deanpeters__product-manager-skills/17JUL26.md

HANDOFF: v0.83 shipped (July 17, 2026) — Market Intelligence Suite

You are working in the Product-Manager-Skills repository. This document is a retrospective + forward-looking handoff: v0.83 was cut and published today, and this file records what shipped, the conventions it established, and the queue of what's next. Read it alongside CLAUDE.md (governance + distillation protocol) and 08JUL26.md / 04JUL26.md (prior handoffs — the release-cut mechanics and skill-authoring recipe there still apply and are not repeated in full here).

If you are a different agent picking this up cold: everything you need is in this file plus CLAUDE.md. You do not need the conversation that produced it.

---

Where Things Stand

v0.83 is the current shipped release (July 17, 2026, tag v0.83 at 0e2296d, Build Release Artifacts green on main and tag, GitHub release published with 80 assets, dist/ shelf regenerated at 70 ZIPs).

The library is 70 skills (24 Component / 27 Interactive / 19 Workflow) + 6 commands. main is clean and in sync with the tag. No deferred count bumps this time — everything is consistent.

The commits of July 16-17 (this release's story)

Commit	What
5ebfee4	README License section clarified + Substack link fix (see "Licensing" below)
4c2277e	PR #13 merged (typo fix, community)
0ad6f9c	Market Intelligence Suite — 13 new skills + 2 upgrades (waves 1-3)
5ab8ecd	intel-discipline-advisor (14th suite skill)
0e2296d	v0.83 release cut: version bumps, release note, dist/ rebuild

Release note: docs/announcements/2026-07-17-v0-83-market-intelligence-suite.md.

---

What the Market Intelligence Suite Is (architecture, not just a list)

Fourteen new skills under theme: market-intelligence (also a new marketplace category of the same name), plus two upgraded skills. The structure matters more than the roster:

• autonomous-investigation (Workflow) is the suite's protocol skill — the counterpart to workshop-facilitation, but for skills where the world holds the context instead of the user. Its seven-clause contract: question budget (proceed on labeled assumptions when spent/silent), search-plan gate, Fact / Inference / Assumption evidence labels (the library's canonical vocabulary — Dean chose this over Confirmed/Inferred/Projected/Unknown), do-not-invent lists, Just Enough Mode, stable diffable output schemas, exactly-4-option Final Step. Plus: the Confidence Stacking Rule (1 channel = watch item, 2 = hypothesis, 3+ = actionable, conflict = dig), "announcements are intent until funding/hiring corroborate," and SCIP guardrails.
• intelligence-collection-disciplines (Component) is the backbone: eight intel disciplines (OSINT, FININT, GEOINT/DEMOINT, TECHINT, HUMINT, SIGINT, MASINT, All-Source Fusion) with free/paid source tables, signal→inference chains, the bottom-up TAM/SAM/SOM recipe, fusion cadence, and the artifact-mapping table. Do not split it into eight skills — fusion is the point.
• intel-discipline-advisor (Interactive) is the backbone's Component + Interactive pairing (Dean's preferred pedagogic model): 3 adaptive questions → discipline mix + cadence + executing skill, teaching the mapping-table logic as it routes, with an honest off-ramp to discovery/win-loss.
• The chain (all Workflow, each consumes the prior's stable schema): market-landscape-scan → competitive-research-snapshot → competitive-intel-watch → battle-card-builder.
• Strategy frameworks, evidence-cited: swot-analysis, porters-five-forces, ansoff-matrix. Ansoff is deliberately cross-referenced with organic-growth-advisor as its investigation sibling (same Ansoff-shaped map; coaching there, evidence here) — do not "deduplicate" them.
• Monitors/miners: voice-of-customer-miner, pricing-packaging-tracker, pestel-delta-monitor (pairs with pestel-analysis; hard prerequisite on a baseline).
• Umbrella: competitive-analysis-process — six-step orchestration (from practitioner field material), delegating to the suite the way discovery-process delegates to discovery skills.
• Upgrades: tam-sam-som-calculator gained three entry modes (own numbers / guided interview / autonomous research with the GEOINT recipe + FININT capture rates); company-intel gained the Executive Signal Refresh rerun pattern (Then/Now diffs, exact-quote rule, Dropped Language section), source priority ladder, do-not-sanitize rule, product-org heuristics (CPO tenure, PM job postings as culture documents), and an optional distress read.

