A product-management toolkit for protoAgent: the tools, skills, subagents, and a console view that an AI agent — or a human PM working through one — needs to research and manage a product end to end, from discovery to strategy, execution, launch, growth, and shipping.
It bundles four things into one plugin:
- 65 PM skills — discovery, strategy, execution, research, go-to-market, marketing/growth, analytics, AI-shipping, and PM utilities. Loaded by description, so the agent reaches for the right framework automatically.
- The PM Brain — a markdown-native, provenance-enforced knowledge base (decisions, hypotheses, stakeholders, knowledge areas, with an ingestion/source audit split), managed by tools so the agent's product memory persists and stays auditable.
- Five PM specialist subagents — a brain operator plus discovery, strategy, execution, and analytics specialists the lead agent can delegate to.
- A PM Brain dashboard — a console view surfacing decision debt, active hypotheses, stale stakeholders, and recent ingestion.
Adapted from Paweł Huryn's excellent open-source work — pm-skills and pm-brain (MIT). This is an adaptation, not a verbatim copy: nine separate Claude Code plugins are unified into one, redundant skill variants are consolidated, and the PM Brain is re-implemented as protoAgent tools + a view. See NOTICE.
In the protoAgent console: Plugins → Download → install from a git URL
(https://github.com/protoLabsAI/pm-plugin), then enable it. It ships disabled — enabling
is your trust decision (ADR 0027). Needs a host serving the DS plugin-kit at /_ds/ (≥ v0.34.0).
Every skill is a SKILL.md loaded by description — the agent picks the right one for the task.
| Area | Skills |
|---|---|
| Discovery (10) | brainstorm-ideas · identify-assumptions · design-experiments · opportunity-solution-tree · prioritize-assumptions · prioritize-features · interview-script · summarize-interview · analyze-feature-requests · metrics-dashboard |
| Strategy (12) | product-vision · product-strategy · value-proposition · lean-canvas · business-model · startup-canvas · swot-analysis · pestle-analysis · porters-five-forces · ansoff-matrix · monetization-strategy · pricing-strategy |
| Execution (16) | create-prd · brainstorm-okrs · outcome-roadmap · sprint-plan · user-stories · job-stories · prioritization-frameworks · pre-mortem · stakeholder-map · strategy-red-team · retro · release-notes · test-scenarios · summarize-meeting · wwas · dummy-dataset |
| Market research (7) | competitor-analysis · customer-journey-map · market-segments · market-sizing · sentiment-analysis · user-personas · user-segmentation |
| Go-to-market (6) | gtm-strategy · gtm-motions · growth-loops · beachhead-segment · ideal-customer-profile · competitive-battlecard |
| Marketing & growth (5) | north-star-metric · positioning-ideas · value-prop-statements · marketing-ideas · product-name |
| Data & analytics (3) | sql-queries · cohort-analysis · ab-test-analysis |
| AI shipping (2) | intended-vs-implemented · shipping-artifacts |
| PM toolkit (4) | review-resume · draft-nda · privacy-policy · grammar-check |
Consolidated from phuryn's 68: the existing/new variants of idea-brainstorming, assumption-identification, and experiment-design are merged into single mode-aware skills.
A second brain for product work — markdown files the agent reads before a task and writes after, so product memory persists and stays greppable, diffable, and human-editable.
Areas: decisions/ · hypotheses/ · stakeholders/ · knowledge/ (strategy, product,
users, market) · ingestion/ (synthesized records) · source/ (verbatim audit anchors) ·
rules/ · maintenance/.
Provenance is enforced, not suggested. Every evidence row on a decision or hypothesis must
carry one tag — a markdown link to ../ingestion/… or ../source/…, or one of
(stakeholder-verbal, …) / (intuition, PM, …) / (industry-knowledge) / (chat, no artifact).
pm_log_decision rejects an untagged evidence row, so a decision always wears how much of its
reasoning is collected-and-fresh vs inherited-and-stale on its face. (Upstream pm-brain enforced
this with an editor hook; here it's baked into the tool.)
Tools:
| Tool | What |
|---|---|
pm_brain_init |
Scaffold the brain (idempotent). |
pm_brain_status |
Maintenance sweep: decision debt, active hypotheses, stale stakeholders, recent ingestion. |
pm_log_decision |
Log a decision record (provenance-validated evidence; observable reverse-condition). |
pm_upsert_hypothesis |
Create/replace a feature's hypotheses (orphan-evidence warnings). |
pm_upsert_stakeholder / pm_touch_stakeholder |
Map a stakeholder; log a touchpoint (updates last-touched). |
pm_ingest |
Write a verbatim source + a synthesized ingestion record; returns a citation tag. |
pm_note |
Append to a knowledge area. |
pm_list / pm_get / pm_search |
Read / browse / full-text search (traversal-guarded). |
The brain lives at brain_dir (Settings ▸ Plugins) or PM_BRAIN_DIR; empty → ~/.protoagent/pm-brain
(instance-scoped). Point it at a repo path to keep the brain git-versioned.
Delegatable PM specialists (task them). The 65 skills are available to all of them; each adds a
focused persona + tool allowlist.
pm_brain— the second-brain operator: load context before a task, write back after; ingest, prep a 1:1, log decisions, run a maintenance sweep. Retrieves before asking.pm_discovery— discovery & user research: ideation, assumptions, experiments, interviews, personas, segmentation, sentiment, competitive/market research, prioritization.pm_strategy— strategy & go-to-market: vision, canvases, SWOT/PESTLE/Porter, pricing, positioning, GTM, growth.pm_execution— delivery: PRDs, OKRs, roadmaps, sprints, stories, pre-mortems, stakeholder maps, retros.pm_analytics— product data: SQL, cohorts, A/B tests, metric definition.
A right-rail PM Brain view, themed by the DS plugin-kit so it follows the console's live theme. Two surfaces in one panel:
- At-a-glance cards — decision debt (pending decisions), hypotheses by status, stale or never-touched stakeholders, recent ingestion, and area counts.
- Full file browser + editor — an All files card listing every doc the brain produced,
grouped by area (not just the curated slices above). Click any file to read it; Edit opens an
inline editor and Save writes it back; New file creates one (pick an area + name). So a PM
can read and edit all brain docs without leaving the console. Two guardrails carry over from the
tools:
source/files are verbatim audit anchors and stay read-only, and saving a decision or hypothesis with an untagged evidence row warns (the human PM has final say — it still saves).
Data flows through gated /api/plugins/pm routes (/files, GET/PUT /file), all traversal-guarded
to the brain root.
| Setting (Settings ▸ Plugins) | Env override | Default | What |
|---|---|---|---|
| PM Brain directory | PM_BRAIN_DIR |
~/.protoagent/pm-brain |
Where the brain markdown lives. Set a repo path to git-version it. |
| Stakeholder staleness (days) | PM_STALE_STAKEHOLDER_DAYS |
21 |
A stakeholder not touched within this window is flagged stale. |
pip install -r requirements-dev.txt
pytest -q
ruff check . && ruff format --check .The suite is host-free (no protoAgent install) — host-only imports stay lazy. CI runs the same on every PR.
Built for protoAgent. Skills & brain model © Paweł Huryn (MIT); adaptation © protoLabs AI (MIT). See LICENSE and NOTICE.