bootloops-setup

BootLoops ships twelve working protocols in skills/ at the repository root. They are a library, not an installation: nothing in skills/ is active until the user chooses to activate it. This skill is the installer. Never activate anything without an explicit choice.

Procedure

1. Show the roster. Present the two groups below (one line each). The protocol layer is discipline for any quantitative work; the research skills are heavier machinery for specific tasks.

   The protocol layer:

   • acceptance-gate — what "done" means: never-fit digits against an independent route
   • constant-recognition — integer-relation discipline: declared rings, height bounds, refusal over invention
   • planted-truth — synthetic-truth controls before real data
   • independence-bookkeeping — provenance of reference values; no oracle that fed a fit certifies the result
   • timing-discipline — measure a small run first; a multi-hour projection means restructure
   • reading-contract — sources-only assertion for document-reading instruments
   • tool-stewardship — consult the toolkit first, write code last, document for the next agent

   The research skills:

   • prove-protocol — proving with agents: adversary-first pipelines, skeptic loops
   • referee-sim — pre-circulation audit: the strongest standard objection from every audience
   • lit-review — the literature audit behind a novelty claim: papers actually read, citation chains both directions
   • ref-check — bibliography verification against authoritative sources
   • prose-lint — scientific-prose linter: hype vocabulary, empty structures, agentless prose, honesty and number checks

2. Ask what to activate. Offer: all twelve, the protocol layer only, the research skills only, an individual selection, or none. Use the host's question/multi-select interface if it has one (in Claude Code, grouped multi-select questions fit within the four-option limit); otherwise ask in plain text. "None" is a fine answer — the library stays available to read or load ad hoc either way.

3. Ask the scope. Project-level (.claude/skills/ in the user's project — active only there) or user-level (~/.claude/skills/ — active in every project). Default to project-level if the user has no preference.

4. Install. Copy each chosen skill's directory from skills/<name>/ in this repository into the chosen location, keeping the directory name. Nothing else: no edits to the files, no extra registration step.

5. Confirm. List what is now active and where. Note that an activated skill still only engages when its work actually comes up; activation makes it discoverable, not mandatory. To deactivate one later, delete its directory from the install location. Re-run this setup any time to add or change the selection.