/ork:dev — bringing it up to skill conventions

The Lane B initial PR shipped a SKILL.md describing a 9-step boot but no executable scripts. Every other user-invocable skill ships scripts/ + rules/ + test-cases.json. This PR closes the gap.

Compliance audit (before vs after)

Required	expect	cover	implement	commit	dev (before)	dev (after)
SKILL.md	✓	✓	✓	✓	✓	✓
references/	✓	✓	✓	✓	✓	✓
scripts/	✓	—	✓	✓	✗	✓ (3)
rules/	✓	—	✓	✓	✗	✓ (6)
test-cases.json	✓	✓	✓	✓	✗	✓ (10 cases)

What changed in src/skills/dev/

Before this PR

SKILL.md only describes the contract
├── SKILL.md
└── references/
├── boot-sequence.md
└── state-schema.md

No scripts
No rules
No test cases

After this PR

Full skill structure
├── SKILL.md
├── test-cases.json (10 cases)
├── references/
│   ├── boot-sequence.md
│   └── state-schema.md
├── scripts/
│   ├── boot.sh (executable, idempotent)
│   ├── stop.sh (SIGTERM reverse, SIGKILL fallback)
│   └── status.sh (--quiet for liveness only)
└── rules/
    ├── _sections.md, _template.md
    ├── lab-stack-prerequisites.md (CRITICAL)
    ├── branch-named-subdomain.md (HIGH)
    ├── idempotent-boot.md (HIGH)
    └── teardown-order.md (MEDIUM)

scripts/status.sh smoke test (run locally during this PR)

$ bash src/skills/dev/scripts/status.sh
ork:dev — no state file. Run /ork:dev to boot.
(exit 1 — expected, no state)

$ bash src/skills/dev/scripts/status.sh --quiet
(exit 1, no output — used by boot.sh idempotency)

3 stale doc strings also fixed (drift caught while auditing)

docs/getting-started/installation.mdx:36
- Skills:     106 loaded (26 commands, 80 reference)
+ Skills:     107 loaded (27 commands, 80 reference)

docs/skills/overview.mdx:156
- Of OrchestKit's 107 skills, **26 are user-invocable**
+ Of OrchestKit's 107 skills, **27 are user-invocable**

docs/reference/index.mdx:13
- Card description: "Reusable knowledge modules — 26 commands + 80..."
+ Card description: "Reusable knowledge modules — 27 commands + 80..."

Why this matters

A skill without scripts is a contract without an implementation — every invocation requires the model to re-read the SKILL.md and translate prose into shell. Other users-invocable skills ship the shell, then describe it. Now /ork:dev matches.

Rules give the AI guardrails it can cite ("per idempotent-boot.md, re-running while live is a no-op") instead of relying on prose embedded in SKILL.md. Test cases give golden behaviors that can be eval'd by /ork:bare-eval.