sccs 2.72.2__tar.gz → 2.74.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {sccs-2.72.2 → sccs-2.74.0}/.claude/settings.local.json +2 -1
- {sccs-2.72.2 → sccs-2.74.0}/.github/workflows/ci.yml +30 -13
- sccs-2.74.0/.remember/logs/autonomous/save-20261108-180917-11155.log +1 -0
- sccs-2.74.0/.remember/logs/autonomous/session-end-090044-85773.log +1 -0
- sccs-2.74.0/.remember/logs/hook-errors.log +2 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/logs/memory-2026-11-08.log +6 -0
- sccs-2.74.0/.remember/logs/memory-2026-11-09.log +12 -0
- sccs-2.74.0/.remember/tmp/last-save-ts +1 -0
- sccs-2.74.0/.remember/tmp/last-summary-failure +1 -0
- sccs-2.74.0/.remember/tmp/ndc-generation +1 -0
- sccs-2.74.0/.remember/tmp/save-session.pid +1 -0
- sccs-2.74.0/.remember/tmp/transcript-scan.3732e73d-145d-4d96-bcde-f3e49c15ef38 +1 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/today-2026-11-08.md +6 -1
- {sccs-2.72.2 → sccs-2.74.0}/CLAUDE.md +4 -3
- {sccs-2.72.2 → sccs-2.74.0}/PKG-INFO +8 -2
- {sccs-2.72.2 → sccs-2.74.0}/README.md +6 -1
- {sccs-2.72.2 → sccs-2.74.0}/RELEASE_NOTES.md +22 -0
- {sccs-2.72.2 → sccs-2.74.0}/docs/architecture.md +26 -2
- {sccs-2.72.2 → sccs-2.74.0}/docs/development/code-map.md +8 -0
- {sccs-2.72.2 → sccs-2.74.0}/docs/development/maintainer-notes.md +19 -0
- sccs-2.74.0/docs/superpowers/plans/2026-11-09-agent-supervisor-setup.md +465 -0
- sccs-2.74.0/docs/superpowers/specs/2026-11-09-agent-supervisor-setup-design.md +254 -0
- sccs-2.74.0/docs/usage/agents.md +163 -0
- {sccs-2.72.2 → sccs-2.74.0}/docs/usage/cli-reference.md +22 -2
- {sccs-2.72.2 → sccs-2.74.0}/docs/usage/doctor.md +36 -0
- {sccs-2.72.2 → sccs-2.74.0}/pyproject.toml +4 -1
- {sccs-2.72.2 → sccs-2.74.0}/sccs/__init__.py +1 -1
- {sccs-2.72.2 → sccs-2.74.0}/sccs/cli.py +252 -1
- {sccs-2.72.2 → sccs-2.74.0}/sccs/config/defaults.py +4 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/config/loader.py +41 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/doctor/installer.py +6 -1
- {sccs-2.72.2 → sccs-2.74.0}/sccs/doctor/reporter.py +7 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/doctor/runner.py +28 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/doctor/schema.py +58 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/doctor/skill_packages.py +99 -1
- sccs-2.74.0/sccs/supervisor/__init__.py +45 -0
- sccs-2.74.0/sccs/supervisor/edit.py +132 -0
- sccs-2.74.0/sccs/supervisor/models.py +99 -0
- sccs-2.74.0/sccs/supervisor/render.py +116 -0
- sccs-2.74.0/sccs/supervisor/templates/agy.md +55 -0
- sccs-2.74.0/sccs/supervisor/templates/codex.md +50 -0
- sccs-2.74.0/sccs/supervisor/templates/opencode.md +57 -0
- sccs-2.74.0/sccs/supervisor/tools.py +107 -0
- sccs-2.74.0/sccs/supervisor/tui.py +521 -0
- sccs-2.74.0/sccs/supervisor/wrappers.py +216 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/conftest.py +56 -0
- sccs-2.74.0/tests/fixtures/supervisor/antigravity-worker.md +50 -0
- sccs-2.74.0/tests/fixtures/supervisor/opencode-worker.md +52 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_doctor_ceilings.py +45 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_doctor_skill_packages.py +136 -0
- sccs-2.74.0/tests/test_supervisor_cli.py +193 -0
- sccs-2.74.0/tests/test_supervisor_edit.py +198 -0
- sccs-2.74.0/tests/test_supervisor_models.py +111 -0
- sccs-2.74.0/tests/test_supervisor_render.py +163 -0
- sccs-2.74.0/tests/test_supervisor_tools.py +70 -0
- sccs-2.74.0/tests/test_supervisor_tui.py +240 -0
- sccs-2.74.0/tests/test_supervisor_wrappers.py +215 -0
- {sccs-2.72.2 → sccs-2.74.0}/usage/AGENT.md +23 -2
- sccs-2.74.0/usage/multi-provider-agents.de.md +207 -0
- sccs-2.74.0/usage/multi-provider-agents.en.md +207 -0
- {sccs-2.72.2 → sccs-2.74.0}/uv.lock +52 -7
- sccs-2.72.2/.remember/now.md +0 -9
- sccs-2.72.2/.remember/tmp/last-save-ts +0 -1
- sccs-2.72.2/.remember/tmp/ndc-generation +0 -1
- sccs-2.72.2/.remember/tmp/now-day +0 -1
- sccs-2.72.2/.remember/tmp/save-session.pid +0 -1
- sccs-2.72.2/.remember/tmp/transcript-scan.3732e73d-145d-4d96-bcde-f3e49c15ef38 +0 -1
- {sccs-2.72.2 → sccs-2.74.0}/.gitignore +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.planning/PROJECT.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.planning/ROADMAP.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.planning/STATE.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.planning/codebase/ARCHITECTURE.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.planning/codebase/CONCERNS.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.planning/codebase/CONVENTIONS.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.planning/codebase/INTEGRATIONS.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.planning/codebase/STACK.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.planning/codebase/STRUCTURE.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.planning/codebase/TESTING.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.planning/phases/1-statusline-detector/CONTEXT.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.pre-commit-config.yaml +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.project-tips +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/.gitignore +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/.install-marker +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/logs/autonomous/save-20261108-120807-69267.log +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/logs/autonomous/save-20261108-162925-24306.log +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/logs/autonomous/save-20261108-163215-29743.log +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/logs/autonomous/save-20261108-163557-37304.log +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/logs/autonomous/save-20261108-164157-49561.log +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/logs/autonomous/save-20261108-164358-54348.log +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/logs/autonomous/save-20261108-164606-59900.log +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/logs/autonomous/save-20261108-164807-64649.log +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/logs/autonomous/save-20261108-165109-71161.log +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/logs/autonomous/save-20261108-165319-77089.log +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/logs/autonomous/save-20261108-165538-82883.log +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/logs/autonomous/save-20261108-165743-88535.log +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/logs/autonomous/save-20261108-175923-96365.log +0 -0
- /sccs-2.72.2/.remember/logs/hook-errors.log → /sccs-2.74.0/.remember/now.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/tmp/capture-alive +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/tmp/capture-alive.d/3732e73d-145d-4d96-bcde-f3e49c15ef38 +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/tmp/capture-gap-reported +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/tmp/case-divergence +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/tmp/handoff-path +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/tmp/handoff-path.3732e73d-145d-4d96-bcde-f3e49c15ef38 +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/tmp/last-ndc.ts +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/tmp/last-save.json +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/tmp/position.3732e73d-145d-4d96-bcde-f3e49c15ef38 +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/tmp/position.dd17b6a8-0601-478c-bb30-3acd459c0830 +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/tmp/post-tool-ran +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/tmp/session-dir-cache +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/tmp/session-slug +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/tmp/start-context.cache +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.remember/tmp/start-context.manifest +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/.gitignore +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/afterwork-docs-report.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/final-fix-report-2.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/final-fix-report.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/progress.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/review-068b66e..a372f09.diff +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/review-1078b36..d29274e.diff +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/review-162f080..5d48f3e.diff +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/review-30cfca4..e9e5470.diff +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/review-5d48f3e..d1f4c57.diff +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/review-7fa4600..1078b36.diff +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/review-8666790..30cfca4.diff +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/review-91cf827..8666790.diff +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/review-a0c1c3d..068b66e.diff +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/review-a465ed9..068b66e.diff +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/review-a465ed9..c3990b7.diff +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/review-c3990b7..162f080.diff +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/review-d1f4c57..91cf827.diff +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/review-d29274e..a0c1c3d.diff +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/review-e9e5470..f20ba5c.diff +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/review-f20ba5c..7fa4600.diff +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/task-1-brief.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/task-1-report.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/task-2-brief.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/task-2-report.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/task-3-brief.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/task-3-report.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/task-4-brief.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/task-4-report.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/task-5-brief.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/task-5-report.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/task-6-brief.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/task-6-report.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/task-7-brief.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/task-7-report.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/task-8-brief.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/task-8-report.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/task-9-brief.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/.superpowers/sdd/2026-09-01-deployment-profiles/task-9-report.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/AGENTS.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/LICENSE +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/PLAN_v2.29.0.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/docs/superpowers/plans/2026-08-25-codex-hooks-export.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/docs/superpowers/plans/2026-09-01-deployment-profiles.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/docs/superpowers/plans/2026-09-15-mirror-parity.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/docs/superpowers/specs/2026-08-25-codex-hooks-export-design.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/docs/superpowers/specs/2026-09-01-deployment-profiles-design.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/docs/superpowers/specs/2026-09-15-mirror-parity-design.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/docs/usage/capacity.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/docs/usage/categories.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/docs/usage/codex.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/docs/usage/deploy.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/docs/usage/memory-bridge.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/docs/usage/opencode.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/docs/usage/pi.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/docs/usage/platforms.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/docs/usage/profiles.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/docs/usage/sync.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/docs/usage/transfer.md +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/__main__.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/capacity/__init__.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/capacity/probes.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/capacity/report.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/capacity/schema.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/config/__init__.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/config/migration.