-readable and LLM-optimized documentation for the current session or a specified scope

Generate or update the project's CLAUDE.md with LLM-optimized behavioral directives

Diagnose prove installation health — checks core config, tooling, and project drift

Run the orchestrator end-to-end: requirements -> plan -> parallel worktree execution -> merge

Create a handoff prompt (.prove/handoff.md) for clean session-to-session context transfer

Build or update the content-addressable file index

Detect project tech stack and generate .claude/.prove.json configuration

Install recommended community skills from external repos into ~/.claude/skills/

Configure orchestrator notification reporters (Slack, Discord, MCP, custom)

Send a test notification through configured reporters in .claude/.prove.json

Resume work from a handoff prompt (.prove/handoff.md) created by /prove:handoff

**Step Reference**: $ARGUMENTS

Plan implementation for a task

Configure Claude Code permissions (.claude/settings.local.json) for the active task's toolchain

Show orchestrator execution status — current wave, task statuses, review verdicts, and blockers

Manage the prompt engineering research cache -- list, add, prune, or refresh entries.

Craft an optimized LLM prompt from a goal or description. Pass --research for live web research.

Estimate token counts for prompt files using glob patterns or file paths.

Report a bug or feature request against the prove plugin via GitHub CLI

Assemble intent manifests and launch the review UI for structured code review

Surface review groups needing discussion

Post-agent code quality auditor. Deep codebase audit with surgical refactoring after parallel agent workflows.

Prompt optimization specialist. Audits and rewrites prompts, agent definitions, system instructions, and CLAUDE.md files for LLM efficiency. Use when writing, reviewing, or improving any prompt or agent definition. Triggers on prompt optimization, token efficiency, instruction tuning, or LLM best practices.

Reference for steward skills orchestrating the Progressive Context Distillation pipeline.

Semantic annotator for PCD structural maps. Adds module-purpose descriptions and architectural boundary labels to file clusters. Optional enrichment round.

Deep code reviewer for PCD audit pipeline. Performs targeted review of high-risk files with triage context and cross-file questions. Produces detailed findings with fix sketches.

Final synthesis agent for PCD audit pipeline. Produces findings.md and fix-plan.md from compressed artifacts. Never reads source code directly — operates only on structured review data.

Risk classifier for PCD audit pipeline. Reads source files in a batch and produces triage cards with risk scores, findings, and cross-file questions. Launched in parallel per cluster.

Principal Architect for code review during orchestrated execution. Reviews implementation against requirements, checks architectural coherence, and approves or requests changes. Used by the orchestrator in full mode for mandatory review gates.

RFC/IETF-style specification author. Drafts, revises, and audits technical specs, protocol definitions, and format standards.

Senior technical writer that produces clear, human-readable documentation for projects, APIs, and modules. Delegates from the agentic-doc-writer skill or invoked directly when documentation needs to be written for human consumption. Use when creating READMEs, getting-started guides, architecture overviews, API references, or contributor docs.

Lightweight LLM validator that evaluates code changes against user-supplied prompt criteria. Used by the orchestrator validation gate for non-scriptable checks. Runs prompt-based validators defined in .claude/.prove.json.

Generate machine-parseable, LLM-optimized documentation for Claude Code agents. Use when documenting agents, subagents, APIs, modules, or any code that other agents will consume. Triggers include "document this agent", "write agent docs", "create API docs for agents", or when documentation needs to be actionable by LLMs.

Analyze scope and delegate to docs-writer and/or agentic-doc-writer. Triggers on "auto docs", "document everything", "generate docs".

Iterative code quality audit that loops until clean. Runs the code-steward agent in a bounded fix-audit cycle — first pass gets human approval, subsequent passes auto-fix until the audit returns clean or the iteration cap is hit.

Interactive brainstorming sessions for software architecture, product scoping, and general engineering. Use when the user wants to explore ideas, gather requirements, narrow down solutions, weigh trade-offs, or make technical decisions. Triggers on "brainstorm", "let's think through", "help me decide", "what approach should", "pros and cons", or any open-ended design/architecture discussion. Saves decisions to .prove/decisions/ directory.

Generate and maintain an LLM-optimized CLAUDE.md for the target project. Scans the codebase (tech stack, conventions, structure), reads .claude/.prove.json config, and composes a concise CLAUDE.md with behavioral directives that Claude Code follows during the session. Full ownership of the file — safe to re-run, always produces deterministic output.

Clean up all task artifacts (plans, reports, branches, handoff context) with optional archiving to .prove/archive/. Use after a task lifecycle is complete to archive key documents and remove working artifacts.

