Skip to content

sealad886/repository-guidance

v0.1.0MIT

Create, adapt, and audit repository instructions against local architecture, tooling, tests, and authority instead of importing another project’s assumptions.

GitHubOpen Website

Repository Guidance

Version: 0.1.0. Create, adapt, and audit coding-agent instructions using the repository’s actual architecture, commands, tests, and delivery policies.

The repository-guidance skill turns useful engineering conventions into evidence-backed local rules. It maps each relevant candidate to adopt, adapt, not applicable, conflict, or unknown; then produces the requested audit, draft, or focused instruction-file patch.

Install and use

Register this marketplace and install the plugin with your client:

codex plugin marketplace add sealad886/sealad886-codex-marketplace --ref main
codex plugin add repository-guidance@sealad886-codex-marketplace

copilot plugin marketplace add sealad886/sealad886-codex-marketplace
copilot plugin install repository-guidance@sealad886-codex-marketplace

Start a fresh task after installation. In Codex, invoke $repository-guidance:repository-guidance. In Copilot, ask it to use the repository-guidance skill; client invocation surfaces can differ.

Example requests:

  • “Use repository-guidance to audit this repo’s AGENTS.md. Report stale commands, unsupported rules, and ownership gaps without editing files.”
  • “Use repository-guidance to adapt the Multica-inspired engineering conventions to this repository and update its existing AGENTS.md.”
  • “Draft instructions for this package from its code, CI, and contribution guide. Keep the root policy authoritative and explain unresolved decisions.”

An audit or draft request does not authorize file edits. An instruction-edit request allows the scoped local patch; it does not itself authorize commits, installation, external writes, or release. Existing user and repository authority continues to apply. Substantial authorized replacements retain a local backup.

How adaptation works

Inspect the target’s applicable instructions and current repository state, locate its real owners and command sources, and evaluate only relevant concerns. Keep shared rules with their canonical owner and scoped rules with their package or platform. Verify referenced paths and commands; state when a command was only inspected rather than executed. Report missing evidence instead of inventing rules.

The adaptation lenses include a worked mapping for this marketplace’s package boundaries, paired manifests, canonical/generated catalogs, Python environment, both client checks, and publishing authority. The optional adaptation record can be replaced by an existing project format or a concise response.

Sources and limits

This original synthesis was informed by Multica’s AGENTS.md at d7d1831. See source provenance for the inspected revision, research date, license distinction, and excluded project-specific assumptions. No upstream source text, runtime, branding, or assets are included; this plugin has no Multica affiliation or endorsement.

The package contains one shared skill for Codex and GitHub Copilot. It includes no MCP server, network client, credentials, shell runtime, or required sibling plugin. The host supplies repository-reading/editing tools and enforces approvals. Browsing is optional unless the task requests current external-source verification. A valid package does not prove that every target host automatically loads AGENTS.md or that future agents follow it. See security boundaries.

Maintainer validation

From the marketplace checkout, use its existing Python environment:

PYTHONDONTWRITEBYTECODE=1 .venv/bin/python scripts/check_plugin.py plugins/repository-guidance --layout source
PYTHONDONTWRITEBYTECODE=1 .venv/bin/python scripts/check_distribution_bundle.py plugins/repository-guidance

Follow the current marketplace contribution guide for both catalog checks, skill validation, isolated native client discovery and installation, installed-file parity, and release scanning. Documentation and package checks establish structural validity; evaluate the actual adaptation’s claims against the target repository before accepting its guidance.