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/config/schema.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/convert/__init__.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/convert/claude_to_codex.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/convert/claude_to_codex_hooks.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/convert/claude_to_opencode.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/convert/fish_to_pwsh.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/convert/fish_to_zsh.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/convert/frontmatter.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/convert/rules.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/convert/templates.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/convert/toml_write.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/convert/zsh_block.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/convert/zsh_rules.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/convert/zsh_templates.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/deploy/__init__.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/deploy/bundle.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/deploy/defaults.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/deploy/install.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/deploy/receipt.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/deploy/resolve.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/deploy/revoke.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/deploy/schema.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/deploy/traces.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/docs/__init__.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/docs/generator.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/doctor/__init__.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/doctor/_paths.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/doctor/cao.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/doctor/defaults.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/doctor/detectors.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/doctor/managed.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/doctor/mirror.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/doctor/profiles.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/doctor/scope_patch.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/doctor/shared_config.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/doctor/state.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/doctor/statusline.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/git/__init__.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/git/operations.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/git/resolve.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/integrations/__init__.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/integrations/antigravity.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/integrations/claude_desktop.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/integrations/codex.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/integrations/codex_hooks.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/integrations/detectors.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/integrations/opencode.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/integrations/pi.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/integrations/skill_limits.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/integrations/sync_all.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/output/__init__.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/output/console.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/output/diff.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/output/json_emit.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/output/merge.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/sync/__init__.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/sync/actions.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/sync/category.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/sync/engine.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/sync/item.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/sync/settings.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/sync/state.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/transfer/__init__.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/transfer/exporter.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/transfer/importer.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/transfer/manifest.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/transfer/ui.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/utils/__init__.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/utils/hashing.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/utils/logging.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/utils/paths.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs/utils/platform.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/sccs-openai.png +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/__init__.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_capacity.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_category_toggle_raw.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_cli.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_cli_capability_card.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_cli_json.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_codex_cli.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_codex_config.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_codex_convert.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_codex_detector.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_codex_export.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_codex_hooks_cli.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_codex_hooks_convert.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_codex_hooks_merge.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_config.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_conflict_resolution.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_console.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_convert.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_convert_zsh.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_deploy_bundle.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_deploy_cli.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_deploy_install.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_deploy_receipt.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_deploy_resolve.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_deploy_revoke.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_deploy_schema.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_deploy_traces.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_diff.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_docs.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_doctor.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_doctor_cao.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_doctor_mirror.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_doctor_orphans.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_doctor_shared_config.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_exporter_security.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_frontmatter.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_git_operations.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_git_resolve.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_hashing.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_importer_security.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_integrations.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_local_repo_path.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_merge.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_migration.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_opencode_convert.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_opencode_detector.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_opencode_mcp.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_opencode_models.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_paths_atomic.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_paths_security.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_pi_detector.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_pi_export.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_platform.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_platform_utils.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_profiles.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_settings.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_skill_limits.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_statusline.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_sync.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_sync_all.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_sync_moved_local_root.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_transfer.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_transfer_ui.py +0 -0
- {sccs-2.72.2 → sccs-2.74.0}/tests/test_windows_documents.py +0 -0
|
@@ -10,13 +10,13 @@ jobs:
|
|
|
10
10
|
lint:
|
|
11
11
|
runs-on: ubuntu-latest
|
|
12
12
|
steps:
|
|
13
|
-
- uses: actions/checkout@
|
|
13
|
+
- uses: actions/checkout@v7
|
|
14
14
|
|
|
15
15
|
- name: Install uv
|
|
16
|
-
uses: astral-sh/setup-uv@
|
|
16
|
+
uses: astral-sh/setup-uv@v10.3.3
|
|
17
17
|
|
|
18
18
|
- name: Set up Python
|
|
19
|
-
uses: actions/setup-python@
|
|
19
|
+
uses: actions/setup-python@v7
|
|
20
20
|
with:
|
|
21
21
|
python-version: "3.12"
|
|
22
22
|
|
|
@@ -37,9 +37,26 @@ jobs:
|
|
|
37
37
|
- name: Bandit security scan
|
|
38
38
|
run: uv run bandit -r sccs/ -ll
|
|
39
39
|
|
|
40
|
-
|
|
41
|
-
|
|
42
|
-
|
|
40
|
+
audit:
|
|
41
|
+
# Kept out of `lint`: a new advisory is not a code defect, and it used to
|
|
42
|
+
# turn the lint badge red. Audits the locked set, not what resolves today;
|
|
43
|
+
# the project itself is not on PyPI under this lock, so it is skipped.
|
|
44
|
+
runs-on: ubuntu-latest
|
|
45
|
+
steps:
|
|
46
|
+
- uses: actions/checkout@v7
|
|
47
|
+
|
|
48
|
+
- name: Install uv
|
|
49
|
+
uses: astral-sh/setup-uv@v10.3.3
|
|
50
|
+
|
|
51
|
+
- name: Audit runtime dependencies (blocking)
|
|
52
|
+
# What `pip install sccs` ships. A finding here fails the build.
|
|
53
|
+
run: uv export --frozen --no-hashes --no-emit-project --no-dev | uvx pip-audit --disable-pip --no-deps -r /dev/stdin
|
|
54
|
+
|
|
55
|
+
- name: Audit dev tooling (report only)
|
|
56
|
+
# pytest, ruff, mypy, pre-commit and their chain (virtualenv, filelock …)
|
|
57
|
+
# never reach users. A finding shows as a warning; fix it with
|
|
58
|
+
# `uv lock --upgrade-package <name>` when convenient.
|
|
59
|
+
continue-on-error: true
|
|
43
60
|
run: uv export --frozen --no-hashes --no-emit-project --all-extras | uvx pip-audit --disable-pip --no-deps -r /dev/stdin
|
|
44
61
|
|
|
45
62
|
test:
|
|
@@ -48,13 +65,13 @@ jobs:
|
|
|
48
65
|
matrix:
|
|
49
66
|
python-version: ["3.11", "3.12", "3.13", "3.14"]
|
|
50
67
|
steps:
|
|
51
|
-
- uses: actions/checkout@
|
|
68
|
+
- uses: actions/checkout@v7
|
|
52
69
|
|
|
53
70
|
- name: Install uv
|
|
54
|
-
uses: astral-sh/setup-uv@
|
|
71
|
+
uses: astral-sh/setup-uv@v10.3.3
|
|
55
72
|
|
|
56
73
|
- name: Set up Python ${{ matrix.python-version }}
|
|
57
|
-
uses: actions/setup-python@
|
|
74
|
+
uses: actions/setup-python@v7
|
|
58
75
|
with:
|
|
59
76
|
python-version: ${{ matrix.python-version }}
|
|
60
77
|
|
|
@@ -70,15 +87,15 @@ jobs:
|
|
|
70
87
|
|
|
71
88
|
build:
|
|
72
89
|
runs-on: ubuntu-latest
|
|
73
|
-
needs: [lint, test]
|
|
90
|
+
needs: [lint, test, audit]
|
|
74
91
|
steps:
|
|
75
|
-
- uses: actions/checkout@
|
|
92
|
+
- uses: actions/checkout@v7
|
|
76
93
|
|
|
77
94
|
- name: Install uv
|
|
78
|
-
uses: astral-sh/setup-uv@
|
|
95
|
+
uses: astral-sh/setup-uv@v10.3.3
|
|
79
96
|
|
|
80
97
|
- name: Set up Python
|
|
81
|
-
uses: actions/setup-python@
|
|
98
|
+
uses: actions/setup-python@v7
|
|
82
99
|
with:
|
|
83
100
|
python-version: "3.12"
|
|
84
101
|
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
18:09:17 [post-tool] save triggered
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
09:00:44 [session-end] flush started
|
|
@@ -0,0 +1,2 @@
|
|
|
1
|
+
09:00:46 [haiku] ERROR: call-haiku error: claude exited 1: Failed to authenticate: OAuth session expired and could not be refreshed
|
|
2
|
+
09:00:46 [session-end] WARNING: save-session.sh --force exited 1 at session end -- this session's unsaved tail may be lost. See /Users/picard/gitbase/PyPi-Projects/sccs/.remember/logs/autonomous/session-end-090044-85773.log for what save-session.sh itself logged.