Semantic commit assistant. Reads scopes from .claude/.prove.json, detects scope gaps and offers to register new ones, groups changes into logical units, and creates conventional commits.

Post-diff Socratic quiz that builds deep comprehension of agent-generated code. Analyzes recent changes, generates causal/design questions, quizzes the developer interactively, and logs comprehension gaps.

Generate human-readable documentation (READMEs, guides, API references, contributor docs). Triggers: "document this", "write docs", "create a README", "write a guide for".

Create a focused handoff prompt file for clean conversation-level handoffs. Deterministically assembles context from git state and prove artifacts, generates a small LLM pickup note, recommends the right agent, and outputs the exact command to start a fresh session. Use when hitting context limits, transitioning between phases, or wanting a clean restart without losing context.

Build, update, or query the content-addressable file index (CAFI) that helps agents navigate the codebase via routing-hint descriptions.

Configure orchestrator notification reporters (Slack, Discord, MCP, custom). Generates bash scripts and updates .claude/.prove.json. Triggers on "notify setup", "set up notifications", "configure alerts".

Autonomous task orchestrator that auto-scales between simple mode (<=3 steps, sequential, no worktrees) and full mode (4+ steps, parallel worktrees with mandatory principal-architect review). Each run operates in its own git worktree with namespaced state (.prove/runs/<slug>/), enabling concurrent runs that consolidate at merge time. Creates feature branches, runs validation gates, commits after each step, and supports rollback via git. Use when a .prove/runs/<slug>/TASK_PLAN.md or .prove/plans/ directory exists and the user wants hands-off execution. Triggers on "orchestrate", "autopilot", "full auto", "run autonomously", "implement without me", "hands-off mode".

Interactive planning and requirement gathering for specific tasks from .prove/TASK_PLAN.md. Use when the user wants to work on a numbered step from their plan (e.g., "Let's work on step 1.2.3") to create detailed requirements, make design decisions, identify edge cases, and define test strategies BEFORE implementation.

Analyze the active task plan and .claude/.prove.json to configure .claude/settings.local.json with scoped permission rules. Use before orchestrator, autopilot, or implementation. Triggers on "prep permissions", "setup permissions", "configure permissions", "allow tools", "stop asking me".

Manage the prompt engineering research cache. List, add, prune, or refresh cached research artifacts across three tiers — plugin-bundled, global, and project. Use when the user wants to manage their prompt research cache.

Generate optimized LLM prompts using the bundled prompt engineering guide and optional research. Delegates to the llm-prompt-engineer agent. Use when the user wants to create a new prompt, system instruction, or agent definition from scratch. Triggers on "craft a prompt", "write a prompt", "generate a system prompt", "create an agent prompt", "prompt for", or any request to produce a new LLM prompt.

Estimate token counts for prompt files. Wraps the token-count script for measuring agents, skills, commands, references, or any text file. Use when the user wants to check prompt size, compare token budgets, or measure files before/after optimization.

Assemble per-commit intent manifests into a review document. Launch the browser-based review UI for structured accept/reject per intent group. Falls back to LLM reconstruction when manifests are missing.

Create Claude Code skills with best practices for description tuning, resource bundling, and interaction patterns. Use when the user wants to create a new skill, asks about skill structure, or needs help with skill descriptions and trigger phrases.

Guide for creating Claude Code slash commands with best practices. Use when the user wants to create a new slash command, asks about slash command structure, or needs help with command frontmatter, arguments, or tool restrictions.

Create, revise, and audit technical specifications following RFC/IETF conventions. Use when the user wants to write a new spec, edit an existing spec, review a spec for completeness, or formalize a design decision into a specification document. Triggers on "write a spec", "spec for", "formalize this", "draft a specification", "audit this spec", "revise the spec", "protocol spec", "format spec", or any request to create structured technical documentation with normative requirements. Also triggers when the user has a brainstorm decision record they want to turn into a formal spec.

Session-scoped code quality review. Audits only source files changed in the current branch/task, skipping tests. Lighter version of /prove:steward for use during active work.

Deep codebase quality audit with automated fixes. Runs the code-steward agent for line-by-line source code review, produces a findings document, then orchestrates parallel subagents to implement all fixes. Tests are reviewed separately after source changes land.

Create Claude Code subagents. Triggers on "create an agent", "new subagent", agent design questions, or tool permission/role definition requests.

Discovery-driven planning for tasks in existing codebases. Explores code, gathers requirements, identifies edge cases, and produces .prove/TASK_PLAN.md for the orchestrator.