Product Manager Skills
Turn Claude Code, Codex, Cursor, or Windsurf into a senior PM who pushes back: diagnose SaaS metrics, critique PRDs, plan roadmaps, run discovery, and coach career transitions.
SKILL.md explicitly forbids auto-running local scripts, update checks are manual-only, README claims no credentials, no telemetry, no privilege escalation, runtime is pure Markdown. Deductions: bin/update-check still makes a network request to GitHub (manual), license metadata is NOASSERTION at the system layer, publisher identity unverified, and data-flow/rollback disclosures rest on README self-attestation rather than verifiable evidence.
Instructions are self-consistent: routing table, mode selection, status protocol, and escalation rules cohere; abnormal input has BLOCKED/NEEDS_CONTEXT feedback paths. Static cap is 10, all example outputs are author-written with no test suite or independent reproduction, and the referenced knowledge/ and templates/ files were not in the provided scope to confirm existence. Scored 9.
Audience, scenarios, trigger intents (routing table), and non-fit ranges are clearly written; Chinese responses supported with a Chinese README. Deductions: very broad trigger surface (40+ frameworks, 7 domains) raises false-trigger risk; no statement on mainland-China reachability of clawhub/npm install paths; cross-platform behavioral fidelity is acknowledged as unvalidated by the author in TODOS.md.
Has a Keep-a-Changelog-format CHANGELOG, VERSION, CONTRIBUTING, ETHOS, bilingual READMEs, and TODOS maintenance path; layering is clear. Deductions: the provided SKILL.md text is truncated mid-sentence in Quality Gate 6, leaving key gate content incomplete; CC BY-NC-SA 4.0 (non-commercial) may restrict some enterprise use without explicit warning; system-layer license metadata is NOASSERTION.
Static cap is 7. README provides three expected-output examples (SaaS diagnostic, PRD review, coaching) that appear directly usable, and quality gates plus anti-pattern flagging add marginal value over generic prompting. Deductions: all examples are author-supplied with no independent reproduction; the comparative-benefit claim is supported only by a self-written comparison table.
Static cap is 5. Auditable primary material exists: CHANGELOG, example files, starter prompts, routing table, and structural claims (7 domains, 12 templates) are checkable in-repo. Deductions: no CI tests or third-party execution evidence; the 'Featured In' list and star counts are external claims not verifiable within this repository; the fact/inference boundary between authored examples and actual behavior is not explicitly separated.
- This is a static source review; no scripts or workflows were executed, and real interaction quality is independently unverified.
- The license is CC BY-NC-SA 4.0 (non-commercial, share-alike); confirm compliance before commercial use.
- bin/update-check is manual but still makes a network request to GitHub; use git clone in network-restricted environments.
- The provided SKILL.md text is truncated in the Quality Gates section; verify the complete file before relying on it.
- README example outputs are author-attested and should not be treated as verified behavior.
What does this skill do, and when should you use it?
Product Manager Skills is an Agent Skill that gives AI coding tools a reusable PM operator, built as one SKILL.md plus 7 knowledge modules, 12 templates, and 8 worked examples, all in plain Markdown. A routing table matches user intent to a domain, loads the relevant frameworks from 40+, and applies tiered quality gates before producing output. The skill is designed to be opinionated: it labels assumptions, names tradeoffs, and flags PM anti-patterns rather than writing polished filler. Installation is a one-line command, and runtime needs no network, scripts, or credentials.
Reads user intent and routes it through a routing table into domains covering discovery, strategy, delivery artifacts, finance, career, growth/PLG, and AI product craft; produces deliverables like PRDs, user stories, positioning statements, roadmaps, and opportunity solution trees from 12 templates; performs quantitative SaaS diagnostics (LTV, CAC payback, NRR, Rule of 40, magic number) from raw metrics; names anti-patterns such as Solution Smuggling, Metrics Theater, and Feature Factory in user input; activates a coaching mode on explicit request with follow-up questions and a closing verdict; ends every output with decisions made, assumptions to validate, a recommended next step, and a completion status (DONE, DONE_WITH_CONCERNS, BLOCKED, or NEEDS_CONTEXT).
- A technical PM who wants PRD review inside their coding tool: paste a draft and the skill flags solution smuggling, metrics without baselines, and overscoped releases.
- A founder or PM with raw business data: feed in 'MRR $50k, 500 customers, monthly churn 8%, CAC $500' and get a quantified health diagnosis naming the real bottleneck (in the documented example, retention rather than acquisition).
- A senior PM preparing for Director or VP interviews: request coaching mode and receive a gap diagnosis plus a concrete action plan based on the Altitude-Horizon framework.
- A growth team designing PLG strategy: get framework-based analysis on activation, viral loops, freemium conversion, and growth experiment design.
- An AI product PM stress-testing AI-native decisions: use dedicated modules for AI-Shaped Readiness, Context Engineering, and Agent Orchestration.
- Anyone wanting end-to-end flow: say 'take this from idea to PRD' and the skill runs a 6-phase PM Sprint covering Discover, Position, Prioritize, Specify, Validate, and Measure.
What are this skill's strengths and limitations?
- Pure, inspectable Markdown: no auto-executed scripts, no network calls, no credentials, fully auditable
- Structured output discipline: labeled assumptions, named tradeoffs, measurable outcomes, and explicit anti-pattern detection
- Broad coverage: 7 knowledge domains, 40+ frameworks, 12 templates, backed by 8 worked examples with full input/output
- Works across four mainstream AI coding tools and responds bilingually
- Featured in prominent curated lists including awesome-agent-skills and awesome-claude-skills
- Licensed CC BY-NC-SA 4.0 (non-commercial), and the GitHub license field is NOASSERTION, so commercial teams need legal review before adopting
- Ships only two small maintenance shell scripts and no automated test suite or benchmark evidence for its performance claims
- Output quality depends on the user providing sufficient context; with thin input it falls back to assumption-labeled guesses that may drift from reality
- Coaching mode and other advanced behaviors need explicit trigger phrases; standard mode will not deeply interrogate you
- Not a collaborative web app: no approvals, comments, or sharing workflows for teams expecting SaaS delivery
How do you install this skill?
Claude Code / OpenClaw: run clawhub install product-manager-skills. Codex / Cursor / Windsurf / GitHub-based skill loaders: run npx skills add Digidai/product-manager-skills. Claude Projects: upload SKILL.md, knowledge/, and templates/. Any LLM with local file loading: point the system prompt at SKILL.md and keep the sibling folders intact. Updates are manual: clawhub update product-manager-skills or npx skills update Digidai/product-manager-skills.
How do you use this skill?
After installing, issue natural-language requests. Simple asks (like a single user story) get direct output; say 'coach me' for coaching mode or 'take this from idea to PRD' for the 6-phase Sprint. Example prompts: 'Analyze these metrics: MRR $50k, 500 customers, gross margin 80%, monthly churn 8%, CAC $500.' or 'Review this PRD draft like a strong PM peer.' The skill replies in the user's language and closes every response with decisions, assumptions, next step, and a status.
How does this skill compare with similar options?
The README explicitly contrasts the skill with generic prompting, arguing that plain LLM chats produce plausible PM text and accept bad framing, while this skill enforces framework routing and quality gates. It also positions itself as a local alternative for teams who refuse to send product context to a third-party SaaS.