|
|
@@ -92,3 +92,9 @@
|
|
|
92
92
|
17:59:26 [write] position -> 2102
|
|
93
93
|
17:59:26 [ndc] now.md -> today-2026-11-08.md
|
|
94
94
|
17:59:26 [ndc] running (PID 96809)
|
|
95
|
+
17:59:34 [ndc] tokens: 2+2082cache->1873out ($0.001465)
|
|
96
|
+
17:59:35 [ndc] 1539->1017b (-33%)
|
|
97
|
+
18:09:17 [hook] save-session: PROJECT_DIR=/Users/picard/gitbase/PyPi-Projects/sccs PIPELINE_DIR=/Users/picard/.claude/plugins/cache/claude-plugins-official/remember/0.43.1 PYTHON=python3 REMEMBER_DIR=/Users/picard/gitbase/PyPi-Projects/sccs/.remember
|
|
98
|
+
18:09:17 [extract] session 3732e73d-145d-4d96-bcde-f3e49c15ef38
|
|
99
|
+
18:09:17 [extract] 13 exchanges (1 human)
|
|
100
|
+
18:09:17 [extract] 1 human msgs < 3, skip
|
|
@@ -0,0 +1,12 @@
|
|
|
1
|
+
09:00:44 [hook] session-end: reason=other session=3732e73d-145d-4d96-bcde-f3e49c15ef38
|
|
2
|
+
09:00:44 [hook] save-session: PROJECT_DIR=/Users/picard/gitbase/PyPi-Projects/sccs PIPELINE_DIR=/Users/picard/.claude/plugins/cache/claude-plugins-official/remember/0.43.1 PYTHON=python3 REMEMBER_DIR=/Users/picard/gitbase/PyPi-Projects/sccs/.remember
|
|
3
|
+
09:00:44 [force] bypassing cooldown + min msgs
|
|
4
|
+
09:00:44 [extract] session 3732e73d-145d-4d96-bcde-f3e49c15ef38
|
|
5
|
+
09:00:44 [extract] 14 exchanges (1 human)
|
|
6
|
+
09:00:44 [haiku] calling (branch: main)
|
|
7
|
+
09:00:45 [haiku] WARNING: this CLI failed authentication (Failed to authenticate: OAuth session expired and could not be refreshed); retrying WITHOUT hook isolation in case the isolated run's own environment is simply missing a credential the normal one has. The nested call will run with your hooks registered -- a hook that blocks it returns its block message as if it were the model's reply (#202).
|
|
8
|
+
09:00:46 [haiku] WARNING: the un-isolated retry failed with the same authentication error (Failed to authenticate: OAuth session expired and could not be refreshed) -- hook isolation was not the cause. The CLI's own saved login has expired; log in again with your coding agent's own CLI. This plugin reads no credential of its own any more -- there is no setting here to configure (#129/#131/#860).
|
|
9
|
+
09:00:46 [haiku] call exited 1; tokens already spent are unknown -- the failure carried no usage block, so this run's reported cost is lower than what it actually cost
|
|
10
|
+
09:00:46 [haiku] ERROR: call-haiku error: claude exited 1: Failed to authenticate: OAuth session expired and could not be refreshed
|
|
11
|
+
09:00:46 [haiku] failure 1/3 on this span -- will retry next run
|
|
12
|
+
09:00:46 [session-end] WARNING: save-session.sh --force exited 1 at session end -- this session's unsaved tail may be lost. See /Users/picard/gitbase/PyPi-Projects/sccs/.remember/logs/autonomous/session-end-090044-85773.log for what save-session.sh itself logged.
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
1794211244
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
3732e73d-145d-4d96-bcde-f3e49c15ef38:2193 1
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
3
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
85814
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
5636409 2167
|
|
@@ -3,4 +3,9 @@ sccs 2.72.0 (`91ff9ae`; `sync -c` mehrfach + unbek. Kategorie = Fehler; `doctor
|
|
|
3
3
|
## 13:04 | main
|
|
4
4
|
sccs 2.71.0–2.71.2 released+tagged GitLab/GitHub (CI grün): per-host `local.yaml` repo path + `config repo-path`; Windows Known-Folder `~/Documents`→OneDrive; `categories enable` writes only `enabled`; per-category sync-root tracking. blockiert: PyPI-publish 2.71.2 manuell (User); Windows-Migration (Schritte 1–9) offen; `ssh -T git@github.com` (cache-tax) offen.
|
|
5
5
|
## 16:32 | main
|
|
6
|
-
sccs 2.72.0 pushed GitLab+GitHub (ungetaggt, nicht auf PyPI); claude-mem auf Ausmusterungsliste; andere Rechner brauchen noch `doctor optimize` + ggf. manuelles `~/.claude-mem`-Löschen; lokal claude-mem komplett entfernt (76 MB Daten); 13 `.remember`-Ordner gelöscht, 4 bleiben bis User committet (v18-kpi, v18-hyserve/modules, herdr-app, memocove); sccs-`.remember` offen; Codex/agy-CLI-Tool-Check läuft.
|
|
6
|
+
sccs 2.72.0 pushed GitLab+GitHub (ungetaggt, nicht auf PyPI); claude-mem auf Ausmusterungsliste; andere Rechner brauchen noch `doctor optimize` + ggf. manuelles `~/.claude-mem`-Löschen; lokal claude-mem komplett entfernt (76 MB Daten); 13 `.remember`-Ordner gelöscht, 4 bleiben bis User committet (v18-kpi, v18-hyserve/modules, herdr-app, memocove); sccs-`.remember` offen; Codex/agy-CLI-Tool-Check läuft.
|
|
7
|
+
## 16:43 | main
|
|
8
|
+
Codex-Plugin 1.0.6 → `doctor.extra_plugins`; lokal via `sccs doctor install`; Antigravity-Skill (`~/.claude/skills/antigravity/`, Gemini-only, Diff-Review getestet) geschrieben, verteilt via `sccs sync --commit --push`; CAO optional; offen: Fix `optimize`-Marktplatz-Bug (`sccs/doctor/installer.py:2081`) als 2.72.1; Task-Delegation-Modus Skill ungetestet.
|
|
9
|
+
|
|
10
|
+
## 16:51-17:59 | main
|
|
11
|
+
sccs 2.72.1 fertig (Abhängigkeiten-Fix bei abgelehnter Deinstallation + Mehrfach-Scopes; 6 neue Tests; 2163 grün; ruff/mypy sauber); Commit mit Datenschutzprüfung gestagtem Diff → `fe4bb13`; 2.72.2 Sicherheitsfix ab 16:55: Importer-Pfadvalidierung (`zip_path`), `bundled_skill.target`-Check (doctor/schema.py), `invocation`-Validierung (installer.py), 2 Tests auf Temp-Home umgestellt, 3 neue Importer-Security-Tests; Nachfrage für eigene npx-Tools/marketplace_source; lokal committet → `5cae4d0`; 2192 grün; Push GitLab/GitHub angefragt + bestätigt, Remotes geprüft, noch nicht gepusht; Tag + PyPI bei User.
|
|
@@ -6,7 +6,7 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
|
|
|
6
6
|
|
|
7
7
|
**sccs** (SkillsCommandsConfigsSync) is a unified YAML-configured bidirectional synchronization tool for Claude Code files and optional shell configurations.
|
|
8
8
|
|
|
9
|
-
**Version**: 2.
|
|
9
|
+
**Version**: 2.74.0
|
|
10
10
|
|
|
11
11
|
### Key Features
|
|
12
12
|
|
|
@@ -22,19 +22,20 @@ incident behind it and the exact edge cases are in [`docs/development/maintainer
|
|
|
22
22
|
**read the matching section there before changing that code**. The release
|
|
23
23
|
history lives in `RELEASE_NOTES.md`.
