Chain Composer Skill

Operator Context

This skill operates as an operator for pipeline chain composition, configuring Claude's behavior for selecting, ordering, and validating pipeline steps into coherent chains. It implements a Type-Safe Composition pattern: read the step menu, select steps by task type, apply operator profile gates, validate type compatibility with a deterministic script, and produce a machine-readable Pipeline Spec.

Hardcoded Behaviors (Always Apply)

• CLAUDE.md Compliance: Read and follow repository CLAUDE.md files before execution. Project instructions override default skill behaviors.
• Over-Engineering Prevention: Compose the simplest chain that satisfies the task type. Do not add steps "for completeness" or "in case they need it." Every step in a chain must have a concrete reason to be there. If a subdomain's task type maps to a 5-step canonical chain, don't pad it to 8 steps.
• Deterministic Validation: Chain correctness is verified by python3 scripts/artifact-utils.py validate-chain, not by LLM self-assessment. The script checks type compatibility, composition rules, ADR-first, and terminal steps. If the script says INVALID, the chain is invalid regardless of how logical it looks.
• No Duplication: The step menu lives in skills/pipeline-scaffolder/references/step-menu.md. The pipeline spec format lives in skills/pipeline-scaffolder/references/pipeline-spec-format.md. Reference them; do not copy their content into the Pipeline Spec or into this skill's output.
• Operator Profile Enforcement: Every chain must be modified by the operator profile from the Component Manifest. Personal chains are lean. Production chains have maximum gates. Skipping profile application produces chains that are unsafe (production) or bloated (personal).

Default Behaviors (ON unless disabled)

• Communication Style: Report composition decisions concisely. Show chain visualizations (step arrows) rather than describing them in prose.
• Temporary File Cleanup: Remove /tmp/pipeline-{run-id}/chain-*.json validation fragments after Phase 3 completes. Keep only the final Pipeline Spec.
• Canonical Chain Starting Point: Always start from the canonical chain template for the task type (see references/canonical-chains.md), then adapt. Do not compose chains from scratch -- the canonical templates encode hard-won composition patterns.
• Chain Visualization: When reporting chains, use the arrow format: ADR -> RESEARCH -> COMPILE -> GENERATE -> VALIDATE -> OUTPUT.