pm-guided-learning

The AI-native PM operating system — deep, framework-grounded PM skills with live MCP tool integrations, chained sub-agent workflows, and persistent user memory. Built for solo PMs and founding PMs who need an AI partner that actually knows their product.

Agent-first PM toolkit with 9 specialist agents and 18 skills for solo developers and small teams

12 PM-specific agent skills, 6 workflow commands, 3 automation hooks for Product Managers

18 production-ready Claude Code skills for Product Managers. Discovery, build, measure, communicate.

Execution and product management skills: PRDs, OKRs, roadmaps, sprints, pre-mortems, stakeholder maps, user stories, prioritization frameworks, and more.

68 product management skills (30 phase + 11 foundation + 12 utility + 15 tool) plus 6 sub-agents (pm-critic, pm-skill-auditor, pm-changelog-curator, pm-release-conductor, pm-workflow-orchestrator, pm-skill-router) for AI agents covering the full product lifecycle from discovery through iteration. v2.31.0 is the zero-drift release: generation becomes the only write path for every derived surface that used to drift by hand (the README catalog tables and badges, both quickstarts, the sub-agent compatibility matrix, these three manifest description headlines, and the release-notes mirrors), closing the class of count drift the prior audit kept finding. It stands up release-please in shadow (its Release PR is observed and diffed against the manual cut, never merged this cycle; authoritative cutover is a later-tag decision) behind a PR-title lint, ports the two awk RSTART/RLENGTH hazard validators (check-count-consistency, validate-skill-family-registration) to single-source Node with proven fixture parity before retiring the shell pairs, lifts trigger-fixture coverage from 31 to 43 of 68 skills (about 63%), ships a dry-run-by-default output-eval CI lane plus a published evals page, and expands SECURITY.md with a new provenance page. No new skills; catalog stays 68 / 6 sub-agents. v2.30.0 is a trust-repair release: a 2026-07-04 audit found the count on the first line a new user reads contradicted its own sub-counts, the curated release zip omitted the agents/ and hooks/ directories the dispatch skills need at runtime while still shipping the internal docs tree, and the skill-builder's inventory could not see 22% of the catalog. This release fixes every instance of that drift across the front-door surfaces (README, QUICKSTART, the docs site, these manifests), adds a new enforcing count-phrase gate plus an advisory heading-canon gate so the drift class cannot silently recur, stages agents/ and hooks/ in the release zip while excluding docs/internal, derives the skill-builder's inventory at run time instead of hand-listing it, fixes the phase-router opt-out and a noisy hook regex, backfills four reciprocal When-NOT-to-Use edges, normalizes headings to a documented skeleton canon, rewrites eight early-cohort descriptions, freezes new dual-shell validator pairs, externalizes the trigger-eval roster, and declares the 15-skill tool family stable at 1.0.0. No new skills; catalog stays 68 / 6 sub-agents. v2.29.1 is a maintenance patch: the docs-site generator had built each skill page from a fixed section allow-list and silently dropped the rest, hollowing about 27 of 68 pages, most visibly foundation-build-risk-review; the generator now renders every SKILL.md section verbatim behind a new completeness gate; no skill behavior change, catalog stays 68 skills / 6 sub-agents. v2.29.0 adds one foundation skill, foundation-build-risk-review, a fast pre-build risk gate that names the single assumption most likely to sink the work and returns a verdict (build small, validate first, pivot first, or don't build yet) with a no-code validation step; alongside it, the pm-skill-router sub-agent makes the new-skill collision gate and the trigger router-eval runnable on the subscription with no API key; the catalog grows from 67 to 68 (foundation 10 to 11), sub-agents from 5 to 6. v2.28.0 adds one foundation skill, foundation-stakeholder-briefings, which turns any source artifact (spec, discovery, GTM, metrics, retro) into a canonical master document plus a set of audience-tailored briefings - one per stakeholder lens, each a traceable projection of the master - so a single piece of work reaches engineering, design, data, exec, legal, and more without the versions disagreeing; the catalog grows from 66 to 67. v2.27.1 is a maintenance patch (the classification sub-count drift gate; no new skills, catalog stays 66). v2.27.0 is the provable-quality release: every measured skill carries trigger-eval fixtures with