feat(tools): add idempotentHint + openWorldHint annotations to all 12 tools (R3 F39) (#42)

R3 § 4 Bundle 2 finding F39: tool annotations were missing
`idempotentHint`, `openWorldHint`, and `title`. Non-Claude MCP clients
(GPT-4, Gemini, Kimi, DeepSeek via Cursor/ChatGPT/Codex/etc.) reason
about side-effect classes using these annotations before invoking;
without them, those models fall back to conservative defaults (refuse,
ask, escalate) — degrading UX.

This PR adds `idempotentHint` and `openWorldHint` to all 12 tools.
`title` was already present at the top level of every tool definition,
so no `title` changes were needed.

Per CLAUDE.md verb-pattern table + MCP 2025-06-18 spec semantics:

| Verb       | readOnly | destructive | idempotent | openWorld |
|------------|----------|-------------|------------|-----------|
| list*, get*| true     | false       | true       | false     |
| reserve*   | false    | false       | false      | false     |
| upload*    | false    | false       | false      | false     |
| confirm*   | false    | false       | true       | false     |
| terminate* | false    | true        | false      | false     |

`openWorldHint: false` is uniform across all 12 tools — the Kobiton
MCP server is a bounded API surface (`api.kobiton.com/mcp`), not an
"open world" like web search or arbitrary HTTP fetch.

`idempotentHint` values derive from MCP/HTTP idempotency semantics
("no additional effect after first request"). `confirmAppUpload` is
idempotent because the second call after a successful first call sees
the existing app record rather than creating a new one (consistent with
R2 finding F5/F25 observations); but `uploadAppToStore` is NOT
idempotent because it allocates a fresh S3 pre-signed URL per call.

Tools touched (all 12):
- tools/devices.yaml: listDevices, getDeviceStatus, reserveDevice,
  terminateReservation
- tools/sessions.yaml: listSessions, getSession, getSessionArtifacts,
  terminateSession
- tools/apps.yaml: listApps, uploadAppToStore, confirmAppUpload, getApp

`outputSchema` per tool is NOT included in this PR — adding outputSchema
without evidence of the actual server response shape would be
speculative. Defer to a follow-up PR backed by empirical response
captures (test-runs/ has the raw JSON dumps from the R2 experiment
campaign that can ground outputSchemas accurately).

Cross-references:
- R3 § 4 Bundle 2 fork issue #26
- MCP spec primitives: modelcontextprotocol.io/specification/2025-06-18/server/tools

Validation:
- `pnpm run validate` passes
- `pnpm test` passes (22 tests)
- `validate-skills-schema.py --enterprise` — 0 ERRORs (all pre-existing
  WARNs unchanged); Skills: 1, Agents: 3; 91/100 average

Signed-off-by: jeremylongshore <jeremylongshore@users.noreply.github.com>
Co-authored-by: jeremylongshore <jeremylongshore@users.noreply.github.com>

Author: jeremylongshore

tools/apps.yaml

@@ -8,6 +8,8 @@ tools:
     annotations:
       readOnlyHint: true
       destructiveHint: false
+      idempotentHint: true
+      openWorldHint: false
     inputSchema:
       type: object
       properties:
@@ -46,6 +48,8 @@ tools:
     annotations:
       readOnlyHint: false
       destructiveHint: false
+      idempotentHint: false
+      openWorldHint: false
     inputSchema:
       type: object
       properties:
@@ -79,6 +83,8 @@ tools:
     annotations:
       readOnlyHint: false
       destructiveHint: false
+      idempotentHint: true
+      openWorldHint: false
     inputSchema:
       type: object
       properties:
@@ -106,6 +112,8 @@ tools:
     annotations:
       readOnlyHint: true
       destructiveHint: false
+      idempotentHint: true
+      openWorldHint: false
     inputSchema:
       type: object
       properties:

tools/devices.yaml

@@ -9,6 +9,8 @@ tools:
     annotations:
       readOnlyHint: true
       destructiveHint: false
+      idempotentHint: true
+      openWorldHint: false
     inputSchema:
       type: object
       properties:
@@ -59,6 +61,8 @@ tools:
     annotations:
       readOnlyHint: true
       destructiveHint: false
+      idempotentHint: true
+      openWorldHint: false
     inputSchema:
       type: object
       properties:
@@ -84,6 +88,8 @@ tools:
     annotations:
       readOnlyHint: false
       destructiveHint: false
+      idempotentHint: false
+      openWorldHint: false
     inputSchema:
       type: object
       properties:
@@ -112,6 +118,8 @@ tools:
     annotations:
       readOnlyHint: false
       destructiveHint: true
+      idempotentHint: false
+      openWorldHint: false
     inputSchema:
       type: object
       properties:

tools/sessions.yaml

@@ -8,6 +8,8 @@ tools:
     annotations:
       readOnlyHint: true
       destructiveHint: false
+      idempotentHint: true
+      openWorldHint: false
     inputSchema:
       type: object
       properties:
@@ -68,6 +70,8 @@ tools:
     annotations:
       readOnlyHint: true
       destructiveHint: false
+      idempotentHint: true
+      openWorldHint: false
     inputSchema:
       type: object
       properties:
@@ -91,6 +95,8 @@ tools:
     annotations:
       readOnlyHint: true
       destructiveHint: false
+      idempotentHint: true
+      openWorldHint: false
     inputSchema:
       type: object
       properties:
@@ -114,6 +120,8 @@ tools:
     annotations:
       readOnlyHint: false
       destructiveHint: true
+      idempotentHint: false
+      openWorldHint: false
     inputSchema:
       type: object
       properties: