PE personas: add marketing persona + enrich product-manager#117
Merged
Conversation
Adds a 4th 'personas/' vertical and a product-manager persona that owns the value+viability product risks Hyperstack currently leaves unowned. Tiered gate (hard for net-new, advisory for tweaks, user override). Spec grounded in adversarial deep research (24 claims confirmed, primary sources: Cagan/SVPG, Torres, Intercom, Christensen, Doshi); corpus persisted alongside. Phase 1 = scaffold+plugin+skill+wiring; gaps stubbed NEEDS-RESEARCH.
3-part TDD plan (plugin / vertical infra / skill+binding), code-complete, grounded in the optimizer plugin pattern + verified test breakpoints (plugin count 13->14, bootstrap compile:context regen).
9 tools (four-risks, jtbd, discovery, anti-patterns, strategy, opportunity-vs- solution, validate-job-statement, score-rice, resolve-product-decision). Corpus prose stored as snippets/*.txt via createSnippetLoader (ecosystem pattern, not inline); logic stays in data.ts. Registered as 14th plugin; audit SOURCES entry added (editorial). bun test 59 pass, tsc clean.
Persona vertical: top-level personas/ holds manifests (content), src/personas/ registry.ts loads them (type-checked, reads top-level personas/ same as the context-compiler reads skills/). product-manager manifest binds plugin+skill+ role with tiered gate_policy.
… B2) SKILL.md gains Layer 4 Personas + Persona Registry sections; context-compiler extracts the registry and emits a Personas block, guarded by new required markers (Personas, product-manager). Bootstrap regenerated (58.1% savings).
product-manager persona contracts (PROFILE/LIFECYCLE/CHECKS), pm-gate skill (tiered: hard net-new, advisory tweaks, user override), sources.ts skill link, and hyper -> product-manager persona-gate transitions in router/transitions.
Secondary-pass validation (1 empirical battery + 2 independent red-team agents) found the 3 'logic' tools were theater: resolve_product_decision could never return PASS/BLOCK (hardcoded NEEDS-INPUT), and the regex classifiers were ~80% accurate, gameable, and confidently wrong - violating the ground-truth-not- judgment philosophy the rest of the stack follows. - resolve_product_decision: now takes value+viability assessments as input and returns a real PASS (both addressed) / BLOCK (net-new, missing) / ADVISORY (tweak) verdict. PASS is earned, not automatic. - opportunity_vs_solution + validate_job_statement: drop the regex verdicts; return the Torres / JTBD rubric for the agent to apply (new snippets/rubrics/). - persona.json: drop the dangling 'agent: product-manager' claim (no such agent exists); persona is engaged_by hyper, realized via the pm-gate skill. - LIFECYCLE: remove the unimplemented 'decision trail' overclaim. bun test 63 pass, tsc clean. The 6 framework tools were sound and untouched.
Generalizes the persona vertical from gate-only to gate OR capability. Adds a marketing persona (capability) that does product marketing for any brand: - marketing MCP plugin: 12 ground-truth tools over a 19-snippet corpus (positioning/Dunford, messaging/StoryBrand+value-prop, copy formulas+Schwartz awareness+Cialdini+headlines, brand archetypes+Sharp+category-design, GTM motions+Traction-Bullseye+AARRR+loops, anti-patterns) + a real brief assembler. - skills/marketing: the position->message->copy->voice->GTM workflow + on-demand mode. - persona manifest (mode: capability), contracts, bootstrap Layer 4 generalized. Tools = ground truth only (PM validation lesson baked in - no regex theater). 15th plugin. bun test 72 pass, tsc clean, bootstrap regenerated.
…s (secondary pass) Adds the highest-value learnings missed in the first build: - Voice-of-Customer / message mining (Copyhackers) - the method for getting marketing words straight from real customers (reviews/support/calls). The literal 'marketing words from learnings' engine. - ICP definition (firmographics, triggers, early-vangelist, anti-ICP). - Pricing & packaging (value-based, good-better-best, anchoring, fences, value metric). - Unit economics (CAC/LTV/payback benchmarks) + the single-minded 'only' proposition. 3 new tools (15 total), 3 new snippets + 2 enhanced (22 total). Workflow now leads with VoC research. bun test 73 pass, tsc clean.
…on (web-sourced) From inline web research (GitHub skill repos are coding-focused, thin on marketing): - Hooks: first-3-seconds frameworks (curiosity/open-loop, pain, pattern-interrupt, contrarian, result) + the 4-beat short-form structure + swipe openers. New tool. - Lifecycle/email flows: welcome -> onboarding -> nurture -> retention -> win-back, behavior-triggered (the retention + monetization engine). New tool. - Blue Ocean differentiation: ERRC grid (eliminate/reduce/raise/create) + value innovation, folded into positioning. 17 tools, 25 snippets. bun test 73 pass, tsc clean.
…etrics Find-more sweep (web; GitHub is coding-skill focused, thin on PM frameworks). Closes 3 of the 4 spec-flagged gaps: - Prioritization: MoSCoW, Kano, ICE, WSJF/cost-of-delay, opportunity scoring (gap #2). - MVP & scope-cutting: RAT, vertical slicing, walking skeleton, the cut test (gap #1). - Decision toolkit: Type-1/2 (reversible), 70% rule, pre-mortem, disagree-and-commit, decision records, decision-vs-outcome quality (gap #4). - Plus roadmapping (now/next/later, outcome-based, GIST) and metrics (NSM, OKR, HEART, AARRR). 5 new tools (14 total), 5 new snippets. bun test 76 pass, tsc clean.
…c gap #3) How product sense is actually built (Jules Walter, Julie Zhuo): customer immersion, product teardowns/critiques, principles-first, closing the predict/outcome loop. 15th PM tool. All 4 spec-flagged gaps now closed. bun test pass, tsc clean.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Continues #116 (which squash-merged only the first slice). This lands everything built since: the marketing persona plus the product-manager enrichments.
What's in this PR (delta over main)
personas/vertical): 17 ground-truth tools over a 25-snippet corpus - VoC/message-mining, positioning (Dunford + ICP + Blue Ocean differentiation), messaging (StoryBrand + value-prop + narrative), copy (formulas + hooks + Schwartz awareness + Cialdini + headlines + landing), brand (12 archetypes + Byron Sharp + category design + voice), GTM (motions + 19 channels/Bullseye + AARRR/loops + lifecycle + pricing + launch), anti-patterns - plus abriefassembler and themarketingskill.Verification
Tools are ground-truth only (PM validation lesson applied - no regex theater); corpus stored as
snippets/*.txt. Frameworks cited per-commit (GitHub had no marketing/PM-framework gold; sourced from practitioner web).