|
|
24
24
|
|
|
25
|
-
- **Trust ceilings** (`deploy/install.py`, `doctor/schema.py`): deploy without a config writes only inside default category roots; `post_install` runs only the tool's own binary, and only bundled steps skip the prompt — the same holds for npx `invocation` (head `npx`/`npm` only), `marketplace_source` and skill-package `source`; `bundled_skill.target` is one `~/.claude/skills/<name>` folder; a bundle's `zip_path` stays inside the staging dir. → [Security ceilings](docs/development/maintainer-notes.md#security-ceilings)
|
|
25
|
+
- **Trust ceilings** (`deploy/install.py`, `doctor/schema.py`): deploy without a config writes only inside default category roots; `post_install` runs only the tool's own binary, and only bundled steps skip the prompt — the same holds for npx `invocation` (head `npx`/`npm` only), `marketplace_source` and skill-package `source`; `bundled_skill.target` is one `~/.claude/skills/<name>` folder; a bundle's `zip_path` stays inside the staging dir; `version_args` (the unattended `doctor check` probe) is one version switch only. → [Security ceilings](docs/development/maintainer-notes.md#security-ceilings)
|
|
26
26
|
- **Optional npx tools** (`doctor/installer.py`): optional + absent = INFO, never installed or updated without `--with-optional`; a missing marketplace *with* a source is added by a runnable action, test via `execute_plan`. → [Doctor](docs/development/maintainer-notes.md#doctor)
|
|
27
27
|
- **Retired plugins** (`doctor/installer.py`, `doctor/schema.py`): `retired_plugins` are removed by plain `optimize`, each confirmed; retired entries leave `effective_plugins()`; a marketplace goes only when nothing uses it; project-scope uninstalls run in the project folder. → [Doctor](docs/development/maintainer-notes.md#doctor)
|
|
28
28
|
- **Mirror parity** (`doctor/mirror.py`): macOS only — any other platform resolves to `off`; the source never receives install/remove actions, a mirror never writes `inventory.yaml`; removals only under `optimize --strict` + `cleanup: true`; never `brew bundle cleanup` or bare `fisher update`. → [Doctor](docs/development/maintainer-notes.md#doctor)
|
|
29
29
|
- **Shared config** (`doctor/shared_config.py`): `sccs_config` is bundled; adoption always asks (`always_confirm`, skipped without a terminal), pulls, backs up, validates, rolls back; never merges; `~/.config/sccs` is never a deploy root. → [Doctor](docs/development/maintainer-notes.md#doctor)
|
|
30
30
|
- **Host-local repo path** (`config/loader.py`): `local.yaml` carries `repository.path` only and is never synced; `save_config` never writes the effective path back; adoption pins the current path before replacing `config.yaml`. → [Doctor](docs/development/maintainer-notes.md#doctor)
|
|
31
|
-
- **Skill packages** (`doctor/skill_packages.py`): membership comes from `~/.agents/.skill-lock.json`, not a glob; `--copy` is load-bearing. → [Doctor](docs/development/maintainer-notes.md#doctor)
|
|
31
|
+
- **Skill packages** (`doctor/skill_packages.py`): membership comes from `~/.agents/.skill-lock.json`, not a glob; `--copy` is load-bearing; a retired package loses only lock-named skills, each removal confirmed and verified afterwards (`skills remove` always exits 0). → [Doctor](docs/development/maintainer-notes.md#doctor)
|
|
32
32
|
- **CAO provider patches** (`doctor/cao.py`): no provider source is ever bundled; `anchor_lost` is reported, never repaired; edits accumulate per file. → [Doctor](docs/development/maintainer-notes.md#doctor)
|
|
33
33
|
- **Profiles / statusline presets** (`doctor/profiles.py`, `doctor/statusline.py`): `installable_npx_tools()` is profile-aware, `effective_npx_tools()` stays blind; installers run as `bash <file>`, never `curl | bash`. → [Doctor](docs/development/maintainer-notes.md#doctor)
|
|
34
34
|
- **Fish coverage / statusline sync** (`config/defaults.py`, `sync/settings.py`): every fish file in exactly one enabled category; `item_pattern` is `statusline*` without the dot; `superseded_patterns` match by pattern. → [Sync categories](docs/development/maintainer-notes.md#sync-categories)
|
|
35
35
|
- **Deployment profiles** (`sccs/deploy/`): `pre_existing` is sticky, foreign targets are never written or removed, ownership is keyed on the resolved target path, `RECEIPT_VERSION` is 2. → [Deploy](docs/development/maintainer-notes.md#deploy)
|
|
36
36
|
- **Codex / Pi / OpenCode** (`sccs/integrations/`, `convert/`): skills go to `~/.agents/skills/`, hooks to `~/.codex/hooks.json` and never into `export-all`; `--replace-foreign` is not implied by `--overwrite`; the two frontmatter parsers stay different. → [Integrations](docs/development/maintainer-notes.md#integrations-codex-pi-opencode)
|
|
37
37
|
- **Capacity probe** (`sccs/capacity/`): never invent a number; `unknown` ≠ `tight`; Gemini tight → Codex, never Antigravity on Claude. → [Capacity](docs/development/maintainer-notes.md#capacity)
|
|
38
|
+
- **Supervisor wrappers** (`sccs/supervisor/`): the model is the only editable token and is validated (pattern, no leading `-`, `gemini-` for Antigravity); templates are read-only only; adoption inserts marker lines textually; `set-model` refreshes the hash only when it matched; plugin and unrecognised agents are never written; auth files presence-only; the core never imports Textual. → [Supervisor wrappers](docs/development/maintainer-notes.md#supervisor-wrappers-v2730)
|
|
38
39
|
- **Converter literals** (`convert/`): fish single-quoted values stay literal; escape backticks and `$(` in double-quoted targets. → [Converters](docs/development/maintainer-notes.md#converters)
|
|
39
40
|
|
|
40
41
|
## Commands
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.5
|
|
2
2
|
Name: sccs
|
|
3
|
-
Version: 2.
|
|
3
|
+
Version: 2.74.0
|
|
4
4
|
Summary: SkillsCommandsConfigsSync - Unified YAML-configured synchronization for Claude Code files
|
|
5
5
|
Project-URL: Homepage, https://github.com/equitania/sccs
|
|
6
6
|
Project-URL: Repository, https://github.com/equitania/sccs
|
|
@@ -29,6 +29,7 @@ Requires-Dist: pydantic<3.0.0,>=2.0.0
|
|
|
29
29
|
Requires-Dist: pyyaml<7.0.0,>=6.0
|
|
30
30
|
Requires-Dist: questionary<3.0.0,>=2.0.0
|
|
31
31
|
Requires-Dist: rich<16.0.0,>=13.0.0
|
|
32
|
+
Requires-Dist: textual<9,>=8.2
|
|
32
33
|
Provides-Extra: dev
|
|
33
34
|
Requires-Dist: bandit>=1.9.0; extra == 'dev'
|
|
34
35
|
Requires-Dist: mypy<3.0.0,>=2.4.0; extra == 'dev'
|
|
@@ -45,7 +46,7 @@ Description-Content-Type: text/markdown
|
|
|
45
46
|
|
|
46
47
|
> **Language / Sprache**: [Deutsch](#deutsche-dokumentation) | [English](#english-documentation)
|
|
47
48
|
|
|
48
|
-
**Version:** 2.
|
|
49
|
+
**Version:** 2.74.0 · **Lizenz / License:** AGPL-3.0 · **Python:** ≥ 3.11
|
|
49
50
|
|
|
50
51
|
---
|
|
51
52
|
|
|
@@ -64,6 +65,7 @@ SCCS ist ein YAML-konfiguriertes, bidirektionales Synchronisierungswerkzeug für
|
|
|
64
65
|
- 🖥️ **Die Statuszeile reist mit** *(v2.66.0)* — die Kategorie `claude_statusline` deckt jetzt beide Hälften ab: das Skript unter `~/.claude/` **und** den `statusLine`-Eintrag in `settings.json`, der darauf zeigt. Vorher wurde nur `statusline.sh` erfasst, während Claude Codes eigenes `/statusline`-Setup nach `statusline-command.sh` schreibt — die Datei, die auf einem echten Rechner die Statuszeile *ist*, wurde nie eingesammelt, und jeder andere Host lief weiter mit dem, was seine eigene `settings.json` zufällig sagte. macOS und Linux bekommen `bash ~/.claude/statusline-command.sh`, Windows den PowerShell-Port. Ein Eintrag, den SCCS nicht selbst geschrieben hat, bleibt unangetastet: nur Werte aus früheren SCCS-Versionen (`superseded_patterns`) werden gehoben, und ein solcher Austausch wird gemeldet → [docs/usage/categories.md](docs/usage/categories.md#die-statuszeile-über-mehrere-rechner)
|
|
65
66
|
- 🪝 **Codex-Hooks-Export** *(v2.59.0)* — `sccs integrations codex export-hooks` überträgt die Hook-Einträge aus `~/.claude/settings.json` nach `~/.codex/hooks.json`. Zehn gemeinsame Events werden übertragen, alles andere entfällt mit Warnung. Skripte bleiben in `~/.claude/hooks/`, nur die Kommandos werden gemerged — non-destruktiv: Einträge, die SCCS nicht geschrieben hat, bleiben unangetastet. Codex vertraut Hooks über einen Hash, deshalb nach dem Export `/hooks` in Codex ausführen. Bewusst nicht Teil von `export-all` — Hooks führen bei jedem Tool-Aufruf Code aus → [docs/usage/codex.md](docs/usage/codex.md#hooks-exportieren)
|
|
66
67
|
- 📉 **Kapazitätssonde** *(v2.60.0)* — `sccs capacity [--json] [--offline]` zeigt, wie viel Plan-Kontingent Claude Code, Codex und Antigravity noch haben, damit ein Orchestrator nach Kapazität statt nach Bauchgefühl verteilt. Codex wird kostenlos aus den eigenen Session-Dateien gelesen (nur so frisch wie die letzte Sitzung, wird bei Überalterung markiert), Antigravity live über `agy -p "/usage"`, Claude Code ehrlich als „nicht auslesbar, Reserve" — es werden keine Zahlen erfunden. Antigravity rechnet Gemini getrennt von den durchgereichten Claude/GPT-Modellen ab: Ist Gemini knapp, ist der Ersatzprüfer **Codex**, nicht Antigravity mit Claude-Modell — sonst prüft Anthropic sich selbst → [docs/usage/capacity.md](docs/usage/capacity.md)
|
|
68
|
+
- 🧭 **Agent-Supervisor einrichten** *(v2.73.0)* — `sccs agents` konfiguriert Claude Code als Supervisor, der Reviews und zweite Meinungen an Codex, Antigravity (Gemini) und OpenCode (Kimi) abgibt: Vollbild-Ansicht und `list`/`tools`/`models`/`set-model`/`new`/`adopt` (alle mit `--json`), schreibgeschützte Wrapper-Vorlagen, Modellwechsel per Liste, vorhandene Wrapper werden einmalig übernommen und aus dem Export der anderen Werkzeuge herausgehalten. Reine Konfiguration — es startet keine Agentenarbeit. Die Vollbild-Ansicht nutzt [Textual](https://textual.textualize.io), das seit 2.73.0 eine reguläre Abhängigkeit ist → [docs/usage/agents.md](docs/usage/agents.md) · Schritt für Schritt: [usage/multi-provider-agents.de.md](usage/multi-provider-agents.de.md)
|
|
67
69
|
- 🩺 **System & Plugin Health-Check** (`sccs doctor`) für Node.js, `claude` CLI, Claude-Plugins, npm-Helper-Tools, Browser-Bundles und Filesystem-Permissions — mit gezielten Reparatur-Plänen → [docs/usage/doctor.md](docs/usage/doctor.md)
|
|
68
70
|
- 🧹 **`sccs doctor optimize`** *(v2.30.0)* — Ein-Schuss-Optimierungslauf: install + update + Drift-Warnung für Plugins/MCP-Server außerhalb der Spec. Mit `--strict` werden Uninstall-Actions per Confirm gequeut.
