chain

Execute YAML-defined multi-step skill workflows with schema validation, gates, and JSONL audit logging.

Updated Jul 12, 2026
One-click install
npx skills add https://github.com/sumitake/agent-collab --skill chain-sumitake
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: chain
Source: https://github.com/sumitake/agent-collab/tree/main/plugins/agent-collab/skills/chain
Command: npx skills add https://github.com/sumitake/agent-collab --skill chain-sumitake

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Running several skills in sequence by hand is inconsistent and leaves no verifiable record. This Skill turns a YAML-defined sequence of skill invocations into a reproducible, audited workflow with input validation, conditional steps, gates, and a JSONL audit trail. ## Core Features & Use Cases - YAML Chain Execution: Parse a chain definition and run each step in order, dispatching to named skills or to the active primary's own reasoning. - Validation, Gates, and Retries: Validate each step's output against a declared schema, retry on malformed output, and enforce filesystem, semantic, and tool gates with per-gate failure policies. - Audit Logging and Isolation: Write an append-only JSONL audit log per run, support per-step git worktree isolation, and route cross-check steps through family-aware verifier-independence rules. - Use Case: A team runs a release-validation workflow repeatedly. They define it once as a chain YAML, then say "run the release-validation chain" to get consistent execution, gate enforcement, and a locatable audit log every time. ## Quick Start Run the validate-release chain defined in the project's chains directory on the current change set.

Frequently Asked Questions about chain

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run a multi-step skill workflow as a single chain?▼

Define the workflow as a chain YAML in the project's chains/ directory, then ask to run that chain by name. The active primary parses the YAML, executes each step in order, validates outputs, and writes a JSONL audit log.

How do I chain multiple AI skills together with validation?▼

Write a chain YAML where each step declares a runner, input templates, and an output schema. The chain runtime renders inputs, dispatches each step, validates output against the schema, retries on malformed output, and applies declared failure policies.

Where does the chain skill look for chain YAML definitions?▼

It searches in priority order: the project's chains/ directory, the user-scoped agent-collab chains directory, then drafts/sample-chains/ for prototypes. Missing required fields cause a halt with a clear error.

Can a chain resume in a new session after interruption?▼

No. This is an instruction-native, single-session runtime, so chain state lives in the current conversation and cannot resume in a new session. The JSONL audit log persists on disk as the record of completed steps.

When should I not use a chain workflow?▼

Skip chains for one-off tasks with no repetition, novel scope with no documented chain, or sequences of only one or two skills without schema or audit needs. Direct skill invocation is simpler in those cases.

How does the chain skill ensure independent cross-checking?▼

Cross-check steps and semantic AI gates route through a family-aware policy that excludes the primary and artifact-author model families and fails closed on unknown lineage. Same-family cross-checks are rejected to preserve verifier independence.