feat: intent-first instructions and consistent worktree placement

[vsem-azamat]

Summary

• Specify observable outcomes, constraints, acceptance evidence, and scope rather than prescribing internal names and implementation details.
• Keep governing architecture and committed handoff interfaces explicit; leave routine implementation choices to the executor.
• Clarify that compliant internal implementation choices are not unrequested extras in review.
• Default new worktrees to <primary-checkout>/.worktrees/<task>, with explicit user/project placement policy taking precedence over native host defaults. Preserve existing isolation and require authorization before cleanup.
• Regenerate repository skills and plugin distributions and add focused regression checks.

Docs updated first: method/README.md, playbooks/planning.md, playbooks/implement.md, playbooks/delegate-slice.md, playbooks/worktrees.md

Verification

• Final local full suite: 812 tests, 808 passed, 4 skipped, 0 failed (stdd verify -- npm test).
• npm run check: exit 0, with warnings.
• npm run selfcheck and git diff --check: passed.
• Independent fresh-context Claude Code review, followed by a targeted re-review of the final worktree-policy corrections: no remaining material findings. This was an external coordinator review, not a recorded stdd review verdict.

Limits and scope

• This is instruction-contract improvement, not a demonstrated agent-speed or token-cost improvement. Touched canonical prose is 57,849 UTF-8 bytes versus 57,896 before; the reduction is negligible.
• Regression tests check generated instruction content and packaging, not real agent behavior.
• No changes to review verdict/gate/budget mechanics, native filesystem, integration runners, or worker runtime.
• No release, consumer-project installation, or merge is included.