|
|
69
71
|
- 🔧 **`doctor.disallowed_hooks`** *(v2.31.0)* — Substring-Patterns entfernen unerwünschte Hooks aus `~/.claude/settings.json` nach jedem doctor-Pass (z.B. Hooks die von npx-Tools re-injiziert werden). Mit Backup, idempotent.
|
|
@@ -146,6 +148,7 @@ sccs doctor check # System & Plugin Health prüfen
|
|
|
146
148
|
| [docs/usage/opencode.md](docs/usage/opencode.md) | OpenCode-Integration: Agents/Commands exportieren, Modell-Mapping, MCP-Merge |
|
|
147
149
|
| [docs/usage/pi.md](docs/usage/pi.md) | Pi-Integration: Skills/Agents/Commands nach Pi (pi.dev) exportieren |
|
|
148
150
|
| [docs/usage/codex.md](docs/usage/codex.md) | OpenAI-Codex-Integration: Skills/Agents/Commands in die Codex CLI exportieren |
|
|
151
|
+
| [docs/usage/agents.md](docs/usage/agents.md) | `sccs agents`: Supervisor-Wrapper für Codex, Gemini und OpenCode einrichten und nutzen |
|
|
149
152
|
| [docs/usage/platforms.md](docs/usage/platforms.md) | Windows/PowerShell-Support, Fish→PowerShell-Konvertierung |
|
|
150
153
|
| [docs/usage/cli-reference.md](docs/usage/cli-reference.md) | Vollständige CLI-Befehlsreferenz |
|
|
151
154
|
| [docs/architecture.md](docs/architecture.md) | Modul-Layout, Test-Setup, Quality-Gate |
|
|
@@ -171,6 +174,8 @@ SCCS is a YAML-configured, bidirectional synchronization tool for Claude Code fi
|
|
|
171
174
|
- 📊 **Statusline presets** *(v2.58.0)* — `sccs statusline list|use|install|show` picks which statusline Claude Code runs; bundled are `builtin` (`~/.claude/statusline.sh`) and [`claude-code-statusline`](https://github.com/glauberlima/claude-code-statusline). Needed because a parked extension takes its own statusline with it. Installing goes through a temp file and `bash <file>` — **never** `curl … | bash` through a shell; install URLs are https- and host-allowlist-checked. The multi-MB binary lands in the sync excludes via `managed_paths`. `sccs doctor check` reports the chosen preset in every state — `installed · in use`, `installed · not in use`, or `MISSING` with the install command *(v2.58.1)*. `~` and `$HOME` in `statusLine.command` are expanded before the check, and `statusline use|install` record the choice in `config.yaml` *(v2.58.2)* → [docs/usage/profiles.md](docs/usage/profiles.md#statusline-1)
|
|
172
175
|
- 🖥️ **The statusline travels** *(v2.66.0)* — the `claude_statusline` category now covers both halves: the script under `~/.claude/` **and** the `statusLine` entry in `settings.json` pointing at it. Previously only `statusline.sh` was picked up, while Claude Code's own `/statusline` setup writes `statusline-command.sh` — the file that on a real machine *is* the statusline was never collected, and every other host kept running whatever its own `settings.json` happened to say. macOS and Linux get `bash ~/.claude/statusline-command.sh`, Windows the PowerShell port. An entry SCCS did not write itself is left alone: only values from earlier SCCS releases (`superseded_patterns`) are lifted, and such a replacement is reported → [docs/usage/categories.md](docs/usage/categories.md#the-statusline-across-machines)
|
|
173
176
|
- 🪝 **Codex hooks export** *(v2.59.0)* — `sccs integrations codex export-hooks` transfers the hook entries from `~/.claude/settings.json` into `~/.codex/hooks.json`. Ten shared events transfer, everything else is dropped with a warning. Scripts stay in `~/.claude/hooks/`, only the commands get merged — non-destructive: entries SCCS did not write are left alone. Codex trusts hooks by a hash, so run `/hooks` in Codex after exporting. Deliberately not part of `export-all` — hooks execute code on every tool call → [docs/usage/codex.md](docs/usage/codex.md#exporting-hooks)
|
|
177
|
+
- 📉 **Capacity probe** *(v2.60.0)* — `sccs capacity [--json] [--offline]` shows how much plan quota Claude Code, Codex and Antigravity have left, so an orchestrator routes by capacity instead of gut feeling. Codex is read for free from its own session files (only as fresh as the last session, flagged when stale), Antigravity live via `agy -p "/usage"`, Claude Code honestly as „not readable, reserve“ — no numbers are invented. Antigravity bills Gemini separately from the passed-through Claude/GPT models: when Gemini is tight, the substitute reviewer is **Codex**, not Antigravity on a Claude model — otherwise Anthropic reviews itself → [docs/usage/capacity.md](docs/usage/capacity.md)
|
|
178
|
+
- 🧭 **Set up the agent supervisor** *(v2.73.0)* — `sccs agents` configures Claude Code as a supervisor that hands reviews and second opinions to Codex, Antigravity (Gemini) and OpenCode (Kimi): a full-screen view plus `list`/`tools`/`models`/`set-model`/`new`/`adopt` (all with `--json`), read-only wrapper templates, model changes from a list, existing wrappers adopted once and kept out of the other tools' exports. Pure configuration — it starts no agent work. The full-screen view uses [Textual](https://textual.textualize.io), a regular dependency since 2.73.0 → [docs/usage/agents.md](docs/usage/agents.md) · Step by step: [usage/multi-provider-agents.en.md](usage/multi-provider-agents.en.md)
|
|
174
179
|
- 🩺 **System & plugin health check** (`sccs doctor`) for Node.js, the `claude` CLI, Claude plugins, npm helper tools, browser bundles and filesystem permissions — with surgical repair plans → [docs/usage/doctor.md](docs/usage/doctor.md)
|
|
175
180
|
- 🧹 **`sccs doctor optimize`** *(v2.30.0)* — one-shot optimisation pass: install + update + drift warning for plugins/MCP servers outside the spec. With `--strict`, uninstall actions are queued per confirm.
|
|
176
181
|
- 🔧 **`doctor.disallowed_hooks`** *(v2.31.0)* — substring patterns strip unwanted hooks from `~/.claude/settings.json` after every doctor pass (e.g. hooks re-injected by npx tools). Backup-aware, idempotent.
