instructions

Audits agent instruction files for duplicated prescriptions and undocumented project helpers.

4|Updated Jun 9, 2026
One-click install
npx skills add https://github.com/laicluse/agent-fieldkit --skill instructions-laicluse
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: instructions
Source: https://github.com/laicluse/agent-fieldkit/tree/main/.agents/plugins/generated/drydry/skills/instructions
Command: npx skills add https://github.com/laicluse/agent-fieldkit --skill instructions-laicluse

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Agent instruction files like AGENTS.md and CLAUDE.md can themselves cause code duplication: they may prescribe two different paths for the same job, or stay silent about existing project helpers, causing agents in fresh sessions to reinvent utilities that already exist. ## Core Features & Use Cases - Two-prescribed-paths detection: Finds instruction rules that prescribe conflicting workflows (e.g., two ways to authorize, commit, or test) and proposes a convergence direction. - Silent-helper detection: Scans project helper directories for reusable symbols the instruction files never mention, flagging high re-invention risk. - Verifier-backed findings: Every finding ships with a runnable grep command and a drift hypothesis; findings without them are dropped. - Use Case: During a drydry duplication audit, pass your project's AGENTS.md and CLAUDE.md paths plus the project root, and receive a structured markdown section of instruction-file duplication findings for the orchestrator's audit artefact. ## Quick Start Ask the drydry audit pipeline to audit your instruction files by supplying the instruction_paths and project_root arguments.

Frequently Asked Questions about instructions

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

FAQPage Schema
How do I audit AGENTS.md or CLAUDE.md files for duplication problems?▼

Pass the instruction file paths as the instruction_paths argument along with the project_root. The skill reads each file, indexes its prescribed rules, and flags cases where two rules prescribe different paths for the same job.

What is a silent-helper finding in an instruction file audit?▼

A silent-helper finding flags a project helper (like a tenancy wrapper or notification utility) that the instruction files never mention. Agents in fresh sessions then reinvent the helper, producing parallel implementations that drift on edge cases.

Can this skill edit or fix my instruction files automatically?▼

No. The skill returns findings only and never edits instruction files. Convergence proposals tell the operator where to add pointers or which path to name as primary, but the operator decides and applies changes.

Why are findings without a verifier command dropped?▼

Every finding must include a runnable grep so the operator can re-confirm the duplication or silence in the codebase. Findings lacking a verifier command or a drift hypothesis are discarded to keep the audit artefact actionable.

Does the audit treat user-level instruction files differently?▼

Yes. User-level files are read and audited if included in instruction_paths, but convergence proposals for them are phrased as operator considerations rather than edit directives, since user-level config is personal.