Fix high-priority factual errors across docs

[rbowen]

Summary

• Six factual errors fixed: stale skill counts in threat model, "proposed" banners on shipped skills, fabricated directory tree in harness guide, wrong CLI syntax in pattern catalogue, and wrong tool path in privacy-llm setup.

Type of change

• Documentation (docs/, README.md, CONTRIBUTING.md)

Test plan

• prek run --all-files passes
• Other: Verified each fix against the actual repo state — tools/agent-guard/ tree via fd, tools/privacy-llm/checker/pyproject.toml entry point, tools/privacy-llm/redactor/pyproject.toml entry points, and skills/*/SKILL.md frontmatter family: fields. Prose-only; no skill behaviour changes.

RFC-AI-0004 compliance

N/A — documentation-only change; no mutations, no tool behaviour, no private content handling.

Linked issues

Notes for reviewers (optional)

The threat-model.md fix adds the three missing skills to their
respective STRIDE family skill lists but does NOT add new STRIDE rows
for them — that's a substantive security analysis task, not a docs
copyedit. A follow-up will add the missing STRIDE coverage.

The process.md change removes 13 *(proposed)* markers from
per-step skill references. All ten release-management skills have
complete SKILL.md files on main.

[potiuk]

NICE!

[potiuk]

The threat-model.md fix adds the three missing skills to their
respective STRIDE family skill lists but does NOT add new STRIDE rows
for them — that's a substantive security analysis task, not a docs
copyedit. A follow-up will add the missing STRIDE coverage.

The threat-model indeed needs more review.