|
|
@@ -252,6 +257,7 @@ sccs doctor check # System & plugin health check
|
|
|
252
257
|
| [docs/usage/opencode.md](docs/usage/opencode.md) | OpenCode integration: export agents/commands, model mapping, MCP merge |
|
|
253
258
|
| [docs/usage/pi.md](docs/usage/pi.md) | Pi integration: export skills/agents/commands to Pi (pi.dev) |
|
|
254
259
|
| [docs/usage/codex.md](docs/usage/codex.md) | OpenAI Codex integration: export skills/agents/commands to the Codex CLI |
|
|
260
|
+
| [docs/usage/agents.md](docs/usage/agents.md) | `sccs agents`: set up and use supervisor wrappers for Codex, Gemini and OpenCode |
|
|
255
261
|
| [docs/usage/platforms.md](docs/usage/platforms.md) | Windows/PowerShell support, Fish→PowerShell conversion |
|
|
256
262
|
| [docs/usage/cli-reference.md](docs/usage/cli-reference.md) | Full CLI command reference |
|
|
257
263
|
| [docs/architecture.md](docs/architecture.md) | Module layout, test setup, quality gate |
|
|
@@ -4,7 +4,7 @@
|
|
|
4
4
|
|
|
5
5
|
> **Language / Sprache**: [Deutsch](#deutsche-dokumentation) | [English](#english-documentation)
|
|
6
6
|
|
|
7
|
-
**Version:** 2.
|
|
7
|
+
**Version:** 2.74.0 · **Lizenz / License:** AGPL-3.0 · **Python:** ≥ 3.11
|
|
8
8
|
|
|
9
9
|
---
|
|
10
10
|
|
|
@@ -23,6 +23,7 @@ SCCS ist ein YAML-konfiguriertes, bidirektionales Synchronisierungswerkzeug für
|
|
|
23
23
|
- 🖥️ **Die Statuszeile reist mit** *(v2.66.0)* — die Kategorie `claude_statusline` deckt jetzt beide Hälften ab: das Skript unter `~/.claude/` **und** den `statusLine`-Eintrag in `settings.json`, der darauf zeigt. Vorher wurde nur `statusline.sh` erfasst, während Claude Codes eigenes `/statusline`-Setup nach `statusline-command.sh` schreibt — die Datei, die auf einem echten Rechner die Statuszeile *ist*, wurde nie eingesammelt, und jeder andere Host lief weiter mit dem, was seine eigene `settings.json` zufällig sagte. macOS und Linux bekommen `bash ~/.claude/statusline-command.sh`, Windows den PowerShell-Port. Ein Eintrag, den SCCS nicht selbst geschrieben hat, bleibt unangetastet: nur Werte aus früheren SCCS-Versionen (`superseded_patterns`) werden gehoben, und ein solcher Austausch wird gemeldet → [docs/usage/categories.md](docs/usage/categories.md#die-statuszeile-über-mehrere-rechner)
|
|
24
24
|
- 🪝 **Codex-Hooks-Export** *(v2.59.0)* — `sccs integrations codex export-hooks` überträgt die Hook-Einträge aus `~/.claude/settings.json` nach `~/.codex/hooks.json`. Zehn gemeinsame Events werden übertragen, alles andere entfällt mit Warnung. Skripte bleiben in `~/.claude/hooks/`, nur die Kommandos werden gemerged — non-destruktiv: Einträge, die SCCS nicht geschrieben hat, bleiben unangetastet. Codex vertraut Hooks über einen Hash, deshalb nach dem Export `/hooks` in Codex ausführen. Bewusst nicht Teil von `export-all` — Hooks führen bei jedem Tool-Aufruf Code aus → [docs/usage/codex.md](docs/usage/codex.md#hooks-exportieren)
|
|
25
25
|
- 📉 **Kapazitätssonde** *(v2.60.0)* — `sccs capacity [--json] [--offline]` zeigt, wie viel Plan-Kontingent Claude Code, Codex und Antigravity noch haben, damit ein Orchestrator nach Kapazität statt nach Bauchgefühl verteilt. Codex wird kostenlos aus den eigenen Session-Dateien gelesen (nur so frisch wie die letzte Sitzung, wird bei Überalterung markiert), Antigravity live über `agy -p "/usage"`, Claude Code ehrlich als „nicht auslesbar, Reserve" — es werden keine Zahlen erfunden. Antigravity rechnet Gemini getrennt von den durchgereichten Claude/GPT-Modellen ab: Ist Gemini knapp, ist der Ersatzprüfer **Codex**, nicht Antigravity mit Claude-Modell — sonst prüft Anthropic sich selbst → [docs/usage/capacity.md](docs/usage/capacity.md)
|
|
26
|
+
- 🧭 **Agent-Supervisor einrichten** *(v2.73.0)* — `sccs agents` konfiguriert Claude Code als Supervisor, der Reviews und zweite Meinungen an Codex, Antigravity (Gemini) und OpenCode (Kimi) abgibt: Vollbild-Ansicht und `list`/`tools`/`models`/`set-model`/`new`/`adopt` (alle mit `--json`), schreibgeschützte Wrapper-Vorlagen, Modellwechsel per Liste, vorhandene Wrapper werden einmalig übernommen und aus dem Export der anderen Werkzeuge herausgehalten. Reine Konfiguration — es startet keine Agentenarbeit. Die Vollbild-Ansicht nutzt [Textual](https://textual.textualize.io), das seit 2.73.0 eine reguläre Abhängigkeit ist → [docs/usage/agents.md](docs/usage/agents.md) · Schritt für Schritt: [usage/multi-provider-agents.de.md](usage/multi-provider-agents.de.md)
|
|
26
27
|
- 🩺 **System & Plugin Health-Check** (`sccs doctor`) für Node.js, `claude` CLI, Claude-Plugins, npm-Helper-Tools, Browser-Bundles und Filesystem-Permissions — mit gezielten Reparatur-Plänen → [docs/usage/doctor.md](docs/usage/doctor.md)
|
|
27
28
|
- 🧹 **`sccs doctor optimize`** *(v2.30.0)* — Ein-Schuss-Optimierungslauf: install + update + Drift-Warnung für Plugins/MCP-Server außerhalb der Spec. Mit `--strict` werden Uninstall-Actions per Confirm gequeut.
|
|
28
29
|
- 🔧 **`doctor.disallowed_hooks`** *(v2.31.0)* — Substring-Patterns entfernen unerwünschte Hooks aus `~/.claude/settings.json` nach jedem doctor-Pass (z.B. Hooks die von npx-Tools re-injiziert werden). Mit Backup, idempotent.
|
|
@@ -105,6 +106,7 @@ sccs doctor check # System & Plugin Health prüfen
|
|
|
105
106
|
| [docs/usage/opencode.md](docs/usage/opencode.md) | OpenCode-Integration: Agents/Commands exportieren, Modell-Mapping, MCP-Merge |
|
|
106
107
|
| [docs/usage/pi.md](docs/usage/pi.md) | Pi-Integration: Skills/Agents/Commands nach Pi (pi.dev) exportieren |
|
|
107
108
|
| [docs/usage/codex.md](docs/usage/codex.md) | OpenAI-Codex-Integration: Skills/Agents/Commands in die Codex CLI exportieren |
|
|
109
|
+
| [docs/usage/agents.md](docs/usage/agents.md) | `sccs agents`: Supervisor-Wrapper für Codex, Gemini und OpenCode einrichten und nutzen |
|
|
108
110
|
| [docs/usage/platforms.md](docs/usage/platforms.md) | Windows/PowerShell-Support, Fish→PowerShell-Konvertierung |
|
|
109
111
|
| [docs/usage/cli-reference.md](docs/usage/cli-reference.md) | Vollständige CLI-Befehlsreferenz |
|
|
110
112
|
| [docs/architecture.md](docs/architecture.md) | Modul-Layout, Test-Setup, Quality-Gate |
|
|
@@ -130,6 +132,8 @@ SCCS is a YAML-configured, bidirectional synchronization tool for Claude Code fi
|
|
|
130
132
|
- 📊 **Statusline presets** *(v2.58.0)* — `sccs statusline list|use|install|show` picks which statusline Claude Code runs; bundled are `builtin` (`~/.claude/statusline.sh`) and [`claude-code-statusline`](https://github.com/glauberlima/claude-code-statusline). Needed because a parked extension takes its own statusline with it. Installing goes through a temp file and `bash <file>` — **never** `curl … | bash` through a shell; install URLs are https- and host-allowlist-checked. The multi-MB binary lands in the sync excludes via `managed_paths`. `sccs doctor check` reports the chosen preset in every state — `installed · in use`, `installed · not in use`, or `MISSING` with the install command *(v2.58.1)*. `~` and `$HOME` in `statusLine.command` are expanded before the check, and `statusline use|install` record the choice in `config.yaml` *(v2.58.2)* → [docs/usage/profiles.md](docs/usage/profiles.md#statusline-1)
|
|
131
133
|
- 🖥️ **The statusline travels** *(v2.66.0)* — the `claude_statusline` category now covers both halves: the script under `~/.claude/` **and** the `statusLine` entry in `settings.json` pointing at it. Previously only `statusline.sh` was picked up, while Claude Code's own `/statusline` setup writes `statusline-command.sh` — the file that on a real machine *is* the statusline was never collected, and every other host kept running whatever its own `settings.json` happened to say. macOS and Linux get `bash ~/.claude/statusline-command.sh`, Windows the PowerShell port. An entry SCCS did not write itself is left alone: only values from earlier SCCS releases (`superseded_patterns`) are lifted, and such a replacement is reported → [docs/usage/categories.md](docs/usage/categories.md#the-statusline-across-machines)
