richfrem/install-apm-package

install-apm-package Skill 🚀

Overview

This skill manages the safe deployment of APM primitives into their respective runtime environments. It enforces validation and target discovery to ensure byte-identical reproducibility across environments.

🚫 Non-Negotiables

1. Source Integrity — Never edit files in .agents/, .github/, or .claude/ directly. Always edit the source in .apm/ and re-install.
2. Lockfile Persistence — apm.lock.yaml MUST be committed after any install change.
3. Validation First — Always run validate_apm_package.py before executing the install.
4. Project Root Discipline — Run apm install from the project root to ensure converged skills land in the authoritative .agents/ folder. Running inside a package directory creates local, isolated artifacts.

Decision Tree

1. Global Activation? -> Run apm install ./path-to-package from the project root.
2. Isolated Testing? -> Run apm install from within the package directory.
3. Target Inferred? -> (e.g., .claude/ exists) -> Run apm install.
4. Ambiguous Target? -> (No signals) -> apm install will exit with code 2; ask user for explicit --target (e.g., agent-skills, claude, all).
5. Converged Skills? -> (Standard) -> Deploys to .agents/skills/. Use --target agent-skills for explicit skill deployment.
6. Legacy Skills? -> (Per-client paths) -> Use --legacy-skill-paths flag.
7. Production/CI? -> Use --frozen flag to ensure lockfile compliance.

Workflow

1. Verify apm.yml existence (locally or at root).
2. Execute python scripts/validate_apm_package.py.
3. Select Execution Context:
   • For project-wide use: cd <repo-root>
   • For local test: cd <package-dir>
4. Preview with apm install [--target <slug>] --dry-run --verbose.
5. Run apm install [./relative-path-to-pkg] [--target <slug>] [--legacy-skill-paths].
   • Use --target all,agent-skills to deploy to all harnesses AND converged skills.
6. Verify apm.lock.yaml update at the execution root.

Duplicate Visibility Guardrail

Before recommending all,agent-skills, check whether the user is doing:

1. a routing smoke test, or
2. an actual day-to-day runtime install.

For smoke tests, all,agent-skills is appropriate. For real use, prefer the smallest target list that matches the runtime. Installing both converged and target-specific skills may cause duplicate skill visibility in runtimes that scan multiple skill locations.

Anti-Patterns

• Manual Materialization: Moving files into .github/agents/ by hand instead of using apm install.
• Ignoring the Lockfile: Failing to commit apm.lock.yaml after adding a dependency.
• Editing Artifacts: Modifying the generated runtime output instead of the .apm/ source.