GitHub's Spec Kit hit 1.0 — is its 'product manager' bundle something a PM can actually install?

The one-paragraph answer

The tool is real and works; the "product manager" bundle is a sketch, not a kit. GitHub's Spec Kit (github/spec-kit) — an open-source toolkit that runs spec-driven development as slash commands inside 30+ AI coding agents — reached v1.0.0 around its one-year anniversary (README's own framing: "One year after the first commit, Spec Kit has reached 1.0.0 — not because the work is finished... but because the project has grown into something coherent") and is at v1.0.4 as of 2026-09-02 (VERIFIED, CHANGELOG.md). It ships a real feature called Bundles: Role-Based Setups, meant to provision "a whole team persona (product manager, business analyst, security researcher, developer, …) with one command," and a product-manager bundle manifest genuinely exists at examples/bundles/product-manager/bundle.yml (VERIFIED — read directly). But that manifest names two components: the agent-context extension, which is real, documented, and installable (VERIFIED, extensions/agent-context/README.md), and a product-discovery preset, which returns 404 at every path a preset would live at — no manifest, no README, nothing (VERIFIED via direct probe). The repository's own README says so, one section down from the bundle table: the four example bundles (product manager, business analyst, security researcher, developer) are "bundle packaging examples, not filled generated feature specs." So the honest read is: GitHub has already decided a PM deserves a named slot in its agent-workflow tooling — a real and useful signal about where the industry is heading — but if you tried to specify bundle install product-manager verbatim today, half of it would fail to resolve.

---

Method — what could and could not be read

Primary source only, and unusually complete. This session's egress policy blocks almost every external domain outright (www.pmi.org, owasp.org, genai.owasp.org, arxiv.org, en.wikipedia.org, hbr.org all returned EGRESS_BLOCKED / CONNECT tunnel failed, response 403 on direct test). github.com and raw.githubusercontent.com are reachable, and the entire finding below was read from those two hosts: github/spec-kit at the main branch — README.md, CHANGELOG.md, spec-driven.md, examples/bundles/product-manager/bundle.yml, and extensions/agent-context/README.md — plus a directory-existence probe (curl -o /dev/null -w '%{http_code}') against every path the bundle manifest references, to check which of its components are real files versus dangling names.

This digest is more evenly sourced than most because the one domain this session can reach — GitHub — happened to be the exact domain the question needed.

---

§1 — What actually ships and works: the core loop

VERIFIED, read directly from README.md. Spec Kit's core workflow is six slash commands, run inside the user's AI coding agent of choice (Claude Code, Copilot, Cursor, Codex CLI, Gemini, and 25+ others per the README's "Supported AI Coding Agent Integrations" claim):

1. /speckit.constitution — one-time project principles ("code quality, testing standards, user experience consistency, performance requirements")
2. /speckit.specify — "Focus on the what and why, not the tech stack"
3. /speckit.plan — tech stack and architecture choices
4. /speckit.tasks — breaks the plan into an actionable task list
5. /speckit.implement — executes the tasks
6. /speckit.converge — checks the build against spec/plan/tasks and appends what's still missing, repeated until "Converged"

This is a genuinely PM-shaped division of labor even without any bundle: steps 1–2 are pure requirements-and-intent work (a PM can own them without touching architecture), and the README is explicit that "Rich specification creation" and "Intent-driven development" are core philosophy, not an afterthought.

§2 — The one component of the PM bundle that is real

VERIFIED, extensions/agent-context/README.md. The agent-context extension is opt-in (specify init does not install it by default) and does one job: it keeps a coding agent's context/instruction file — CLAUDE.md, .github/copilot-instructions.md, AGENTS.md, GEMINI.md — in sync inside a delimited managed block, refreshed automatically via after_specify and after_plan hooks, or on demand via speckit.agent-context.update. Concretely: every time a PM runs /speckit.specify or /speckit.plan, this extension can rewrite the relevant section of the project's own AGENTS.md/CLAUDE.md to point at the current plan — the exact pattern this repository already uses by hand (AGENTS.md → self/README.md, docs/decisions/0001, etc.). This is the one piece of the "PM bundle" a person could install and get value from this week, independent of whether the rest of the bundle exists.

§3 — The component that is not there yet

VERIFIED, direct probe. examples/bundles/product-manager/bundle.yml declares:

provides:
  extensions:
    - id: "agent-context"
  presets:
    - id: "product-discovery"
  steps:
    - id: "draft-spec"
    - id: "review-spec"
  workflows:
    - id: "spec-to-roadmap"

Only agent-context resolves to a real, documented component (§2). presets/product-discovery/ returns 404 for both a manifest and a README, at every plausible path (presets/product-discovery/, .specify/presets/templates/product-discovery/). The draft-spec/review-spec steps and the spec-to-roadmap workflow were not independently verified as existing components either — the README's own caveat, one paragraph after introducing the four role bundles, already tells you why: "These are bundle packaging examples, not filled generated feature specs; for end-to-end community examples, see the community walkthroughs." In plain terms: this is GitHub showing the shape a role-based PM kit would take, not shipping one.

