docs: add validate-ldcli-changes skill and PR video requirement

[nieblara]

Requirements

• I have added test coverage for new or changed functionality — N/A (docs/skill only; no runtime code changed)
• I have followed the repository's pull request submission guidelines
• I have validated my changes against all supported platform versions — validated the documented workflow by running the built ldcli locally (see demo below)

Related issues

None.

Describe the solution you've provided

Equips agents to validate CLI changes the way a human would, and requires proof in every PR:

• New skill .cursor/skills/validate-ldcli-changes/SKILL.md covering:
  • Building the CLI locally (make build → test the ./ldcli you just built).
  • Pointing the CLI at staging (https://app.staging.launchdarkly.com) via LD_BASE_URI / LD_ACCESS_TOKEN from Cursor Secrets — never production, never committed tokens.
  • A change-type → validation matrix so agents run the validation that actually exercises the change (resource lifecycles, output formats, config precedence, auth/error paths, dev-server, interactive setup/login).
  • Safe testing on shared staging (scratch project, unique keys, clean up created resources).
  • Recording a human-style screen recording of the CLI in action and attaching it to the PR, with no secrets on screen.
• AGENTS.md now has a "Validating CLI changes" section pointing at the skill and stating the staging-validation + PR-video requirement, so agents discover it automatically. (CLAUDE.md symlinks to AGENTS.md.)

Describe alternatives you've considered

• Putting the guidance only in AGENTS.md: kept a short pointer there but moved the detailed workflow into a skill so it loads on demand for CLI-change tasks and stays maintainable.
• Committing a staging token/base URI: rejected — credentials stay in Cursor Secrets (LD_ACCESS_TOKEN), never in the repo, config file, logs, or the video.

Additional context

Demo of the documented workflow (this is also the style of human-driven CLI video the skill requires in every PR). It builds the binary, points LD_BASE_URI at staging, and runs commands live. whoami and flags list intentionally show the auth-required error — the flags list message deriving https://app.staging.launchdarkly.com/settings/authorization confirms the staging wiring. No secrets appear on screen.

Note: authenticated staging validation (e.g. whoami, real resource lifecycles) requires an LD_ACCESS_TOKEN staging secret, which is not present in this environment. The demo validates the local, no-auth portions of the documented workflow (build, config, output formatting, and the auth-required error path that confirms staging wiring). Add an LD_ACCESS_TOKEN staging secret to enable full authenticated staging runs.