Your AI co-pilots for p roduct management — translate updates for a ny audience, make data-backed decisions, buil d financial models, and sequence them into a credible roadmap, in seconds, not hours.
[ ](https://claude.ai /code) [](https://github.com/features/ copilot)
Four purpose-built agents that hand le the grunt work so you can focus on strateg y: a Stakeholder Translator, a Decision Engine, a Financial Analyst, and a R oadmap Architect — each designed to compr ess hours of work into seconds, with consiste nt structure and reusable templates.
� �️ Disclaimer: All data in this project i s entirely synthetic and mock-generated for d emonstration purposes. Customer names, compan y names, financial figures, market data, and all agent outputs are fictional. No real cust omer data, proprietary information, or actual business metrics were used.
Landing pa ge — 3 specialized agents with one-click la unch
Decision Engine — multi-framework strategic analysis with s coring methodology
Fina ncial Analyst — assumption sourcing, key me trics, and sensitivity analysis
St akeholder Translator — sensitivity classifi cation with 5 audience-tailored outputs
Product manage rs spend a disproportionate amount of time on communication translation, decision structur ing, and financial justification — not on t he strategic thinking itself. A single produc t update requires five different versions, on e per audience. A prioritization decision req uires manually applying three or four framewo rks. A feature business case requires hours o f spreadsheet modeling.
This project demonst rates how purpose-built AI agents — eac h with a distinct cognitive role — can comp ress these workflows:
One product update → 5 audience-tailored communications (en gineering, exec, board, customer, sales)
One strategic question → 4-framework analysis with synthesized recommendation
One feature description → Full financial model with s ensitivity analysis and ship/no-ship decision
These four agents are designed to chain — each one consumes the prior agents outpu t and adds its own cognitive layer. Each agen t doesn't just generate text — it applies s tructured reasoning: sensitivity classificati on, multi-framework scoring, crossover analys is, pre-mortem scenarios, and audience-specif ic framing tuned to the actual decision being made.
Cognitive function: Audi ence Adaptation
Takes a single product upd ate and produces five tailored communications — each with the right tone, detail level, technical depth, and framing for its audience . Includes sensitivity classification (Safe / Caution / Internal Only) so PMs know what's shareable and what's not.
| Output | Audienc e | Framing |
|---|---|---|
| Engineering Update | Dev team | Technical decisions, code references, debt trade-offs |
| Executive Summary | Leadership | Business impact, metrics, decisions needed |
| Board N arrative | Board of Directors | Strategic pos itioning, speaker notes |
| Customer Changelo g | End users | Benefits-focused, no internal details |
| Sales Enablement | Sales team | Objection handling, competitive positioning, talk tracks |
Demo scenarios: AI feature launch, missed deadline communication, compe titive response strategy.
→ [View Skill Definition](skills/stakeholder-translator/SKI LL.md)
Cogn itive function: Strategic Reasoning
Applie s four distinct analytical frameworks to a pr oduct decision, then synthesizes them into a single prioritized recommendation with confid ence scoring and a pre-mortem analysis.
| Fr amework | What It Evaluates | |-----------|-- -----------------| | Impact × Confidence Mat rix | Revenue impact weighted by execution ce rtainty | | Strategic Alignment | Weighted sc oring against company goals | | Second-Order Effects | Downstream consequences (positive a nd negative) for each option | | Pre-Mortem A nalysis | "It's December and this failed — what went wrong?" with probability estimates |
Demo scenarios: Quarterly prioritizati on (3 competing initiatives), ship/iterate/su nset decision for an underperforming feature.
→ [View Skill Definition](skills/decisi on-engine/SKILL.md)
Cognitive function: Quantitative M odeling
Builds rigorous financial models f rom natural language inputs. Fills gaps with SaaS benchmarks, runs sensitivity analysis ac ross multiple variables, and produces ship/no -ship/de-risk recommendations.
| Output | De scription |
|---|---|
| Assump tions Table | Every input labeled by source ( PM Input, SaaS Benchmark, Estimated) with edi t affordance |
| Key Metrics Dashboard | Visu al metric cards with color-coded status |
| F ull Model | Unit economics, revenue projectio ns, NPV, payback period |
| Sensitivity Matri x | Multi-variable sensitivity showing break- even boundaries |
| Decision Framework | Ship / Do Not Ship / De-risk with specific condit ions for each |
Demo scenarios: Feature ROI analysis, TAM/SAM/SOM market sizing, pric ing change impact modeling.
