| name | install-user-instructions |
| description | Install, update, or audit general user-level AI coding instructions across Claude Code, OpenAI Codex, GitHub Copilot CLI, and Cursor. Use when the user asks to set up global instructions, personal AGENTS.md or CLAUDE.md defaults, configure instructions for multiple agent systems, avoid repeated downstream project rules, install model-routing defaults, or migrate Claude-only user rules to cross-agent user scope. |
Install User Instructions
Install shared personal defaults at each selected system's supported user
surface without overwriting custom content or copying global rules into every
repository.
Workflow
- Detect available systems with read-only checks (
command -v plus existing
target files). Do not infer that every detected system should be modified.
- Ask the user which of Claude Code, Codex, Copilot CLI, and Cursor to
configure. This selection is required before any user-home write.
- Read
references/platform-paths.md for the selected systems. Resolve
documented home overrides before default paths.
- Choose the instruction content:
- use
assets/global-instructions.md for the project defaults;
- use a user-supplied file with
--template when requested;
- never silently copy a repository's project-specific instructions into
user scope.
- Run
scripts/install_user_instructions.py --systems <csv> without
--apply. This is the mandatory dry-run and diff preview.
- Show every destination and diff. For Cursor, show the copy/paste block for
Cursor Settings > Rules; do not invent a filesystem destination.
- Ask for confirmation of the exact selected destinations and changes.
- After confirmation, rerun with
--apply. Writing outside the workspace may
require the runtime's normal filesystem approval.
- Verify each selected system using the command or UI check in
references/platform-paths.md.
- Audit the current repository for duplicated general rules. Report exact
duplicates or conflicts, but remove project content only with separate
user approval.
Non-destructive contract
- Manage only the block between
<!-- research-skills:global-instructions:start --> and
<!-- research-skills:global-instructions:end -->.
- Preserve all content outside the managed block byte-for-byte.
- Fail on unmatched or repeated markers instead of guessing.
- Default to preview mode.
--apply is required for writes.
- Write only explicitly selected systems.
- Keep Cursor manual until its official User Rules surface exposes a stable,
supported file path.
- Keep general personal defaults at user scope. Repository
AGENTS.md,
CLAUDE.md, .github/copilot-instructions.md, and .cursor/rules contain
only project facts or system-specific/path-scoped deltas.
- Do not duplicate shared project rules in
CLAUDE.md; import AGENTS.md and
append Claude-only deltas.
Script examples
Preview Claude, Codex, and Copilot targets:
uv run python scripts/install_user_instructions.py \
--systems claude,codex,copilot
Apply the approved preview:
uv run python scripts/install_user_instructions.py \
--systems claude,codex,copilot --apply
Preview Cursor's manual User Rules block:
uv run python scripts/install_user_instructions.py --systems cursor
Use absolute paths to the skill script when the current working directory is
not this skill directory.
Content policy
The bundled defaults encode the central agent-fanout routing contract:
strongest-tier lead judgment, intermediate approved-plan elaboration, and
worker-tier implementation with explicit tests and gates. They also require
closing/removing completed one-off agents.
Use semantic line breaks in prose source generally. GitHub issue and
pull-request bodies are the exception: keep each paragraph on one source line,
separate paragraphs with blank lines, and do not insert sentence- or
clause-level newlines inside a paragraph.
