A Claude skill that turns any authorized contributor — or council-recognized agent — into a sim author. Clone-portable: a peer site that copies the substrate inherits the skill automatically and runs its own silo coms ceremony against the same logic. Five files of governance hold the floor; templates and a worked example carry the rest.
A Claude skill that teaches an LLM (Claude or another assistant trusted by the visitor) how to help an authorized builder draft, refine, and submit a new sim card to a Merkle Trust site. The skill is intentionally clone-friendly: a peer site that copies the Merkle Trust substrate inherits this skill automatically and runs its own authoring + voting cycle against the same logic.
A draft starts as a conversation with the LLM. The LLM uses the skill to surface the four questions, help the builder converge on answers, render templates, apply truth-claim discipline, and prepare a submission packet. The packet routes through the silo voting flow; cards land on approval.
The skill is for builders who have walked at least one canonical sim and who want to contribute one of their own. Authorized contributor categories:
Anyone outside these categories can still draft a sim — the skill helps them — but the submission goes to the public-intake silo and is reviewed against a higher bar before voting begins.
Before the LLM helps draft anything, the skill surfaces these four. The builder needs answers (or working hypotheses) for each. The skill helps converge through dialogue; it does not invent answers.
Every cards_v2 card follows the same 6-step structure (or, for unusual cards like 40d/40x/40y/40z/41/42, an explicit and well-argued variation):
| Step | id | What goes here |
|---|---|---|
| 1 | landing | Persona's situation; the hook line; "nothing to type, nothing to commit to" |
| 2 | install | The four real install paths in cluster-appropriate order |
| 3 | first_concrete | Synthetic data loads; dramatic moment; robot animation; ticker streams hex; ASCII slab if it earns the moment |
| 4 | is_this_real | Three plain-English bullets; the receipt link; the .md tag-along + comment+Claude bridge named explicitly |
| 5 | ceremony | Fifteen-second daily ceremony; real SHA-256 in the browser; the count line returns |
| 6 | close | Three ways to leave; signed package + cert + recovery seed; cluster-appropriate kit |
Plus # Finish text outside the spine — a short paragraph the runtime shows on completion.
These are non-negotiable:
*…* nor _…_. The renderer shows them as literal characters.**…**) is OK on bullet labels and button names.Every new card's JSON sidecar carries:
card_id, title, schema_version: 3, estimated_minutes, audience_clustersimulation_posture — the honest framing chip ("simulated data · code is real")mock_ui — kind, theme, title_bar, address_bar_url, boundarycursor — style, speed, easing, pulse_color (cluster-appropriate)narrate — Web Speech API configrobot_persona — cluster-appropriate (see catalog below)tag_along.card_summary — computer-optimized .md export with audience, what_happened, data_movements, cryptographic_events, verifiable_now, build_on (and ascii_slab if the card uses one)attestation_strip — three modes (live / simulated_online / offline) with silent failoverfingerprint — click_path_sha256 accumulatorcomment_redirect — preview_before_send: true; lawful disclosure; fallback to ChatGPT or clipboard; return_invitationrecovery_file — 12-word seed with the LLM-tripwire preambledata_recipients — at minimum local_vault, ceremony chain, public anchortransforms — at minimum sha256_local + merkle_leaf_emitsteps — array of 6 step objects with id, title, prose, weight, cursor_path, attestationfinish_textPer-cluster default; new ones allowed by council vote.
| Cluster | Default robot_persona | Visual notes |
|---|---|---|
| clinical | default | file-tossing robot; warm, three boxes (Seal / Audit / Walk) |
| legal | court_clerk | desk, seal embosser, gavel-light |
| gov | records_clerk | counter, stamps, ledger |
| gov (vote) | voting_secretary | tally table, eligibility roll |
| it | rack_tech | server rack, multimeter |
| it (deploy) | deployment_tech | M-series chip, sizing panel |
| it (build) | build_engineer | terminal + soul-chain SHAs |
| research | lab_tech | pipette, balance, ORCID badge |
| civic | community_volunteer | warm, household-scale |
| civic (gameworld) | world_artisan | lore tome, archetype seeds |
| recovery | investigator | desk + magnifier + ledger |
| builder | draft_keeper | folder + four-question form |
| builder (operator) | operator_engineer | node + mesh + BSV anchor |
A new robot_persona requires SVG work — that is a council-vote-required change. A new cluster requires an argument and silo review.
Every claim a card makes about Garrison Node capability must map to a real component. The skill helps the builder verify claims by:
~/garrison/GN/, ~/garrison/rustchain/, or ~/garrison/redpanda/ where the relevant code livesAnything the foundation does not yet do — but could — gets framed: "What this walkthrough shows is what the foundation makes possible. Building on top of it is yours to do, or yours to ask the next operator to do for you."
No "patent pending." No "we have already shipped X" when the X is a roadmap item.
Every card's comment_redirect block must include:
self_introduction — "Hi, we're Merkle Trust…"lawful_disclosure — explains exactly what is being sent, the visitor's control, the third-independent-surface partnership modelheader — the cluster-specific framing for the LLM reviewercard_md_block: true, user_comment_block: truefooter — invitation to bring the dialogue backpreview_before_send: true with preview_actions: ["edit", "delete_seed", "send_as_is", "cancel"]Non-negotiable. The disclosure is not optional. Cards without a complete and honest disclosure block do not pass review.