CI gates for routing drift and new-skill collisions, the catalog surfaces (skill-manifest.json + the generated AGENTS.md catalog) are built from frontmatter behind enforcing staleness gates, the output-quality eval harness + asset gate ship, and the creator/validator family bakes the eval contract into skill creation; no new skills (catalog stays 66). v2.26.0 adds the utility-pm-workflow-builder skill (guided authoring from a workflow idea or a promoted chain to a staged Workflow Implementation Packet) and the /chain command (ad-hoc ordered skill chains routed to the pm-workflow-orchestrator Mode B under a written chain-expression contract, with a --thread flag for declared linear dependency); the catalog grows to 66. v2.25.2 is a maintenance patch (a unified validator-inventory manifest with an enforcing CI parity referee that closes the bash/PowerShell/CI drift class, plus the remaining 2026-06-06 Codex audit fixes); no new skills (catalog stays 65). v2.25.1 is a maintenance patch (documentation-site Pattern S reorg, a generated resource index, an em-dash-scar cleanup with new CI guards, dependency bumps, and a pre-tag validator fix); no new skills (catalog stays 65). v2.25.0 adds the plugin's first hooks: opt-in house-rule guardrails (a PreToolUse hook that blocks em-dash and en-dash characters when enabled in .claude/pm-skills.local.md) plus a confident-only SessionStart phase router that suggests the right Triple Diamond skills for the current repo, and an advisory output-quality CI tier (deterministic invariants over the recorded samples); no new skills (catalog stays 65). v2.24.0 adds the pm-workflow-orchestrator sub-agent and its utility-pm-workflow-orchestrator dispatch skill, a governed runner that takes a saved foundation-prioritized-action-plan (or a user-named chain) and runs an ordered sequence of pm-skills with per-step go/no-go checkpoints; foundation-prioritized-action-plan grows a --run handoff (v1.1.0) that offers to run its own runnable prompts through the orchestrator; the catalog grows to 65. v2.23.0 adds one foundation skill, foundation-prioritized-action-plan, which turns any PM input into an evidence-grounded prioritized action plan using Theory of Constraints and Cynefin; the catalog grows from 63 to 64. v2.22.0 removes the 63 redundant per-skill command wrappers (each skill now appears once, invoked directly by name: /pm-skills:<name> on Claude Code, $<name> on Codex) and adds a native Codex .codex-plugin manifest so Codex discovers the skills; all existing skills unchanged. v2.21.0 is an additive distribution launch: pm-skills becomes installable through the new product-on-purpose marketplace (the recommended home for Product on Purpose plugins) while the existing install path keeps working unchanged, so no existing user has to act; no new skills (catalog stays 63). v2.20.0 adds slash commands for the three workshop sprint methodologies (Foundation Sprint, Design Sprint, and the end-to-end arc) plus documentation-count validator hardening; no new skills (catalog stays 63). v2.19.0 hardened the release-validation tooling so the library polices its own counts, cross-references, and links. v2.18.0 added four highest-consensus content skills: discover-market-sizing (TAM/SAM/SOM multi-framework triangulation), define-prioritization-framework (RICE, ICE, MoSCoW, Weighted Scoring, and Kano run in parallel with a cross-framework comparison), discover-journey-map (stages, touchpoints, emotional curve, and moments of truth), and measure-survey-analysis (honest survey analysis with limitation warnings), each with three thread-aligned samples. It carries forward v2.17.0 native Claude Code sub-agent registration: the sub-agent definitions moved to the canonical agents/ directory (the coordination directory was renamed to _agent-context/ to free the name), so all 4 sub-agents now auto-discover via @-mention on Claude Code; the dispatch skills continue to provide the same capability on Codex CLI, Cursor, Windsurf, Copilot, and Gemini CLI. v2.17.0 also migrates skill frontmatter to the metadata-nested structure per the agentskills.io specification and makes the CI validators bash-3.2 portable. Doc-stack on Astro 6.3.x + Starlight 0.39.x. Carries forward v2.15.0 Sprint Skills (Foundation Sprint family + Design Sprint family + tool-note-and-vote under classification:tool), v2.12.0 OKR Skills set, v2.11.0 Meeting Skills Family, lean canvas, persona, structured templates, real-world examples, 12 workflows including the foundation-to-design end-to-end arc, an interactive skill builder, and lifecycle tools for validating and iterating skills. Follows the agentskills.io specification.