Describe rook 0.1.6 in the agent skill

[samyakLambda]

Update the Rook skill for CLI 0.1.6.

The original workflow-generated change corrected the fields described for
rook doctor. A subsequent audit found behavior changes that the supplied
release notes/help did not describe precisely enough; those corrections were
added manually without another model run.

Changes

• Mark the JSON-string raw_response decoding limitation as fixed in 0.1.6;
  document the 8 MiB decode limit and preserve guidance for uncheckable JSON
  and missing evidence.
• Document pass rate as Pass / (Pass + Fail + Unable to Verify);
  decided remains Pass + Fail. All Unable to Verify yields 0; no recorded
  results yields null. Unable to Verify remains separate from Fail.
• Explain per-environment project selections in settings.json under
  selections[environment], including an environment with no selected project.
• Retain the doctor correction and synchronize both skill mirrors.

Validation

• Published @testmuai/rook@0.1.6 reports version 0.1.6; the skill flag harness
  checked 48 distinct command lines using isolated local state.
• Mirror/frontmatter checks pass; offline CI contract passes 16/16 cases;
  skill installer tests pass; git diff --check passes.
• JSON-path behavior was compared across the 0.1.5 and 0.1.6 release revisions
  without model calls: a valid JSON-string response changed from Unable to
  Verify to Pass. Pass-rate and environment-selection prose was checked against
  the released behavior.

Original generation used Claude Opus 5.5. These additional edits are manual;
no paid generation, skill merge, or skill publication was performed.
Human review must assess prose truth/completeness; format and flag checks alone
do not establish those properties. Compare with current main before merging.