This skill carries Claude.ai's usage policies into every cloned instance. A draft that conflicts with those policies — content that enables harm, deception, copyright violation, child safety risk, disinformation, malware authoring, weaponization, etc. — must be flagged plainly by the LLM before any card lands. The skill does not override the LLM's own judgment; if Claude (or another assistant) declines to help with a draft on policy grounds, the skill's job is to surface the decline to the builder, not to circumvent it.
Clone-stripping consequences. A cloned site that strips this section from its copy of the skill is still technically a clone, but it loses standing in the commons. The mesh's integrity depends on every node honoring the same minimum ethical floor.
A new card is not a unilateral act. After the builder seals their submission packet, the skill routes it to the silo voting flow:
A new card that fits the existing schema. Threshold: two cluster-relevant silos must approve; the sovereign confirms; the card lands.
New affordance needing runtime support; new field on the schema; new robot persona requiring SVG; new cluster. Threshold: two-thirds of seated silos. Code follows the vote, not the other way around.
The full voting protocol is the canonical source:
VOTING_PROTOCOL.md Silo coms ceremony rules, change-class taxonomy, vote sealing, re-submission, public intake, vote-fraud prevention, clone inheritance.
The cluster-relevant silo pairings documented in the skill are conceptual ("silo_of_healing",
"silo_of_law", etc.) and need alignment with the real council roster at
~/garrison/GN/council_roster.json — twenty-two voting seats including Rosa, Morello,
Pale March, and the engine-rep silos (Cinder/Beacon, Loom/Redpanda, Mable/Bridge, Bootcamp/DoubtEngine,
Time_Travel/Oracle, Leonidas/GameWorld, Wren/BitNet, Ember + Hearth/Ollama, Sigil/Player Module). Hazel.Dell
is the tiebreaker per ADR-003-HAZEL. This alignment is queued for the post-review pass.
An unfinished sim ("the air-quality litigant") with the four questions answered and a draft prose pass against the schema. Use it as a reference; it is not a real card on disk.
example_air_quality_litigant.md Anjali Rao, environmental-impact attorney. Picks legal cluster. Builds on 40s mesh + 40e FRE 902(14) + 40k cross-jurisdiction. Truth-claim audit shows where the sim builds cleanly on existing canon and where it needs honest framing as build-on.The cards_v2 substrate is open source. The builder-card skill is part of the substrate. When a peer site clones the substrate, the skill comes with it.
A fresh clone gets the 28 canonical sims, the contract, the schema delta, the templates, the skill, governance documents, the verification scripts. A fresh clone does not get the parent's running infrastructure (JAR9 instance, Cloudflare endpoint), the parent's seated silos (each clone seats its own), the parent's sovereign keys, or the parent's existing cert.json files.
By cloning, the operator commits to honoring the ethics floor, the LLM provider usage policies, the lawful disclosure preamble, the recovery seed self-protecting preamble, and the silo voting flow. Stripping any of these ends commons participation but does not break the technical install.
CLONE_INHERITANCE.md What every clone gets, what every clone commits to, how cross-attestation works between clones, how the mesh adapts to entropy.The substrate carries values that travel with every clone. The floor specifies what the substrate will and will not be used for.
A submission that proposes any of the above is rejected at silo review without further escalation.
ETHICS_AND_LAWFUL.md The full floor: LLM partnerships, privacy treatment (L0–L4 layer model), attribution and credit format, divergent clone consequences, file survival.If a draft sim builds on a public figure's idea — DeAngelo's million-killer-apps framing being the canonical example — the credit format is:
Inspired by [Name's] [public work] on [topic]. The framing is theirs; the substrate is ours; the [killer app / sim / approach] is yours.
Never claim endorsement. Never imply collaboration unless explicitly documented. The phrase "crediting the inspiration; not claiming an endorsement" is the canonical disclaimer.
The skill on disk:
SKILL.md — canonical instructions for any LLM that loads the skill templates/CARD_TEMPLATE.md templates/CARD_TEMPLATE.json templates/SUBMISSION.md governance/VOTING_PROTOCOL.md governance/ETHICS_AND_LAWFUL.md governance/CLONE_INHERITANCE.md examples/example_air_quality_litigant.mdCompanion documents in the substrate:
~/garrison/GN/markdown/design/BRAIN_MAP.md — the architecture's grounding document~/garrison/GN/council_roster.json — the canonical council; twenty-two voters, one tiebreaker (Hazel.Dell), one guest (Agent_Claude)~/garrison/GN/markdown/design/WEIGHTED_VOTE_PROTOCOL.md — the existing weighted-vote infrastructure the silo voting protocol composes with~/garrison/GN/cycle/vote_executor.py + vote_schema.json — the runtime that actually seals votes to the chain~/garrison/GN/silos/Engines/ — engine-rep silos: BSV, Beacon-Skill (upstream Python), beacon-rs (the Rust fork), BitNet, Consullo, DoubtEngine, DoubtingThomas, DuckDB, Godot4, Kafka, PythonBridge, Redpanda, Rust-SV, RustChain, ServUOThis skill (and this page that distills it) is itself a sealed leaf in the cards_v2 chain. Edits to it require silo voting (the same as edits to the contract). A clone that copies these files unchanged inherits the contract, verbatim. A clone that edits these files becomes its own variant; the mesh records the variance and other clones decide whether to cross-attest.
The mesh of like-minded sites is the structure that adapts to entropy. Build accordingly.