use-levistack
Skill0 starsChanged 18 days agoMITView on GitHub
Start here. The entry point that routes any prompt through levistack in seven fixed steps: route, prime, invoke, hunt, apply, prove, deliver. Picks the first skill (codebase-learner, trace, second-opinion, challenge-me, ready-spec, or a named one), hunts the catalog with find-skills when nothing obvious fits, runs the chain, proves every claim, and shapes the reply. Use when the user does not know which skill fits, says 'where do I start', 'help me with', 'what should I use', 'just handle it', gives a goal with no skill named, or types /use-levistack with a task. A named skill inside it runs as a one-shot.
In this repository57 other files it gives its agents.
- AGENTS.mdAGENTS.md
- .cursor/rules/comments.mdcCursor rule
- .cursor/rules/commit-messages.mdcCursor rule
- .cursor/rules/git-workflow.mdcCursor rule
- .cursor/rules/manage-stack.mdcCursor rule
- .cursor/rules/no-worktrees.mdcCursor rule
- manage-stackSkill
- convert-as-skillSkill
- create-agentSkill
- create-agents-mdSkill
- create-commandSkill
- create-hookSkill
- create-ruleSkill
- create-skillSkill
- retrospectiveSkill
- super-architectureSkill
- super-cachingSkill
- super-elixirSkill
- super-nodejsSkill
- super-protobufSkill
- super-tech-blueprintSkill
- atomic-semantic-commitSkill
- code-reviewSkill
- create-bruno-collectionSkill
- create-documentationSkill
- create-ticketsSkill
- ready-specSkill
- write-adrSkill
- write-changelogSkill
- write-mr-descriptionSkill
- write-planSkill
- write-postmortemSkill
- write-runbookSkill
- noslopSkill
- output-saversSkill
- safe-refactorSkill
- tddSkill
- verifySkill
- get-movingSkill
- later-listSkill
- standupSkill
- challenge-meSkill
- codebase-learnerSkill
- researchSkill
- second-opinionSkill
- traceSkill
- color-kitSkill
- design-system-kitSkill
- diagramSkill
- infographicSkill
- product-polish-workflowSkill
- prototype-wireframeSkill
- ui-ux-kitSkill
- evidenceSkill
- find-skillsSkill
- handoffSkill
- rumblingSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 28
Do not stop at a recommendation.
- Distinctive: line 42
Never reorder. A claim without fresh proof does not enter the reply.
- Very specific: line 62
Read-only skills (codebase-learner, trace, second-opinion, challenge-me) never edit code.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Use levistack
Do not stop at a recommendation. Run the steps below in order, top to bottom, for every prompt.
Orchestration (run in order)
| Step | Do | Playbook | Skip when |
| 1 Route | Accept the prompt. If the user named a skill, mark it a one-shot | — | Never |
| 2 Prime | output-savers Step 0 on the user's prompt: safety override, job, reply level. Never rewrite their words | prime.md | Never |
| 3 Invoke | Pick the first skill from the intent map, or the named one-shot. No clear intent: go to step 4 | invoke.md | Never |
| 4 Hunt | find-skills: gather what is missing, search the catalog, compose a chain of at most four skills, go upstream only for a real gap | find-skills | Step 3 found the skill, or the user named it |
| 5 Apply | Run the chain. Load each skill when its step starts. Pass each result to the next in one sentence | the chosen SKILL.md files | Never |
| 6 Prove | verify before any "done", "fixed" or "passing". Every outward-facing action stops at the user | prove.md | Nothing was changed and nothing is claimed |
| 7 Deliver | polyglot-copywriter on prose a person will read, then output-savers last on your reply | deliver.md | Never |
Never reorder. A claim without fresh proof does not enter the reply. An outward action without a yes does not run.
Around the steps
| Layer | What | Role |
| Rules | rules/core/ (always on) and topic rules by glob | Apply to the whole session, not one step |
| Hooks | careful-guard, generated-file-guard, large-read-guard, session-start, notify-done (opt-in) | Enforce at tool-call time, even when a step forgets |
| Agents | context-harvester, explorer, planner, tdd-runner, implementer, reviewer, security-reviewer, bulk-reader | Optional spawn in steps 3 to 6 for one bounded job. Results go back to this session, which keeps the steps |
| Focus | get-moving, later-list | Any step. A tangent goes on the later list, not into the work |
One-shots
The user named a skill inside /use-levistack (for example "use-levistack: second-opinion on my branch"): run steps 1, 2, 3 with that skill, skip 4, then 5, 6 and 7. A bare /skill call without this router runs that skill alone; prime and deliver do not run.
Hard rules
- Steps run in order. Prime before invoke. Deliver after the work. Those are two
output-saverstouches on two different targets. - Closed catalog first. Go to skills.sh only through
find-skillsUpstream, and never install without a yes. - At most one round of questions in the whole run, at most three questions, each with a recommended answer. Read the workspace before asking.
- Read-only skills (
codebase-learner,trace,second-opinion,challenge-me) never edit code. - Push, open or update an MR or PR, post a comment, publish, install, or anything destructive: show exactly what will happen and wait for a yes.
When the route gets stuck
| Symptom | Do |
| Two skills tie | Read both descriptions' "Not for" lines; the one that does not exclude the job wins |
| The opened skill does not fit | Say so, go back to step 3 or 4. Do not force it |
| The user is overwhelmed, not confused about skills | get-moving |
| Nothing fits | Say so plainly. Offer to do it directly, or create-skill if the job will repeat |
Example chains: references/worked-chains.md.
Related skills
find-skills: step 4 hunt and skills.sh installs.output-savers: step 2 prime and step 7 last gate.verify: step 6 proof.polyglot-copywriter: step 7 prose for people.codebase-learner,trace,second-opinion,challenge-me,ready-spec: step 3 first picks.get-moving: the user is stuck on the work, not on the skill.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.