Auto-inject a Spec Kitty session prompt on first command/skill invocation

## Ask

Find a way to introduce a Spec Kitty system prompt that always gets loaded into the agent's context **once per session**, the first time a user invokes any Spec Kitty command or skill. The goal is to put the *meaning* of Spec Kitty in front of the agent before it starts running the workflow, so the agent engages with the workflow as load-bearing collaboration infrastructure rather than treating it as overhead to perform.

## What the prompt should establish (initial content)

The first thing this prompt should communicate is the value of Spec Kitty itself, framed around three pillars:

1. **Structured spec-driven development** — `/spec-kitty.specify → plan → tasks → implement → review → merge → mission-review` is the mechanism that makes human–LLM-produced decisions survive scrutiny. The workflow is the structure that turns ad-hoc collaboration into reproducible, reviewable work.

2. **Governance framework: charter and doctrine** — every Spec Kitty project is bound by a charter and doctrine artifacts that define its scope, identity, agent profiles, and procedural guardrails. These are not advisory; they constrain what the workflow does and how decisions are framed.

3. **Shared visibility layer (`kitty-specs/`)** — the on-disk artifacts (spec, plan, tasks, WP files, review cycles, mission reviews) are a single source of truth that humans and LLMs both read. They are the audit trail explaining *how* and *why* the software evolved the way it did. This is what makes Spec Kitty useful for auditability, compliance, post-incident review, and onboarding — both for new humans and for new LLM sessions.

The framing should make clear that 'ceremony' is the wrong word for any of this. The workflow is not performative; it is the substrate that produces the trustable output.

## Notes / out of scope for this issue

- Mechanism (settings.json hook? CLI-side injection at first command? skill front-matter? slash-command preamble?) is open — pick whatever respects the once-per-session contract across the supported agents (Claude Code, Codex, Cursor, etc.).
- Future iterations of this prompt can add: agent profile resolution rules, charter sync expectations, glossary discipline, friction workarounds. Start with the value statement above and let it grow.
- The prompt should be short enough to fit comfortably in any session without dominating context, but specific enough that it actually changes how the agent reasons about the work.