Conventions the suite established (respect these in future skills)

1. Investigation skills declare the protocol — References names autonomous-investigation as governing; body states question budget, do-not-invent list, output schema ("do not reorder"), Final Step block.
2. Free/paid two-column source tables are the sanctioned house style for tool mentions (neutral, always a free path). This is NOT the same as vendor backlinks in templates, which we reject (see PR #15 below).
3. Fact / Inference / Assumption everywhere. Unknowns are a gaps list, not a fourth label.
4. Output schemas are stable contracts — diffability enables the monitors; "improving" a schema between runs is a regression.

Sources (all Dean's IP; see licensing + anonymization below)

~/Code/product-manager-prompts/market-intelligence/ (12 prompts), (local, not in repo), ~/Downloads/competitive-research-on-steroids-compendium.16jul26.md (the backbone), ~/Downloads/m365-agent.company-research-help.md + m365-agent.tam-sam-som.md (Productside-internal — techniques generalized, internals excluded).

---

Non-Suite Work Shipped July 16 (context you'll need)

Licensing clarity (5ebfee4)

Critics called the repo a licensing "mixed bag" because 39 skills say "Adapted from prompts/... in product-manager-prompts" (an MIT repo). That claim is wrong: Dean authored both repos; a license binds licensees, not the copyright holder. Resolution was README-only: the ## License section now has a bolded "everything is CC BY-NC-SA 4.0, no mix" statement, the same-author explanation, and a plain-English do/don't block (✅ use at work / adapt / teach; ❌ don't sell without expressed written permission; 🤔 unsure → open an issue). Do NOT rewrite the 39 "Adapted from" Provenance lines — they're intentional lineage. Dean plans to relicense product-manager-prompts to CC BY-NC-SA 4.0 as well. Product Porch podcast links (4 career skills) are intentional shout-outs to friends — keep them. MITRE usage is cleared (Dean has been in touch).

Customer anonymization protocol (CRITICAL, permanent)

No Productside customer names — and no example scenario keeping the recognizable shape of a real engagement — may appear in the repo. Examples use bracket placeholders ([TARGET], [Competitor A]) or clearly fictional companies in rotated domains. company-intel's examples were fictionalized on Jul 16 (Parker Hannifin/Emerson/Honeywell/Novo Nordisk → Helix Motion Systems, Northfield Automation, Corvid Industrial, Brightwater Biologics). Real names only for canonical public framework cases (e.g., Amazon for PR/FAQ). Before committing new content, sweep new files (grep -E '\b[A-Z][a-zA-Z]+ (Inc|Corp|GmbH|AG|Ltd|LLC|Group)\b' plus any names Dean supplies in-session). Never commit a customer denylist to the repo — the list itself would be the leak. Podcast transcripts also stay out of the repo (CLAUDE.md's Phase 8 pointer to research/product_porch_transcript.*.md refers to files deliberately not committed — that's correct, not a bug).

PR triage (July 16)

• #13 (typo) — merged.
• #15 (kriptoburak, "X source log") — closed as vendor link spam (Xquik backlink; the PR body's own validation list gave it away). Policy: no vendor/product links in skills/templates; the source-log idea is fine vendor-neutral if ever wanted.
• #17 + issue #16 (director-readiness Step 0 trim) — closed; premise wrong. Context-dump and Best-guess modes ARE implemented via workshop-facilitation (line 62 delegation; v0.4 single-source-of-truth design). This is the exact "efficiency edit strips facilitation scaffolding" regression class. Legitimate kernel noted in the closing comment: a one-line "(modes defined in workshop-facilitation)" pointer in Step 0 menus would help discoverability — small open item, never done.
• #20 (dNzZiz, OpenCode compatibility) — open, awaiting contributor. We verified OpenCode's docs: it reads .claude/skills/ and .agents/skills/ natively, needs only name+description, ignores unknown frontmatter; its metadata field is a string-to-string map (the PR's nested lists wouldn't even validate). The PR's frontmatter restructure would break app/main.py and find-a-skill.sh (they read type: top-level). Comment posted asking for a docs-only slim-down (docs/INSTALL-OPENCODE.md + README/START_HERE mentions). If no response by ~end of July: write the docs-only version ourselves, credit @dNzZiz in commit + changelog. Note our existing dist/packages/pm-skills-codex.zip ships .agents/skills/ and is already OpenCode-compatible — say so in the docs.

