project-agent-instructions

Create, refresh, adopt, or retire selected-project AGENTS.md rules with deterministic ownership and recovery.

7|5|Updated Jul 22, 2025
One-click install
npx skills add https://github.com/nebius/nebius-ps-services --skill project-agent-instructions-nebius
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: project-agent-instructions
Source: https://github.com/nebius/nebius-ps-services/tree/main/skills/project-agent-instructions
Command: npx skills add https://github.com/nebius/nebius-ps-services --skill project-agent-instructions-nebius

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Project-specific agent instruction files (AGENTS.md) drift out of sync with requirements and design documents, get edited by hand without provenance, or are created when they are not needed. This Skill maintains the smallest durable, evidence-backed project instruction contract for one exact selected project, with deterministic rendering, explicit ownership tracking, and guarded recovery. ## Core Features & Use Cases - Receipt-bound decisions: Replays a maintainer-issued spec-validation receipt before trusting requirements or design, then decides between created, attached, refreshed, adopted, retired, existing-sufficient, or not-needed outcomes. - Deterministic rendering: Renders rules into six fixed sections with hard limits (12 rules, 256 bytes per rule, 4 KiB body) without mutating the repository until an authorized apply step. - Ownership and recovery: Tracks managed regions through a workspace-private registry, preserves human-authored prefixes byte-for-byte, and fails closed on lock files, backups, legacy markers, or conflicting evidence. - Use Case: After validating docs/requirements.md and docs/design.md, route an explicit project-instruction decision here to attach a managed compatibility-rules tail to an existing AGENTS.md while preserving every human-written byte. ## Quick Start Ask the agent to use project-agent-instructions to inspect the selected project with its current spec receipt and decide whether an AGENTS.md update is needed.

Frequently Asked Questions about project-agent-instructions

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

FAQPage Schema
How do I create or update an AGENTS.md file for my project?

Invoke this Skill only after a spec owner issues a validation receipt for the current requirements and design files. It inspects the selected project, decides whether rules are needed, renders them deterministically, and applies the change through a guarded helper rather than direct file edits.

When should project agent instructions not be created?

A missing AGENTS.md is not evidence that one is needed. When no durable, project-specific, evidence-backed rule remains, the correct outcome is not-needed: no file is created or changed, and a missing target stays absent.

Does this work with Claude Code as well as Codex?

Yes. Claude requires an existing tracked CLAUDE.md containing an @AGENTS.md import, plus a native runtime declaration of instruction and settings file digests. Codex uses its profile and config discovery chain; both hosts pass explicit --agent and --agent-home values.

What happens if a human edits the generated AGENTS.md region?

Human edits to the prefix remain allowed and are preserved byte-for-byte. Any edit to the managed marker or body transfers that region out of automation ownership, and further mutation requires exact-digest adoption approval.

Why does the workflow fail with RECOVERY_REQUIRED or OWNERSHIP_CONFLICT?

A surviving lock or backup artifact blocks every transition until a human resolves it, and missing or stale ownership evidence blocks mutation. These fail-closed blockers prevent unsafe writes and are never bypassed automatically.