Evaluate a new tool before it joins the stack — the problem-first framing (tools answer needs, not demos), the trial designed with success criteria upfront, the stack-fit check (integration, overlap, the tool-sprawl tax), and the security/data review sized to the stakes. Use when asked should we buy this tool, evaluate this software for the team, we have three tools that do this already, or run a proper trial before committing. Produces the need statement, the trial design with pre-set criteria, the stack-fit audit, and the adopt/decline verdict with its reasoning.
▸The problem, not the tool — what's broken/slow/manual today, for whom, costing what; "I saw this cool tool" gets reverse-engineered into its implied need, which sometimes evaporates on contact
▸The current stack — what's owned that's adjacent (the overlap audit needs the inventory — [contract-renewal-tracker](../contract-renewal-tracker/SKILL.md)'s list is the source); most orgs own 30% more capability than they use
▸What the tool would touch — customer data? Credentials? Just public content? The security review's depth follows ([skill-vetting](../skill-vetting/SKILL.md) blast-radius thinking, applied to SaaS)
▸The trial population — who'd actually test it (the enthusiast *and* a skeptic — enthusiast-only trials always pass)
✅ The bar it holds itself to
Every skill in this library self-verifies — these are this skill's own quality checks, straight from its definition.
✓The need statement predates vendor contact
✓The overlap audit ran against the real inventory with honest coverage
✓Trial criteria were written before day one and include a skeptic
✓The security review preceded real data entering the trial
✓The verdict is logged with reasons, adopt or decline
⚠️ What it refuses to do
Do not evaluate backwards from the demo — the need statement is the eval's spine
Do not skip the overlap audit — the cheapest tool is the one already owned and unconfigured
Do not run criteria-free trials — warm feelings always vote adopt
Do not pipe customer data into "just a trial" — the gate runs first at exactly that moment
Do not decline silently — the unlogged rejection is next year's rematch, at full cost
Install
npx pm-claude-skills add --agent claude # or codex · cursor · gemini · hermes
# or one-line MCP (every skill, any client):
claude mcp add pm-skills -- npx -y pm-claude-skills-mcp