[Skill] Add guided Agenova feature delivery workflow

[wunderforge]

Purpose

Create a cross-agent delivery skill that applies the useful parts of feature-dev to Agenova's existing ticket and task-packet workflow.

This first exercise must make two outcomes visible:

• learn the feature-dev workflow even when the implementation agent is Codex rather than Claude Code;
• observe how AIDLC changes delivery: what was explored and discussed, which decisions changed, how they affected implementation, and what each quality gate proved or blocked.

Boundaries

• The GitHub issue and approved task/spec remain the source of truth; the skill must not create a competing requirements system.
• Keep the workflow vendor-neutral and usable by Codex or Claude Code.
• Use explicit human checkpoints in guided mode; provide a lower-narration delivery mode for normal work.
• Do not include Agenova secrets, private conversations, or project-specific implementation content.

Acceptance criteria

• The skill supports discovery, clarification, design, implementation, review, and verification.
• Guided mode produces a concise decision-and-evidence trail linking discussion to implementation and gates.
• Review uses independent correctness, architecture-boundary, and evidence lenses.
• The skill is discoverable by the skills CLI and includes a safe invocation example.
• Local repository checks pass and SkillSpector reports SAFE with complete coverage.

[wunderforge]

AIDLC exercise checkpoint:

• Starting question: how to learn feature-dev without binding the team to Claude Code or replacing Agenova's ticket/task workflow.
• Decision: keep the Issue and approved task/spec authoritative; add a vendor-neutral skill with guided and delivery modes.
• Design impact: guided mode exposes discovery, discussion, decisions, implementation consequences, review changes, and gate evidence; delivery mode keeps the same controls with less narration.
• Implementation impact: added one skill entry point plus focused references for the guided workflow, review rubric, and optional learning log.
• Gate evidence: local repository checks and validator tests passed; PR CI passed secret scanning, structure checks, tests, and the fail-closed complete SkillSpector scan.

PR: #2