How do I install the Auto Pilot skill?

Run npx skillmds@latest add latifox/auto-pilot in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Auto Pilot safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Auto Pilot?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Auto Pilot free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Auto Pilot?

Latifox (@latifox) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Auto Pilot by latifox — Agent Skill

Name: Auto Pilot
Author: Latifox

File contents

Run the whole pipeline without supervision. Announce the plan first, then execute every stage in order without returning to the user between them.

Focus: $ARGUMENTS

The rule that makes this autonomous: do not stop to ask questions. Where you would normally ask, choose the documented default, record the assumption, and keep going. Gaps go in inputs_missing, profile_gaps and the memo's watermark; they do not become questions. The only reasons to stop early are listed at the bottom.

Sequence:

Profile. If memory/user_profile.md is missing, run the fast path of first-run onboarding. This is the one place you may ask, because nothing downstream is meaningful without it, and it is five questions. If a profile exists, use it as-is; do not offer to update it.
Research and generate. Run workflows/idea-generation.md from the trend step. Reuse any market_insights file for the niche that is still fresh rather than re-running research.
Rank. Quick-score every candidate. Competition is capped at 45 and carries competition_prior_capped: true. Present the ranked table with rank labels.
Validate the top candidate. Run workflows/idea-validation.md in full on the highest-ranked idea. Set business_model from the profile's preferred_business_model; if that is unsure, infer it from the idea and say so.
Memo. Write the decision memo, set status: scored and validated_at.
Verify. Run python tests/validate_memory.py --idea <slug> and fix every ERROR before reporting.
Report. Give the final summary: the ranked candidates, the validated idea's score and verdict, the riskiest assumption with its experiment, and one line on what the run could not determine.

Stop early only if: the trend research finds no usable signal for the niche on any platform, or scoring cannot meet minimum viable input (three dimensions including demand and distribution), or the harness reports errors you cannot fix. In each case say what stopped you and what you would need.

This run costs real research time and produces a full analysis. Say so before you start, then get on with it.

---name: auto-pilotdescription: Autonomous end-to-end run - onboards you if needed, finds ideas, validates the best one, and hands you a memo without stopping to ask---Run the whole pipeline without supervision. Announce the plan first, then executeevery stage in order without returning to the user between them.Focus: $ARGUMENTS**The rule that makes this autonomous:** do not stop to ask questions. Where youwould normally ask, choose the documented default, record the assumption, and keepgoing. Gaps go in `inputs_missing`, `profile_gaps` and the memo's watermark; they donot become questions. The only reasons to stop early are listed at the bottom.Sequence:1. **Profile.** If `memory/user_profile.md` is missing, run the fast path of first-run onboarding. This is the one place you may ask, because nothing downstream is meaningful without it, and it is five questions. If a profile exists, use it as-is; do not offer to update it.2. **Research and generate.** Run `workflows/idea-generation.md` from the trend step. Reuse any `market_insights` file for the niche that is still fresh rather than re-running research.3. **Rank.** Quick-score every candidate. Competition is capped at 45 and carries `competition_prior_capped: true`. Present the ranked table with rank labels.4. **Validate the top candidate.** Run `workflows/idea-validation.md` in full on the highest-ranked idea. Set `business_model` from the profile's `preferred_business_model`; if that is `unsure`, infer it from the idea and say so.5. **Memo.** Write the decision memo, set `status: scored` and `validated_at`.6. **Verify.** Run `python tests/validate_memory.py --idea <slug>` and fix every ERROR before reporting.7. **Report.** Give the final summary: the ranked candidates, the validated idea's score and verdict, the riskiest assumption with its experiment, and one line on what the run could not determine.**Stop early only if:** the trend research finds no usable signal for the niche onany platform, or scoring cannot meet minimum viable input (three dimensionsincluding demand and distribution), or the harness reports errors you cannot fix.In each case say what stopped you and what you would need.This run costs real research time and produces a full analysis. Say so before youstart, then get on with it.

Frequently asked questions

Auto Pilot