Most PMs use AI as a smarter search bar. This is something different — a structured pack of 16 agent skills covering the entire product workflow, from the first customer call to the engineering handoff, built for teams, not just individuals.
Every time a PM opens a new chat, they re-explain the same context: what the product does, who the customers are, what's on the roadmap, what the latest pilot data says. Then they ask for a PRD.
The AI produces something generic. The PM pastes in more context. The output improves slightly. They repeat this every session, with every new task, across every team member.
The real problem isn't the model. It's that there's no shared product intelligence that any skill can read before it does anything.
The architecture is simple: one master file holds your product's intelligence. Every skill reads it before acting. Build it once, share it with the team, and every AI output — PRDs, competitive briefs, pilot plans — starts from the same foundation.
Crawls your wiki, JIRA roadmap, Figma links, and competitor URLs. Produces one scored PROJECT_CONTEXT.md — the shared product brain every other skill reads first.
Reads context first, then structures call transcripts into JIRA tickets that match your team's exact field format.
PRDs and roadmaps that already know your personas, constraints, and strategic goals — no briefing prompt needed.
Pilot plans and leadership decks that reflect your actual roadmap format, risks, and dependency chain.
Organized by where in your week you'd reach for them — not by technical category.
Turns raw call notes or transcript into structured JIRA tickets, themes, and a signal summary — formatted to match your team's exact field patterns.
Builds a structured competitive brief: positioning, gaps, differentiation angles — grounded in your product's actual goals from context.
Processes demo recordings, customer interviews, or conference talks into timestamped summaries with extractable quotes and themes.
Drafts PRDs, roadmaps, and scope docs with full product context pre-loaded. Knows your personas, constraints, and goals without a briefing prompt.
Structures interaction flows, surfaces edge cases, and suggests IA improvements — starting from the product's current design principles.
Builds a polished HTML/CSS prototype from a rough description. Dark-theme first, no frameworks, ready to share with stakeholders in minutes.
The foundation skill. Crawls wiki, JIRA, Figma, and competitor URLs. Run once — every other skill reads this file before acting.
Translates product requirements into technical direction: architecture trade-offs, sequencing recommendations, and risks flagged before the sprint starts.
Generates pilot plans, risk matrices, and dependency maps. Produces HTML status decks formatted for leadership review — risks, owners, next two weeks.
React Native implementation support: component scaffolding, navigation patterns, and platform-specific edge case handling.
Reviews diffs against your team's conventions and the product's technical constraints — not just style, but architectural coherence.
Generates test plans, edge cases, and regression checklists from feature specs — organized by user journey, not just function.
Plans multi-step work as phases, pauses for your approval between each phase, and routes to the right skill. The conductor, not a doer.
Generates internal wiki pages, API docs, runbooks, and onboarding guides — formatted to your team's conventions from context.
Audits code for OWASP top 10 vulnerabilities, flags risky patterns, and suggests hardened alternatives without blocking the build.
Produces a structured security assessment for a feature or system — threat model, attack vectors, risk rating, and mitigation roadmap.
Estimated uplift in output quality for common PM tasks — with full project context loaded vs. starting cold each session. The multiplier comes from the AI knowing your product, not from a better model.
| PM Task | Uplift | What changes |
|---|---|---|
| Writing a PRD | 3×quality | Output already knows your personas, strategic constraints, and JIRA format — skips the briefing paragraph, starts at the substance |
| Competitive analysis | 4×quality | Structured against your actual positioning and roadmap gaps, not a generic feature comparison matrix |
| Weekly roadmap session | ~45 minsaved | Status, risks, and next-two-weeks generated from context — no manual deck assembly for recurring check-ins |
| Exec review prep | ~2 hrssaved | Leadership-formatted HTML decks with pilot status, metrics, and risks, produced from context in one pass |
| Onboarding a new PM | Minutesnot 3 briefing calls | A new team member runs project-context-builder once and has the full product brief, key links, personas, and roadmap format instantly |
Packaging AI behavior as reusable, team-shareable skills requires the same product thinking as any shipped feature. These are the non-obvious decisions.
Most PM AI toolkits treat context as something you paste in before each query. Here it's a versioned artifact — PROJECT_CONTEXT.md — that lives in the repo, gets refreshed as the product evolves, and is read by every skill automatically.
Cursor and Claude Code store skills in different hidden folders. Rather than forcing a choice, the pack includes both layouts. PMs who prefer Cursor and engineers who prefer Claude Code use the same skill set on the same repo — no divergence, no duplicate maintenance.
The project-orchestrator plans work in phases and pauses for your approval between each one. Autonomous multi-step execution in high-stakes PM work (a PRD that auto-generates tickets) has too high a blast radius. The orchestrator is a planner with a human in the loop.
Rather than prescribing a JIRA field format, skills read your team's actual reference tickets from context. Output mirrors your team's exact Epic and Bug structure — even unconventional ones — without a configuration file.
Installation is three terminal commands — documented with Windows and Mac variants, with fallbacks for no-Git environments. A PM who has never touched a dotfile can install from a ZIP, copy two folders, and be running in under 10 minutes.
Personal prompt libraries stay personal. Publishing on a shared org repo makes it a team asset — another PM can use it immediately, contribute a skill back, and the whole team improves together. The value compounds when context and skill improvements are shared, not siloed.
→ Published on AIFoundations