§4 — Why this matters more than a changelog nitpick

The interesting fact is not that an example is incomplete — examples are allowed to be incomplete. The interesting fact is which role got a dedicated example at all. GitHub's own reference project treats "product manager" as one of four personas worth naming alongside business analyst, security researcher, and developer, in a tool whose entire premise is that specs — not code — are now the durable artifact AI agents work from. That is a second, independent data point (after the "PM Documentation Rigor Levels" already in frameworks/spec-driven-development) that the industry expects PMs to be direct authors inside the AI-agent loop, not consumers of its output. It also updates that framework's existing "Spec-kit (aspires)" note in the spec-anchored row: as of v1.0.4, Spec Kit is a real, working spec-first tool with an aspirational spec-anchored role layer sitting on top of it.

How a PM applies this — and what to try this week

1. Don't wait for the PM bundle. Install the tool and the one real extension: specify init on a real small feature, then specify extension add agent-context, and run the constitution → specify → plan → tasks → implement → converge loop on something concrete — a client deliverable, a PMPilot feature, a training-offer draft. The value is in living through the loop, not in a bundled starter kit.
2. Read /speckit.specify and /speckit.plan as the PM/architect handoff, made explicit. The tool draws the line exactly where a PM-to-tech-lead handoff already happens; writing it as two separate, reviewable commands is a forcing function for keeping "what and why" out of "how."
3. Treat any vendor's "role bundle" claim as a probe, not a fact. The concrete, repeatable check used in this digest — does the referenced component resolve to a real file? — is copy-pasteable to the next tool that claims a "PM edition" or "PM mode."
4. If evaluating this for a cohort or client demo: say what is real (the core SDD loop, agent-context) and what is a sketch (the PM bundle's discovery/roadmap half), rather than reselling GitHub's own example as a finished PM toolkit.

Course relevance

Feeds the DevOps-literacy module with a live, checkable example of spec-driven development — and feeds the tech-literacy module with a reusable verification habit: read the primary repo, don't repeat the vendor's bundle description. Directly extends frameworks/spec-driven-development's existing but unverified "Spec-kit (aspires)" note with a dated, sourced status check.

Open questions for a future session

• Do any of draft-spec, review-spec, or spec-to-roadmap exist as real steps/workflows elsewhere in the repo, under a name this session didn't guess? Only presets/product-discovery was conclusively absent; the others were not exhaustively searched.
• Has GitHub published a roadmap or issue tracking a real product-discovery preset? github.com/github/spec-kit/issues was not queried (this session's GitHub tool access is scoped to anhvu-pm/anhvu only).
• Whether the still-blocked domains (pmi.org, owasp.org) would change today's "3 options" shortlist if reachable — see the parked candidates below.

Parked candidates (not researched today — network-blocked)

Two of today's three candidate topics could not be researched with primary sources in this session and are logged here rather than written up on guesswork:

• PMI Pulse of the Profession 2026, AI-skills-gap angle. www.pmi.org is blocked, but the full report PDF is already sitting in raw/Pulse of the Profession_2026_PDF.pdf (used for the Aug 11 complexity digest) — the skills-gap and AI-adoption figures were not pulled from it that time and are a same-day-doable follow-up with zero network dependency.
• OWASP Top 10 for Agentic Applications 2026 (ASI01–ASI10), as a PM governance checklist. owasp.org and genai.owasp.org are both blocked; WebSearch summaries of it disagree with each other on the exact list and ordering, which is itself disqualifying for a VERIFIED or even confident SECONDARY writeup. Worth revisiting the moment a primary fetch (or a GitHub-hosted mirror) is reachable.

---

Cross-References

Related: frameworks/spec-driven-development, frameworks/agent-spec-anatomy, concepts/spec-as-agent-contract, concepts/documentation-as-ai-infrastructure, frameworks/ai-dlc, digests/2026Sep/2026Sep03_Amazon_AI_DLC_v1_vs_v2

Sources

Primary (VERIFIED — read directly, github.com / raw.githubusercontent.com)

• github/spec-kit, main branch, README.md — core workflow, Bundles feature, role list, "packaging examples" caveat, 30+ integrations claim
• github/spec-kit, main branch, CHANGELOG.md — current version 1.0.4, dated 2026-09-02
• github/spec-kit, main branch, examples/bundles/product-manager/bundle.yml — full manifest content
• github/spec-kit, main branch, extensions/agent-context/README.md — extension behavior and install command
• Directory-existence probes (HTTP status only, no content) against presets/product-discovery/, .specify/presets/templates/product-discovery/, extensions/agent-context/ — confirms which bundle components resolve to real files

Not reachable this session (recorded, not guessed): www.pmi.org, owasp.org, genai.owasp.org, arxiv.org, en.wikipedia.org, hbr.org, gartner.com, asana.com — all EGRESS_BLOCKED / proxy CONNECT tunnel failed, response 403 on direct test.