The interface is localized. Skill descriptions and documents are machine-translated by AI into your language; code, commands, paths and technical terms stay in the original, and the author's document governs.
Explore skills
Development · SKILL.md

agent-integration

agent-integration

Install Graphify assistant skills, always-on guidance, and hooks safely across supported agent platforms.

Independent GitHub snapshot · Automated classification · Original documentation governs

Skill documentation

GitHub

name: agent-integration description: "Install Graphify assistant skills, always-on guidance, and hooks safely across supported agent platforms." metadata: disco-role: operating disable-model-invocation: true license: NOASSERTION

agent-integration

Use this sub-skill when the user wants to install, inspect, repair, uninstall, or package Graphify's assistant integration layer: packaged SKILL.md files, references/ sidecars, always-on project guidance, PreToolUse/BeforeTool hooks, Git hooks, strict mode, and user-vs-project scope decisions.

Start from the graphify root router when the complete runtime tree includes one and the request may involve non-install workflows too.

Route here for

  • graphify install, graphify uninstall, graphify <platform> install, or graphify <platform> uninstall.
  • Choosing a supported assistant platform, including aliases such as --platform skills for agents.
  • Deciding user scope versus project scope, and interpreting project-scoped git add hints.
  • Installing, inspecting, or removing graphify hook install/status/uninstall and assistant PreToolUse/BeforeTool hooks.
  • Claude strict mode and the GRAPHIFY_HOOK_STRICT runtime override.
  • Troubleshooting graphify not found, wrong PyPI package name, stale installed skill versions, missing references/ sidecars, Codex's no-op hook, or platform-specific install artifacts.
  • Maintainer-only questions about how Graphify renders and packages its platform skill bodies.

Route elsewhere

  • Building, updating, watching, or validating graphify-out/ graphs: use graph-building.
  • Querying an existing graph with query, path, explain, MCP serving, or query-first answer behavior: use query-navigation.
  • Export, merge, database, PR, or visualization workflows: use exports-integrations.
  • File-format/parser/extractor failures after install succeeds: use extractor-troubleshooting.

Safety defaults

  1. Do not mutate a real HOME or project config unless the user explicitly asked for that install/uninstall scope. For exploration, use the bundled temp probe from the graphify repo-skill root: python sub-skills/agent-integration/scripts/install_platform_probe.py --platform codex --scope project.
  2. Distinguish the package name from the command. Install the official PyPI package as graphifyy; the CLI and import package are graphify.
  3. Prefer project scope for team-visible assistant guidance. graphify install --project --platform <platform> writes artifacts under the current repository when that platform supports project install and prints git add hints.
  4. Use user scope only for personal assistant skills. graphify install --platform <platform> writes under the user's assistant config tree; ask before doing this in a real HOME.
  5. Treat skill-only installs and always-on guidance as related but not identical. Some graphify install --platform ... commands copy only a skill body, while platform subcommands such as graphify codex install wire project instructions or hooks.

Operating workflow

  1. Confirm target and scope. Identify the assistant platform, user/project scope, whether the user wants skill-only install or always-on guidance, and whether uninstall should also purge graphify-out/.
  2. Verify CLI availability. Run graphify --help or python -m graphify --help. If missing, use troubleshooting before attempting install.
  3. Probe safely when uncertain. Run install_platform_probe.py with a temporary HOME/project and inspect the summarized files.
  4. Install with the smallest correct command. Use platforms.md to choose the platform command and expected artifacts.
  5. Inspect hooks and always-on files. Use always-on-and-hooks.md for hook/strict behavior and graphify hook status for Git hooks.
  6. Repair or uninstall precisely. Use platform-specific uninstall or graphify uninstall --project --platform <platform> for project scope; use graphify uninstall --purge only when the user also wants graph outputs removed.
  7. For maintainer packaging work only, load platform-skill-packaging.md.

Quick command patterns

# Recommended package install when Graphify is missing.
uv tool install graphifyy

# Skill-only default: Claude Code on Linux/macOS, Windows variant on Windows.
graphify install

# Named user-scope skill install. Ask before using a real HOME.
graphify install --platform codex

# Project-scoped install with commit hints.
graphify install --project --platform codex

# Platform always-on guidance in the current project.
graphify codex install
graphify claude install

# Claude Code project strict mode.
graphify install --project --strict

# Project Git hooks that keep graphify-out fresh after commits/branch switches.
graphify hook install
graphify hook status
graphify hook uninstall

# Remove assistant integrations; add --purge only to delete graphify-out/.
graphify uninstall
graphify uninstall --project --platform codex
graphify uninstall --purge

Verification cues

A healthy progressive platform install has SKILL.md, .graphify_version, and usually a references/ sidecar in the platform's graphify skill directory. Always-on platforms additionally have a project instruction file such as AGENTS.md, CLAUDE.md, GEMINI.md, .cursor/rules/graphify.mdc, .github/copilot-instructions.md, .kiro/steering/graphify.md, or .agents/rules/graphify.md. Hook-enabled platforms have JSON settings, native plugin files, or Git hook files as described in the references.

Saved skills