[Epic]: Ship AutoGenesis 1.1.1 Skills Preview through one Skill and stdio MCP #25

Goal

Ship AutoGenesis 1.1.1 Skills Preview through one canonical autogenesis Agent Skill and the existing Appium and Windows standard stdio MCP runtimes, so the widest practical set of Coding Agents can execute real tests without client-specific AutoGenesis artifacts.

Product version and artifact identity

The public product release is AutoGenesis 1.1.1 Skills Preview. This product version is intentionally decoupled from component/package versions: it names the coherent user-facing Preview, while each independently distributed technical artifact retains its frozen semantic identity for package resolution, protocol compatibility, and immutable verification. Do not rewrite package coordinates or the GitHub tag to 1.1.1.

User value

A user adopts one Skill, asks directly for natural-language execution or supplies an exact BDD Feature/Scenario, and lets that Skill prepare and route the applicable stdio Runtime. The user does not need to choose between Setup/Run Skills or agent-specific variants. In this safe Preview, an explicit recording request may produce a read-only generated-code preview, but it cannot commit generated files.

Frozen public contract

• One Skill: lowercase canonical autogenesis, distributed as an immutable Agent Contract release asset and discovered through the generic Agent Skills contract.
• One compatibility seam: Agent Skills discovery plus standard stdio MCP. No agent-specific Skill source variants or compatibility adapters unless a real client blocker is separately proven and approved.
• Existing Runtimes: Appium stdio MCP for macOS/iOS/Android targets and Windows stdio MCP for Windows desktop targets.
• Readiness states: Machine Ready, Target Prepared, and Live Ready.
• Execution: direct natural-language tasks and exact BDD Feature/Scenario execution remain available.
• Safe recording: before_gen_code and preview_code_changes remain read-only; live Runtime schemas omit confirm_code_changes; the canonical recording workflow stops after preview and reports that generated-file writes are unavailable in this Preview.
• Trusted writes: [Security] Bind generated writes to trusted approval and directory handles #51 remains the future trusted host/UI approval capability. No chat, task-data, shell, copy/save instruction, alias, or alternate stdio route may bypass it.
• Intended unreleased version line: Agent Contract 2.1.0, Appium Runtime 0.5.0, Windows Runtime 0.5.0, CLI 0.2.0. No target collision existed at refinement time; a public collision requires a new Human Product Owner decision.

Evidence-bound support tiers

Target platforms are macOS, iOS, Android, and Windows only to the extent demonstrated by linked Runtime/client evidence. A result qualifies only the exact client, version, host, target, artifact, and journey exercised.

Scope

• Generate and release the one canonical autogenesis Skill and matching authenticated Agent Contract workflow.
• Keep low-level Runtime wheels independently usable through standard stdio MCP without requiring Skills.
• Preserve direct execution and exact BDD execution.
• Ship recording as a read-only preview with no generated-file commit path reachable through public stdio.
• Provide product-first onboarding that uses final immutable public coordinates and states support tiers honestly.
• Harden the existing release workflow so one reviewed candidate can be published completely, verified exactly, and safely resumed without identity drift.
• Publish the exact candidate as one accepted release unit across PyPI, the immutable GitHub Skill Release, and MCP Registry through the existing main-only release workflow.
• Qualify only environments that pass independent clean final-artifact acceptance.

Non-goals

• No second Setup or Run Skill and no two-Skill release contract.
• No agent-specific Skill variants or speculative client adapters.
• No trusted generated-file writes in this Preview; [Security] Bind generated writes to trusted approval and directory handles #51 owns that future capability.
• No VS Code BDD AI Toolkit or Setup Panel work.
• No Linux or WSL support claim.
• No silent Appium/driver/SDK installation.
• No claim that experimental clients or the Windows Skills journey passed without final-artifact evidence.
• No publication, merge, or protected-branch change during capability implementation or issue refinement.

Capability and release issues

• [Prompt]: Guide installation and environment readiness from user goals #65 — canonical operational autogenesis Skill baseline
• [Capability]: Ship generated-code recording as a read-only safe Preview #69 — remove generated-file commit reachability and ship read-only recording preview
• [Capability]: Make AutoGenesis 1.1.1 Skills Preview publication resumable #71 — make the existing publication workflow immutable, exact-verifying, and safely resumable
• [Security]: Harden generated install prompts and checksum closure #74 — harden Candidate/public install prompts, disposable-environment rollback, sanitized package-source disclosure, and checksum closure
• [Documentation]: Publish AutoGenesis 1.1.1 Skills Preview onboarding #28 / draft PR Publish AutoGenesis 1.1.1 Skills Preview onboarding #29 — root onboarding against final public coordinates and support tiers
• [Release]: Publish AutoGenesis 1.1.1 Skills Preview atomically #70 — atomic AutoGenesis 1.1.1 Skills Preview publication and public-artifact acceptance
• [Compatibility]: Qualify OMP through canonical autogenesis Skill + stdio MCP #21 — OMP compatibility, executed only against this current contract and final artifacts
• [Compatibility]: Qualify Codex through canonical autogenesis Skill + stdio MCP #22 — Codex CLI experimental evidence against the same artifacts
• [Compatibility]: Qualify Copilot CLI through canonical autogenesis Skill + stdio MCP #24 — GitHub Copilot CLI experimental evidence against the same artifacts
• [Compatibility]: Track Pi MCP prerequisite for canonical autogenesis Skill #23 — stock Pi remains unsupported unless a future version gains standard MCP; do not treat Skill-only behavior as AutoGenesis acceptance

Any child issue text that still names autogenesis-setup, autogenesis-run, or two canonical Skills is historical/obsolete and must be refined before execution. This body is the current Epic contract.