Repo settings

main now has branch protection: force pushes blocked, deletion blocked, nothing else (no required PRs/status checks — Dean pushes directly to main; don't add those). Set via API July 16.

---

Gotchas (new ones from this release; 08JUL26's all still apply)

• check-skill-triggers.py wants the literal string "Use when" in description — "Use before/on/to/for" all warn. Five suite descriptions were rewritten for this; write new descriptions with "Use when" from the start. (1 known pre-existing warning: incoming-request-advisor.)
• README ASCII banner width is display-columns, not bytes — • and ║ are multibyte. When editing the version line, verify with display-width math or line up against the skills-count line, not wc -c.
• The marketplace metadata.description count ("70 battle-tested…") moves with the README's three count surfaces — four prose surfaces total plus CLAUDE.md.
• dist/ is committed — after any release-worthy skill change, bash scripts/build-dist.sh regenerates it; the ZIPs land in the release commit. (Still not wired into CI — open item below.)
• 3 pre-existing smoke warnings (interactive option counts) remain known/unrelated.

---

What's Next (priority order)

1. PR #20 follow-through (see above) — the only open PR. Docs-only OpenCode support, by contributor or by us with credit.
2. Theme backfill — 36 pre-suite skills still lack theme frontmatter and land in the Streamlit "All other skills" expander. All 14 suite skills are tagged; the gap is the older library. (Open since 04JUL26.)
3. Regional Source Overlays (EU/MENA companion to the compendium) — referenced in intelligence-collection-disciplines as a planned future Component. Dean has the companion document in his prompt-library world; ask him for it before building. APAC/LATAM/Sub-Saharan Africa are named build-out candidates.
4. Phase 6 remainder (AI PM Orchestrator): ai-product-evals, ai-observability-framework, ai-maintenance-planning (Components) + ai-product-orchestrator (Workflow). Long-standing.
5. Small opens: the Step 0 "(modes defined in workshop-facilitation)" pointer across interactive skills; "Invoking skills with arguments" subsection in docs/Using PM Skills with Claude.md; wire build-dist.sh into CI/release automation; decide fate of the v0.80 AI Product Builder Track brief (15MAY26.md, never executed).
6. Possible Phase 10: Pricing & Monetization Suite (7 skills, listed in CLAUDE.md "What's Next") — note pricing-packaging-tracker and finance-based-pricing-advisor now cover the tracking/decision corners; scope against them to avoid overlap.
7. product-manager-prompts relicensing (MIT → CC BY-NC-SA 4.0) happens in that repo on Dean's timeline; nothing needed here — the README licensing note stays accurate either way.

---

TL;DR for whoever picks this up

1. v0.83 is out: 70 skills, Market Intelligence Suite complete (protocol + backbone + advisor + chain + frameworks + monitors + umbrella + 2 upgrades). All surfaces consistent; CI green.
2. Investigation skills follow the autonomous-investigation contract; F/I/A labels; stable schemas; free/paid source tables. Don't split the compendium; don't merge Ansoff into organic-growth-advisor.
3. Licensing is settled (same-author; README explains it) and customer anonymization is a hard rule — fictional surrogates only, sweep before committing, no denylist in the repo.
4. Only open PR is #20 (OpenCode) — docs-only slim-down requested; build it ourselves with credit if the contributor goes quiet.
5. Dean cuts releases explicitly ("commit and push and tag"). Don't tag on your own initiative.