→ [View Skil l Definition](skills/financial-analyst/SKILL. md)
**Cogni tive function: Sequencing & Capacity Planning **
Takes a set of prioritized initiatives an d turns them into a credible quarterly plan: themes, dependency graph, capacity check, and an explicit commit-vs-stretch split with con fidence bands. Designed to run after the De cision Engine and before the Stakeholder Tr anslator — it answers the middle question: in what order, against what capacity, with wh at trade-offs.
| Output | Description | |--- -----|-------------| | Theme Map | 3–5 stra tegic themes with a one-sentence "why now" fo r each | | Sequenced Plan | Initiatives place d into Now / Next / Later with rationale | | Dependency Graph | Hard, soft, and reverse de pendencies with risk callouts | | Capacity Ch eck | Effort vs. team capacity with explicit overcommit / slack flags | | Commit vs. Stret ch | Confidence bands on committed scope; str etch goals named, not snuck in | | Narrative One-Pager | Stakeholder-ready summary: bets, trade-offs, what we're not doing |
Demo scenarios: Quarterly planning across 10+ in itiatives, mid-quarter resequencing after sli ps, board-readout roadmap narrative.
→ [ View Skill Definition](skills/roadmap-archite ct/SKILL.md)
The four agents are designed to chain into a single planning loop:
Financial Analy st → Decision Engine → Roadmap Ar chitect → Stakeholder Translator
(is it worth (which of these (in what order, (tell each audience
b uilding?) do we do?) a gainst capacity?) in their language)
You can use any agent standalone, but feedin g the output of one into the next is where th e compounding leverage shows up — a financi al model becomes a decision becomes a sequenc ed plan becomes five tailored communications, with traceability at every hop.
Each agent is available as a standalone Claude Code skill:
# Clone and copy all skills
git clo ne https://github.com/varunk130/pm-copilots.g it
cp -r pm-copilots/skills/* ~/.claude/skill s/
# Or copy individual skills
cp -r pm-copi lots/skills/stakeholder-translator ~/.claude/ skills/
cp -r pm-copilots/skills/decision-eng ine ~/.claude/skills/
cp -r pm-copilots/skill s/financial-analyst ~/.claude/skills/
cp -r p m-copilots/skills/roadmap-architect ~/.claude /skills/Restart Claude Code after copyi ng skills. Use via slash commands or natural language prompts.
- Structure d reasoning over free-form generation: Each agent applies named analytical frameworks (I mpact × Confidence, Pre-Mortem, Crossover An alysis) rather than open-ended generation
-
- Sensitivity classification as a first-class concept*: The Stakeholder Translator classif ies information by sensitivity level before generating — knowing what not to say to wh ich audience
- *Multi-framework convergence *: The Decision Engine applies four independe nt frameworks and checks whether they converg e — divergence surfaces genuine uncertainty
- Explicit assumption sourcing: The Fina ncial Analyst labels every model input with i ts source (user-provided, benchmark, estimate d) — making it visible where confidence is high vs. interpolating
- Audience-aware gen eration at scale: Five distinct communicati ons from one input, each correct for its audi ence
- Capacity as a hard constraint, not a vibe: The Roadmap Architect converts T-shi rt sizes into person-weeks, compares to actua l team capacity, and refuses to silently over commit — overflow shows up as explicit cuts or stretch goals, never as quietly inflated scope
pm-copil ots/
├── README.md
├── screenshot s/
│ ├── ai-pm-dashboard.png
│ ├── decision-engine.png
│ ├── financial-analyst.png
│ └── stakeh older-translator.png
└── skills/
� �── stakeholder-translator/
│ └ ── SKILL.md
├── decision-engine /
│ └── SKILL.md
├── financial-analyst/
│ └── SKILL. md
└── roadmap-architect/
� ��── SKILL.md
MIT — se e LICENSE for details.
Built by Varun Kulka rni
Powered by Claude Co de Opus 4.7 + GitHub Copilot
The copilots/ package wraps the three PM copilots in plain Python so they can be invoked from a CLI, embedded in a script, or wired into another agent harness.
Requires Python 3.10+.
# Run the end-to-end demo (exercises all three copilots)
python examples/run_demo.py
# Or use the CLI dispatcher
python main.py comms --audience "Exec" --objective "Q3 launch" --bullets "NPS up|burn down"
python main.py decide --question "Build vs buy?" --options "Build,Buy,Partner"
python main.py model --name "Pro" --users 1000 --arpu 29 --margin 0.75 --cac 100Zero runtime dependencies — pure Python standard library.