[P2][security] Insecure patterns taught in skill docs (auth-bypass examples, secrets in argv/stdout)

[Toqsick]

Source: RepoLens Security-Audit 2026-08-16 (findings security/authorization/001–005+007, security/secrets/002+004–006, adversarial-verified ✅)

Skill docs teach insecure patterns that get copied into real code:

• 001 Telegram wildcard * allow-everyone example
• 002/006 WS/REST patches re-enabling ?token= in query strings
• 003 _TokenPasswordProvider static credential pattern
• 004 PUBLIC_API_PATHS documented as auth-bypass knob
• 007 gateway-adapter-development documents allow_all_env as first-class option
• secrets/002 .gitignore misses credential filename patterns (partially fixed in 182b3fd)
• secrets/004 Harbor default password Harbor12345 as CLI fallback
• secrets/005 Telegram bot token passed as curl argv (visible in process list)
• secrets/006 api-discovery.sh echoes token prefix to stdout

Suggested fix: rewrite doc examples to the secure variants (env vars, headers, fail-closed), add a doc-lint grep to CI.

[Toqsick]

Triage 2026-08-16: ProjectZomboiD #3–#7 sind heute gefixt und geschlossen (PR #8). my-agent-tools-Härtung ist das nächste RepoLens-Backlog, geplante Reihenfolge: #9 (Path-Traversal) → #8 (Skill-Docs) → #7 (Security-Header) → #6/#5 (Rate-Limiting). Die Visibility-Entscheidung (public vs. private) fällt nach der Härte-Runde.