pm-skills

PM Skills Arsenal

AI produces polished analysis that changes no decisions. These skills fix that.

12 PM skills — 1,000–1,300 lines each — for senior PMs, founders, AI engineers, and the agents that route between them.

Personal research, separate from my day-job at Microsoft. MIT-licensed.

Quality benchmark (self-administered — read with that lens)

Condition	Score
Baseline (Claude, no skill)	47 / 105
Anthropic's PM Skill	81 / 105
PM Skills Arsenal	98 / 105

5 prompts × 7 dimensions × 3 conditions. The same author wrote the rubric AND scored every output — this is the author's own assessment, not an independent evaluation. The full methodology, the 7-dimension rubric, and all 15 raw outputs are published at benchmark/ so you can re-score them yourself. If you do and reach a different conclusion, open an issue — independent verification beats author-administered scores every time.

Live site with interactive showcases → · Machine-readable manifest (AGENTS.md) →

Competitive & Market Analysis	Executive Writing
Metric Design & Experimentation	Product Strategy

Install

Via APM (recommended — one install, every harness)

APM — the Agent Package Manager — is a multi-harness manifest. One command configures the plugin across GitHub Copilot CLI, Claude Code, Cursor, OpenCode, Codex, and Gemini from a single apm.yml.

# pin to the released version for reproducibility
apm install Avyayalaya/pm-skills-arsenal#v2.1.0

# or floating on main (will warn about unpinned dependency)
apm install Avyayalaya/pm-skills-arsenal

The install brings the 12 PM skills, the pm-skills MCP server, and lifecycle telemetry hooks. Skills auto-register with each compatible harness.

First-time setup tip. APM auto-detects your harness from directory markers (.claude/, CLAUDE.md, .cursor/, .github/copilot-instructions.md, .codex/, .gemini/, etc.). In a fresh project with no markers yet, pass --target explicitly:

apm install Avyayalaya/pm-skills-arsenal#v2.1.0 --target claude
# or: --target copilot, cursor, codex, gemini, opencode, windsurf

Run apm targets to see the full list of supported harnesses.

Via awesome-copilot marketplace (curated as of 2026-05-20)

github/awesome-copilot — the canonical APM aggregator marketplace — accepted this package into its plugins/external.json on 2026-05-20. If you already have that marketplace registered, install via:

apm marketplace add github/awesome-copilot   # first time only
apm install pm-skills@awesome-copilot

Same source repo (this one), same release ref (v2.1.0). Useful when awesome-copilot is already your team's go-to APM index.

Direct (per-harness fallback)

Claude Code:

claude plugin marketplace add Avyayalaya/pm-skills-arsenal
claude plugin install pm-skills

GitHub Copilot (Agency marketplace):

Skills are listed in the Agency marketplace. Full setup: GitHub Copilot Installation.

MCP server (Claude Desktop, Cursor, Cline, custom agents):

{
  "mcpServers": {
    "pm-skills": {
      "command": "python",
      "args": ["<absolute-path>/pm-skills-arsenal/mcp/pm_skills_mcp_server.py"]
    }
  }
}

Five tools: list_skills, get_skill, list_agents, get_benchmark, run_skill. MCP setup →

Any other LLM (ChatGPT, Gemini, etc.):

Copy the relevant SKILL.md into your context window. Each is self-contained.

Built for agents

This system is designed to be discovered, evaluated, and routed by other AI agents — not just read by humans.

AGENTS.md — a machine-readable capability manifest at the repo root. Lists every skill, its frameworks, line count, version, and benchmark score. An orchestrator can decide whether to load a skill without reading 1,300 lines first.