|
|
132
134
|
- 🪝 **Codex hooks export** *(v2.59.0)* — `sccs integrations codex export-hooks` transfers the hook entries from `~/.claude/settings.json` into `~/.codex/hooks.json`. Ten shared events transfer, everything else is dropped with a warning. Scripts stay in `~/.claude/hooks/`, only the commands get merged — non-destructive: entries SCCS did not write are left alone. Codex trusts hooks by a hash, so run `/hooks` in Codex after exporting. Deliberately not part of `export-all` — hooks execute code on every tool call → [docs/usage/codex.md](docs/usage/codex.md#exporting-hooks)
|
|
135
|
+
- 📉 **Capacity probe** *(v2.60.0)* — `sccs capacity [--json] [--offline]` shows how much plan quota Claude Code, Codex and Antigravity have left, so an orchestrator routes by capacity instead of gut feeling. Codex is read for free from its own session files (only as fresh as the last session, flagged when stale), Antigravity live via `agy -p "/usage"`, Claude Code honestly as „not readable, reserve“ — no numbers are invented. Antigravity bills Gemini separately from the passed-through Claude/GPT models: when Gemini is tight, the substitute reviewer is **Codex**, not Antigravity on a Claude model — otherwise Anthropic reviews itself → [docs/usage/capacity.md](docs/usage/capacity.md)
|
|
136
|
+
- 🧭 **Set up the agent supervisor** *(v2.73.0)* — `sccs agents` configures Claude Code as a supervisor that hands reviews and second opinions to Codex, Antigravity (Gemini) and OpenCode (Kimi): a full-screen view plus `list`/`tools`/`models`/`set-model`/`new`/`adopt` (all with `--json`), read-only wrapper templates, model changes from a list, existing wrappers adopted once and kept out of the other tools' exports. Pure configuration — it starts no agent work. The full-screen view uses [Textual](https://textual.textualize.io), a regular dependency since 2.73.0 → [docs/usage/agents.md](docs/usage/agents.md) · Step by step: [usage/multi-provider-agents.en.md](usage/multi-provider-agents.en.md)
|
|
133
137
|
- 🩺 **System & plugin health check** (`sccs doctor`) for Node.js, the `claude` CLI, Claude plugins, npm helper tools, browser bundles and filesystem permissions — with surgical repair plans → [docs/usage/doctor.md](docs/usage/doctor.md)
|
|
134
138
|
- 🧹 **`sccs doctor optimize`** *(v2.30.0)* — one-shot optimisation pass: install + update + drift warning for plugins/MCP servers outside the spec. With `--strict`, uninstall actions are queued per confirm.
|
|
135
139
|
- 🔧 **`doctor.disallowed_hooks`** *(v2.31.0)* — substring patterns strip unwanted hooks from `~/.claude/settings.json` after every doctor pass (e.g. hooks re-injected by npx tools). Backup-aware, idempotent.
|
|
@@ -211,6 +215,7 @@ sccs doctor check # System & plugin health check
|
|
|
211
215
|
| [docs/usage/opencode.md](docs/usage/opencode.md) | OpenCode integration: export agents/commands, model mapping, MCP merge |
|
|
212
216
|
| [docs/usage/pi.md](docs/usage/pi.md) | Pi integration: export skills/agents/commands to Pi (pi.dev) |
|
|
213
217
|
| [docs/usage/codex.md](docs/usage/codex.md) | OpenAI Codex integration: export skills/agents/commands to the Codex CLI |
|
|
218
|
+
| [docs/usage/agents.md](docs/usage/agents.md) | `sccs agents`: set up and use supervisor wrappers for Codex, Gemini and OpenCode |
|
|
214
219
|
| [docs/usage/platforms.md](docs/usage/platforms.md) | Windows/PowerShell support, Fish→PowerShell conversion |
|
|
215
220
|
| [docs/usage/cli-reference.md](docs/usage/cli-reference.md) | Full CLI command reference |
|
|
216
221
|
| [docs/architecture.md](docs/architecture.md) | Module layout, test setup, quality gate |
|
|
@@ -1,5 +1,27 @@
|
|
|
1
1
|
# Release Notes
|
|
2
2
|
|
|
3
|
+
## Version 2.74.0 (16.11.2026)
|
|
4
|
+
|
|
5
|
+
### Added
|
|
6
|
+
|
|
7
|
+
- **`doctor.retired_skill_packages`: drop a skill package on every host.** A package installed through the `skills` CLI (bundled: `hyperframes`) could be installed and refreshed, but never removed again. Listing it under `doctor.retired_skill_packages` makes `sccs doctor check` show a yellow row with the number of skills still present, and a plain `sccs doctor optimize` removes them after a confirm prompt — no `--strict` needed, as with `retired_plugins`. `install` and `update` leave a retired package alone. Only skills the `skills` lock file (`~/.agents/.skill-lock.json`) attributes to the package are removed; without a lock nothing is touched, so a private skill that shares a name with a package skill stays. Because `skills remove` reports success even when it matched nothing, a second step checks that the folders are gone and fails otherwise. A package cannot be listed in `skill_packages` and `retired_skill_packages` at once. Skills parked by `sccs profile off` are not on disk in `~/.claude/skills` and stay in the parking area.
|
|
8
|
+
|
|
9
|
+
## Version 2.73.0 (09.11.2026)
|
|
10
|
+
|
|
11
|
+
### Added
|
|
12
|
+
|
|
13
|
+
- **`sccs agents`: set up the multi-CLI reviewers from sccs.** A Claude Code session delegates reviews and second opinions to Codex, Antigravity (Gemini) and OpenCode (Kimi) through thin wrapper agents in `~/.claude/agents/`. `sccs agents` opens a full-screen view (Textual, now a regular dependency) with one row per agent: tool, version, login state, model and mode; `list`, `tools`, `models`, `set-model`, `new` and `adopt` do the same on the command line, each with `--json`. It is configuration only: nothing here starts, dispatches or monitors agent work.
|
|
14
|
+
- **New wrappers are read-only.** Three bundled templates (Antigravity, OpenCode, Codex), each with that tool's read-only mechanism; there is no write-capable variant. Pi has no tested template yet and is refused with a hint. A model name is validated before it is written, because the wrapper runs its command line verbatim: a restricted character set, no leading `-`, and the `gemini-` prefix for Antigravity, so a Gemini reviewer never ends up on a Claude or GPT model.
|
|
15
|
+
- **Adopt the hand-written wrappers once.** `sccs agents adopt` adds a small marker to the frontmatter and leaves the rest of the file byte for byte unchanged (CRLF and BOM included); `set-model` then edits only the model in the command line. Plugin agents and agents matching no template are shown and never written.
|
|
16
|
+
- **Wrappers stay out of the other tools' export.** Creating or adopting a wrapper adds its name to `opencode.exclude`, `pi.exclude` and `codex.exclude` once, with a backup of `config.yaml`.
|
|
17
|
+
- **Usage guide and approach.** [docs/usage/agents.md](docs/usage/agents.md) (German and English, also reachable with `?` in the view, together with the approach document on `Tab`) covers setup, working with the supervisor, timeouts and the known traps; the reasoning is in the new architecture section „Supervisor and wrapper agents“.
|
|
18
|
+
|
|
19
|
+
## Version 2.72.3 (09.11.2026)
|
|
20
|
+
|
|
21
|
+
### Security
|
|
22
|
+
|
|
23
|
+
- **The version probe of a tool runs only a version switch.** `sccs doctor check` fills its Version column by running a tool's `detect_command` with its `version_args`, unattended, on every check. Both came straight from `config.yaml`, which reaches every host through the sync repo, so `detect_command: python3` with `version_args: ["-c", "…"]` ran arbitrary code on every host without a prompt — the gap the 2.72.2 ceilings closed for install steps, left open for detection. `version_args` now accepts exactly one of `--version`, `-version`, `-V`, `-v`, `version` (every bundled tool uses `--version`), for npx tools and CLI tools alike, and an npx tool's `detect_command` passes the same name check as a CLI tool's. A config that sets anything else fails validation with the allowed list. Found by a cross-model review (Codex), 09.11.2026.