Epic acceptance

Scenario: Qualified OMP user executes a direct test through one Skill
  Given a clean native macOS host with OMP 18.18.3
  And the accepted public AutoGenesis 1.1.1 Skills Preview artifacts
  When the user follows the published Agent-first onboarding
  And asks autogenesis to complete a direct real-application task
  Then OMP discovers the single canonical autogenesis Skill through Agent Skills
  And initializes the applicable standard stdio MCP Runtime
  And reaches Live Ready without an agent-specific AutoGenesis artifact
  And completes the requested task through real MCP interaction

Scenario: Exact BDD execution remains available
  Given a supported target is Live Ready
  When the user supplies an exact BDD Feature and Scenario for execution
  Then the canonical autogenesis Skill preserves the exact scenario contract
  And routes each step through the applicable stdio Runtime
  And reports observed execution without requiring generated-file writes

Scenario: Recording is read-only in the Preview
  Given a Coding Agent follows the public canonical Skill
  When the user explicitly requests recording or generated test code
  Then the Runtime may produce a read-only preview
  And the workflow stops after preview
  And reports that generated-file writes are unavailable in this Preview
  And no live schema, dispatch path, alias, documentation, copy/save instruction, or alternate stdio route can commit the generated files

Scenario: Low-level packages remain independently usable
  Given an expert installs an AutoGenesis Runtime wheel directly
  When the expert configures and initializes its standard stdio MCP entrypoint
  Then the Runtime executes its supported live behavior without Skill installation
  And package installation has not modified Agent Skills or client MCP configuration

Scenario: Compatibility claims match evidence
  Given the same canonical Skill and Runtime artifacts are evaluated across clients and hosts
  When support status is published
  Then OMP 18.18.3 on native macOS is the qualified path only after clean final-artifact acceptance
  And Copilot CLI and Codex remain experimental and unqualified
  And stock Pi is unsupported because it has no MCP
  And Windows Skills remains unqualified until independent final-artifact acceptance passes

Scenario: Release surfaces form one accepted unit
  Given #69, #71, #74, and #28 have passed required independent reviews
  When #70 is executed by an authorized release operator from approved merged main
  Then all four exact PyPI versions, immutable Skill Release agent-contract-v2.1.0, and both matching MCP Registry entries are publicly verified
  And no partial publication is promoted as success
  And rollback is applied if any public identity, digest, asset, or Registry surface is missing or mismatched

Required gates and ownership

• [Capability]: Ship generated-code recording as a read-only safe Preview #69 owns Agent Contract generation/workflow assets, Runtime generation tools and focused fixtures/tests, technical package documentation, and manifest/pins.
• [Capability]: Make AutoGenesis 1.1.1 Skills Preview publication resumable #71 owns release-workflow hardening and must serialize workflow ownership with [Capability]: Ship generated-code recording as a read-only safe Preview #69; it cannot edit the same workflow concurrently.
• [Security]: Harden generated install prompts and checksum closure #74 owns canonical Candidate/public Agent Install Prompt hardening and authenticated checksum closure; it blocks [Documentation]: Publish AutoGenesis 1.1.1 Skills Preview onboarding #28, [Release]: Publish AutoGenesis 1.1.1 Skills Preview atomically #70, and this Epic.
• [Documentation]: Publish AutoGenesis 1.1.1 Skills Preview onboarding #28 / PR Publish AutoGenesis 1.1.1 Skills Preview onboarding #29 owns root README.md and onboarding assets.
• [Release]: Publish AutoGenesis 1.1.1 Skills Preview atomically #70 owns release readiness, publication evidence, no-partial enforcement, rollback, and final public-artifact acceptance; publication requires Human Product Owner approval.
• Implementers cannot approve their own work. Independent correctness/security review is required for [Capability]: Ship generated-code recording as a read-only safe Preview #69, [Capability]: Make AutoGenesis 1.1.1 Skills Preview publication resumable #71, [Security]: Harden generated install prompts and checksum closure #74, and release inputs. An independent acceptor must exercise built/public artifacts from a clean environment.
• Release publication remains blocked until [Capability]: Ship generated-code recording as a read-only safe Preview #69, [Capability]: Make AutoGenesis 1.1.1 Skills Preview publication resumable #71, [Security]: Harden generated install prompts and checksum closure #74, and [Documentation]: Publish AutoGenesis 1.1.1 Skills Preview onboarding #28 are accepted, repository immutable releases are enabled, required reviews pass, and [Release]: Publish AutoGenesis 1.1.1 Skills Preview atomically #70's clean public-artifact evidence is complete.

Definition of Done

• The active repository and project contract names one canonical autogenesis Skill and the generic Agent Skills + stdio MCP seam.
• Safe Preview generated-write boundaries pass independent correctness/security review.
• Publication safely resumes the exact reviewed candidate, exact-verifies both Registry versions, and produces only a complete immutable Skill release.
• Public onboarding matches the exact released coordinates and support tiers.
• [Release]: Publish AutoGenesis 1.1.1 Skills Preview atomically #70 verifies every public artifact and digest as one release unit and records rollback readiness.
• OMP 18.18.3/native macOS clean final-artifact acceptance passes.
• Experimental, unsupported, and unqualified rows remain explicitly labeled; they are not silently promoted.
• The Human Product Owner approves the verified user experience and remaining risks.