Arkiv skills

Agent skills for working with Arkiv — the Web3 database, powered by $GLM. Each skill follows the Agent Skills format and provides focused guidance for compatible agents.

Available skills

Skill	What it does
arkiv	Routes Arkiv tasks to the appropriate focused skill.
arkiv-first-write	Sets up Tiramisu and verifies a first entity and read-back.
arkiv-data-modeling	Designs attributes, payloads, relations, and schema mappings.
arkiv-query	Builds typed queries, pinned pagination, counts, and historical reads.
arkiv-entity-lifecycle	Explains flags, patching, ownership, deletion, and backup/restore.
arkiv-entity-expiration	Chooses expiration deadlines and reconciles Lifetime Extension.
arkiv-write-safety	Serializes bounded batches and reconciles uncertain transaction outcomes.
arkiv-app-integration	Connects browser wallets, authenticated server routes, DTOs, and events.
arkiv-security-trust	Defines public-data, publisher, signer, and authorization boundaries.
arkiv-troubleshooting	Diagnoses exact errors, missing results, and stale reads.
arkiv-encryption	Encrypts payloads and explains key custody and remaining metadata exposure.
arkiv-large-files	Stores and verifies chunks/images with SDK 0.8.1 releases; plans hybrid pointers and recovery.
arkiv-social-graph	Models relationships and scopes arkiv-graph visualization.
arkiv-indexing	Builds Arkiv-to-app projections and explains the separate EVM-to-Arkiv sync package.
arkiv-feedback	Prepares sanitized bug/feature reports and submits only with authorization.
arkiv-mcp	Explains existing MCP profiles, read-only checks, consent-gated feedback and service privacy boundaries.
arkiv-best-practices	Deprecated compatibility entrypoint; start with arkiv.

The library guides cover published arkiv-chunking@0.1.1, arkiv-images@0.1.2, arkiv-sync@0.3.0 and create-arkiv-sync@0.3.0, verified with SDK 0.8.1. Read the installed package's AGENTS.md and relevant skill for configuration, tested scope and recovery limits.

Installation

Start with the router:

# npm
npx skills add Arkiv-Network/skills --skill arkiv

# pnpm
pnpm dlx skills add Arkiv-Network/skills --skill arkiv

Replace arkiv with a name from the index to install a specific skill. The default scope is the current project; add --global for your user account. Check npx skills --help for supported hosts and flags. --all installs every skill into every supported agent without prompts; choose a specific skill for a focused installation.

Load arkiv for an Arkiv task, then the task skill it selects. Use arkiv-first-write to install the SDK, configure Tiramisu, verify a first entity and propose the inline Arkiv rules for the project's AGENTS.md while preserving existing guidance.

The guidance targets SDK >=0.8.1 <0.9 and Tiramisu (chain ID 7738577). Before using another version, inspect its installed declarations and current official documentation. Keep credentials out of source code, URLs and chat; writes require an authorized signer and spending budget.

Each skill declares its checked SDK range, network, date and MIT license in frontmatter. These declarations describe the documented scope and do not certify every provider, application or future release.

For the combined Claude Code plugin, use the separate Arkiv plugin repository. It imports a fixed canonical revision and bundles the existing MCP connection and project rules. Read its installation and verification scope before use.

Contributors and agents should read AGENTS.md. Recent guidance covers complete atomic batch validation, separately authorized removal after readonly corrections, transfer provenance and the distinction between the General connection and Global Tour Stop's Devfolio profile. The resumable importer saves intent before wallet entry, distinguishes receipt-confirmed progress from verified fields and recovers without resubmitting a known transaction. Its version-2 journal fails closed on older journals; preserve those files and reconcile with the original importer.

Installed-import check

The router includes an advisory installed-import check, derived from a sanitized SDK 0.7.0 friction report. It reads named imports, package exports and installed declarations without executing application code or making network requests. Unsupported evidence stays unmeasured. In Claude Code the router's frontmatter hook runs it after Write/Edit on files that import the SDK; other hosts can run the script explicitly. The Arkiv plugin measures each release's guidance with a context-only ablation before shipping it.

node --test tests/snippets/import-check.test.mjs tests/snippets/skill-references.test.mjs
node scripts/check-skill-references.mjs
# Explicit network fixture: verifies official tarball integrity, never runs package scripts.
node tests/snippets/published-imports.mjs