AUTO.Prompt guides an AI agent to plan before acting, challenge its plan once, execute the shortest suitable path, and save only reusable knowledge.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install auto-prompt
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install auto-prompt using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is AUTO.Prompt?
AUTO.Prompt is a strategy-first problem-solving protocol for AI agents. It turns a topic, problem, or requested outcome into a concise strategy card, checks existing resources in a defined order, reviews the plan against the original request, and then guides execution and a conclusion-first response. For Openclaw Skills users, it adds a consistent method for deciding how to think, act, and wrap up without expanding the task unnecessarily.
The protocol prioritizes reuse and local resources, limits research and self-review, and avoids unnecessary questions except when a task involves irreversible actions, payment, or contradictory requirements. After a problem is resolved, it applies four tests before saving any durable insight: reuse potential, factual or actionable value, uniqueness, and clarity. It is a method layer—not a content library, execution engine, or replacement for specialist skills.
AUTO.Prompt Use Cases
- A user asks how to accomplish a multi-step goal, solve a problem, or create a plan.
- A task may be solvable by reusing a local script, template, memory, skill, or expert instead of starting from scratch.
- The agent needs to decide whether current information requires official documentation, community research, or search.
- A proposed solution needs a focused check for relevance, shorter alternatives, unsupported assumptions, or unnecessary scope.
- A completed task yields a reusable fact, project convention, preference, or repeatable workflow worth retaining.
- A specialist skill already matches the domain, so the shortest path is to delegate directly rather than apply this protocol.
How AUTO.Prompt Works
- Select a mode: Use problem-solving mode by default, strategy-request mode when the user explicitly asks for a plan, direct delegation when a specialist skill applies, and no activation for small talk, simple factual queries, or single-step operations.
- Normalize the request: Compress it into a goal sentence containing an action, object, and completion criterion. Set a reasonable criterion without asking follow-up questions unless the task is irreversible, paid, or internally contradictory.
- Build a strategy card: Check reuse sources in order: local assets and skills, model knowledge for stable facts, official documentation when authoritative or version-specific details matter, tier-1 communities when needed, and search as a fallback. Stop at the first sufficient source.
- Review once: Compare the card with the original request: confirm it addresses the actual question, uses the shortest path, does not skip required verification, and stays within scope. Revise at most once; if a conflict remains, report it rather than forcing a solution.
- Execute and deliver: Follow the approved card, update it if execution deviates, and respond conclusion first with only necessary evidence and any user action required.
- Distill after resolution: Apply the four memory tests. Save only reusable, factual or actionable, non-duplicate, clearly expressed information; do not save unresolved-task notes or process logs.
AUTO.Prompt Setup
AUTO.Prompt is described as a local, zero-dependency method skill. The source does not specify a package manager, installation command, or required configuration, so use the installation mechanism provided by your Openclaw-compatible agent or skill manager.
- Add the skill's
SKILL.mdto the agent's skill directory using your platform's documented process. - Ensure the agent can access project memory paths if you want it to retain qualifying insights. The skill does not require the separate
autocontent library. - Optionally provide the reference file
references/sources.mdfor routing to official documentation and tier-1 communities by subject. - Invoke it with a strategy-oriented request such as “give me the shortest path” or “help me solve this.” Avoid invoking it for simple lookups or single-step actions.
# No installation command is specified in the supplied SKILL.md.
# Follow your agent platform's documented skill-installation process.
AUTO.Prompt Data Schema & Taxonomy
AUTO.Prompt is primarily a workflow protocol; it does not define a generated project format or require a content database. Its strategy card is a compact working artifact with these fields:
| Field | Purpose |
|---|---|
| Goal | One-sentence action, object, and completion criterion |
| Reuse | Highest successful source level and any remaining gap |
| Path | Ordered actions, matching the actual number of steps |
| Tools | Required tools, skills, or commands |
| Deliverable | Output form and length limit |
| Risk | Irreversible actions and points requiring user confirmation |
The reuse-source taxonomy is L0 local assets, L1 stable model knowledge, L2 official documentation, L3 relevant tier-1 communities, and L4 search-engine fallback. Stop when a sufficient source is found; search is limited to 1–3 attempts.
Qualifying memory is routed by scope:
- Project-specific dated notes:
.workbuddy/memory/YYYY-MM-DD.md(append rather than overwrite). - Project-level conventions, boundaries, or approved wording:
.workbuddy/memory/MEMORY.md. - Cross-project preferences or habits:
~/.workbuddy/MEMORY.md. - Reusable multi-step procedures: promote to a skill rather than storing as memory.
Temporary paths, timestamps, raw tool errors, one-off search results, unverified guesses, emotional conclusions, and activity logs are excluded. The optional references/sources.md contains subject-based routes to official documentation and tier-1 communities; AUTO.Prompt itself neither searches nor writes to the separate auto content library.
AUTO.Prompt Advanced Features
- Tiered reuse with early stopping: Prefer available local assets and stop researching as soon as a sufficient source is found.
- Authority-aware verification: Require official documentation for APIs, mechanisms, standards, policies, and version-dependent details rather than relying on memory.
- Bounded adversarial review: Check relevance, efficiency, verification, and scope in one review pass, with at most one strategy revision.
- Explicit mode and delegation boundaries: Skip the protocol for trivial tasks and hand matching work to specialist skills or experts.
- Controlled research budget: Use search only as a fallback and cap retrieval at 1–3 attempts.
- Memory quality gate: Apply four criteria before recording insights and route approved knowledge by project or cross-project scope.
- Repeatable-workflow promotion: Treat durable multi-step procedures as candidates for a dedicated skill instead of accumulating them in memory.
- Concise delivery discipline: Lead with the conclusion, include only necessary support, and avoid unrequested scope or process narration.
SKILL.md
Loading
Related Openclaw Skills
Learning5D Agent Safety Check and Enrolment
Learning5D helps agents assess safety risks, review skills before installation, and optionally study consent-first relational safety.
Marketing Engineer
Marketing Engineer turns marketing operations into an evidence-based, auditable workflow with structured workspaces, content tickets, GEO/SEO playbooks, and a live dashboard.
forktree
forktree helps coding agents and automation create, manage, and clean up Git worktrees with safety checks and predictable shell output.
Medical Device Registration Dossier Assistant
A regulatory writing assistant that organizes medical device evidence into market-specific, submission-ready registration dossiers.
Regulation HK Sandbox
A regulatory pathway designer that maps cross-border RWA projects to jurisdiction-specific rules, Hong Kong sandbox options, SPV structures, and required compliance gates.
TS Term Review — Investment Term Sheet Risk Review
Review an investment term sheet clause by clause, prioritize red lines, and get practical negotiation language and alternatives.
Featured*
