harness-operatorlisted

# Harness Operator Skill Use this skill to enter the generated harness without relying on hidden chat memory. 1. Read root `AGENTS.md`. 2. Run or review `./init.sh` unless the user forbids local checks. 3. Read `feature_list.json`, `progress.md`, and `session-handoff.md`. 4. Read `harness/shared/ROLE_FILE_INDEX.md`. 5. Load `harness/operators/claude-code/AGENT.md`. 6. Load `harness/shared/ACTIVE_SNAPSHOT.md`, `harness/shared/PROJECT_PROFILE.json`, and `harness/shared/WORKSTREAM_PROFILE.json`. 7. Load `harness/shared/OBSERVABILITY.md` and `harness/shared/VISUALIZATION_SPEC_POLICY.md`. 8. Continue only through file-backed state and task artifacts. Preserve shared memory: - Operator memory: `harness/shared/`, `feature_list.json`, `progress.md`, `session-handoff.md`, `harness/tasks/` - Team memory: `harness/teams/*/TEAM_CONTEXT.md` - Do not treat Claude Code skill memory or private chat as authority. - Use `python3 scripts/harnessctl.py report` for a local compiled HTML status view when needed. The report is not canonical memory. - Use `python3 scripts/harnessctl.py viz-export --backend local_file` for local event payloads when a viz backend worker needs smoke evidence. - Treat visualization and diagram information architecture as Claude-owned; event plumbing may be delegated only after approved `VISUALIZATION_SPEC.md`.