Describe rook 0.1.5 in the agent skill

[samyakLambda]

Update the Rook agent skill for CLI 0.1.5.

The generated change removes technical_injection from the documented rook generate --category values; the released CLI help does not accept it. A manual audit also found gaps in the existing skill:

• Explain that profile add may make several real target calls and can retain an unverified draft, with a path to test or fix it.
• Explain that verified profile authoring can store environment values it used, while shell values take precedence.
• Scope unrunnable guidance to observed conversation and usage capabilities; file and image input kinds alone do not establish that a scenario cannot run.
• Describe the readable project directory layout, profile README, explicit run --profile selection, and paid run --rca option.

The canonical skill and its Claude Code and Codex mirrors are synchronized. Locally, all seven public script harnesses passed with the published @testmuai/rook@0.1.5 binary, including 48 documented command lines, 16 offline CI contract cases, and 19 installer cases. git diff --check passed. The 0.1.5 changelog has no detailed release notes, so these edits are limited to behavior checked against the released CLI help and release source.

Please review the prose before merging. The automated harnesses cannot establish every behavioral claim.