|
|
24
|
+
|
|
3
25
|
## Version 2.72.2 (08.11.2026)
|
|
4
26
|
|
|
5
27
|
### Security
|
|
@@ -34,6 +34,7 @@ sccs/
|
|
|
34
34
|
│ ├── runner.py # Subprocess-Wrapper (no shell, no sudo)
|
|
35
35
|
│ ├── managed.py # Doctor-managed Files (vom Sync ausgenommen)
|
|
36
36
|
│ └── state.py # Doctor-State-Persistenz
|
|
37
|
+
├── supervisor/ # Agent-Wrapper: Erkennung, Vorlagen, Modelländerung, Vollbild-Ansicht
|
|
37
38
|
├── transfer/ # Export/Import-Modul
|
|
38
39
|
│ ├── manifest.py # ZIP-Manifest (Pydantic)
|
|
39
40
|
│ ├── exporter.py # Scan + ZIP-Erstellung
|
|
@@ -58,6 +59,17 @@ sccs/
|
|
|
58
59
|
└── platform.py # Plattformerkennung
|
|
59
60
|
```
|
|
60
61
|
|
|
62
|
+
### Supervisor und Wrapper-Agenten
|
|
63
|
+
|
|
64
|
+
Hintergrund zu [`sccs agents`](usage/agents.md). Die Arbeitsweise: **Claude Code ist der Supervisor**. Er gibt Reviews und zweite Meinungen an andere KI-CLIs ab (Codex, Antigravity/Gemini, OpenCode/Kimi) und prüft jeden zurückgekommenen Befund selbst am Code.
|
|
65
|
+
|
|
66
|
+
- **Warum Claude Supervisor ist.** Dort läuft die Sitzung mit dem Projektkontext, den Regeln und den Werkzeugen. Eine zweite Orchestrierungsschicht daneben müsste all das nachbauen. Ein Befund eines anderen Modells ist eine Behauptung; erst der Abgleich mit dem Code macht daraus einen Fehler oder eben nicht. Deshalb wird jeder Befund ausdrücklich als ungeprüft gekennzeichnet und vom Supervisor einzeln geprüft.
|
|
67
|
+
- **Warum dünne Haiku-Wrapper mit nur `Bash` und `Read`.** Ein Wrapper soll einen CLI-Aufruf weiterreichen und die Antwort zurückgeben, nicht selbst urteilen. Das kleinste Modell reicht dafür, und ohne Schreibwerkzeuge kann der Wrapper selbst nichts verändern. Der Befehl steht fest in der Wrapper-Datei; sccs ändert daran nur das Modell, und der Modellname wird vorher streng geprüft (Zeichenmuster, kein führendes `-`), weil der Wrapper den Befehl wörtlich über Bash ausführt.
|
|
68
|
+
- **Warum nicht CAO.** Ein eigener Orchestrator mit Sitzungsverwaltung, Provider-Patches und Handoff-Protokoll ist für „drei Prüfer parallel, einer wertet aus“ mehr Maschinerie als nötig. Claude Code kann Unteragenten im Hintergrund starten, wird benachrichtigt und besitzt bereits den Kontext. Was bleibt, ist reine Konfiguration, und die ist Sache von sccs.
|
|
69
|
+
- **Warum Prüfer nur lesen.** Ein Prüfer, der den Code verändern darf, verändert das, was er prüfen soll. Jedes Werkzeug erzwingt Read-only auf seine Weise: Antigravity mit `--mode plan` und einer Erlaubnisliste lesender Befehle, OpenCode mit einem eingebetteten Agenten `claude-ro` (keine Schreibwerkzeuge, Shell nur für lesende Befehle, keine Web-Abrufe, keine Unteragenten), Codex mit `codex exec -s read-only --ephemeral`. Nie `--dangerously-skip-permissions`, nie OpenCode `--auto`, nie `--mode accept-edits`. Eine schreibfähige Variante gibt es bewusst nicht.
|
|
70
|
+
- **Warum ein Gemini-Prüfer nie auf ein Claude-Modell ausweicht.** Antigravity verkauft neben Gemini auch Claude- und GPT-Modelle weiter und rechnet Gemini getrennt ab. Fällt der Gemini-Prüfer auf ein Claude-Modell zurück, prüft Anthropic sich selbst: Das ist Selbstreview mit Umweg, nicht die zweite Modellfamilie, um die es geht. Darum verlangt die Vorlage für Antigravity ein Modell mit dem Präfix `gemini-`, und der Wrapper bricht ab, statt auszuweichen. Dieselbe Regel steht hinter `sccs capacity`: Ist Gemini knapp, springt der Prüfer zu Codex, nie zu Antigravity auf einem Claude-Modell.
|
|
71
|
+
- **Aufbau im Code.** `sccs/supervisor/` hält Erkennung und Wrapper-Modell (ohne Textual-Import, damit `--json` überall läuft); `sccs agents …` in der CLI ruft nur diesen Kern; die Vollbild-Ansicht (Textual) ruft ebenfalls nur den Kern.
|
|
72
|
+
|
|
61
73
|
### Entwicklung
|
|
62
74
|
|
|
63
75
|
```bash
|
|
@@ -72,7 +84,7 @@ mypy sccs/ # Typenprüfung
|
|
|
72
84
|
bandit -r sccs/ # Security-Scan
|
|
73
85
|
```
|
|
74
86
|
|
|
75
|
-
Querverweise: [doctor.md](usage/doctor.md), [sync.md](usage/sync.md), [memory-bridge.md](usage/memory-bridge.md)
|
|
87
|
+
Querverweise: [agents.md](usage/agents.md), [doctor.md](usage/doctor.md), [sync.md](usage/sync.md), [memory-bridge.md](usage/memory-bridge.md)
|
|
76
88
|
|
|
77
89
|
---
|
|
78
90
|
|
|
@@ -104,6 +116,7 @@ sccs/
|
|
|
104
116
|
│ ├── runner.py # Subprocess wrapper (no shell, no sudo)
|
|
105
117
|
│ ├── managed.py # Doctor-managed files (excluded from sync)
|
|
106
118
|
│ └── state.py # Doctor state persistence
|
|
119
|
+
├── supervisor/ # Agent wrappers: detection, templates, model change, full-screen view
|
|
107
120
|
├── transfer/ # Export/Import module
|
|
108
121
|
│ ├── manifest.py # ZIP manifest (Pydantic)
|
|
109
122
|
│ ├── exporter.py # Scan + ZIP creation
|
|
@@ -128,6 +141,17 @@ sccs/
|
|
|
128
141
|
└── platform.py # Platform detection
|
|
129
142
|
```
|
|
130
143
|
|
|
144
|
+
### Supervisor and wrapper agents
|
|
145
|
+
|
|
146
|
+
Background to [`sccs agents`](usage/agents.md). The working model: **Claude Code is the supervisor**. It hands reviews and second opinions to other AI CLIs (Codex, Antigravity/Gemini, OpenCode/Kimi) and checks every returned finding against the code itself.
|
|
147
|
+
|
|
148
|
+
- **Why Claude is the supervisor.** That is where the session with the project context, the rules and the tools lives. A second orchestration layer next to it would have to rebuild all of that. Another model's finding is a claim; only the comparison with the code turns it into a bug or not. So every finding is explicitly marked unverified and checked one by one by the supervisor.
|
|
149
|
+
- **Why thin Haiku wrappers with only `Bash` and `Read`.** A wrapper forwards one CLI call and returns the answer; it does not judge. The smallest model is enough for that, and without write tools the wrapper cannot change anything itself. The command is fixed in the wrapper file; sccs changes only the model in it, and the model name is validated strictly beforehand (character pattern, no leading `-`) because the wrapper runs the command verbatim through Bash.
|
|
150
|
+
- **Why not CAO.** A separate orchestrator with session management, provider patches and a handoff protocol is more machinery than „three reviewers in parallel, one evaluates“ needs. Claude Code can start subagents in the background, gets notified and already owns the context. What remains is plain configuration, which is sccs' job.
|
|
151
|
+
- **Why reviewers are read-only.** A reviewer allowed to change the code changes what it is supposed to review. Each tool enforces read-only its own way: Antigravity with `--mode plan` and an allowlist of reading commands, OpenCode with an inline agent `claude-ro` (no write tools, shell for reading commands only, no web fetches, no sub-agents), Codex with `codex exec -s read-only --ephemeral`. Never `--dangerously-skip-permissions`, never OpenCode `--auto`, never `--mode accept-edits`. A write-capable variant is left out on purpose.
|
|
152
|
+
- **Why a Gemini reviewer never falls back to a Claude model.** Besides Gemini, Antigravity resells Claude and GPT models and bills Gemini separately. If the Gemini reviewer fell back to a Claude model, Anthropic would review itself: self-review by detour, not the second model family the setup is for. So the Antigravity template requires a model with the `gemini-` prefix and the wrapper stops instead of switching. The same rule stands behind `sccs capacity`: when Gemini is tight, the reviewer moves to Codex, never to Antigravity on a Claude model.
|
|
153
|
+
- **How it is built.** `sccs/supervisor/` holds detection and the wrapper model (it never imports Textual, so `--json` works everywhere); `sccs agents …` in the CLI only calls that core; the full-screen view (Textual) also calls only the core.
|
|
154
|
+
|
|
131
155
|
### Development
|
|
132
156
|
|
|
133
157
|
```bash
|
|
@@ -142,4 +166,4 @@ mypy sccs/ # Type checking
|
|
|
142
166
|
bandit -r sccs/ # Security scan
|
|
143
167
|
```
|
|
144
168
|
|
|
145
|
-
See also: [doctor.md](usage/doctor.md), [sync.md](usage/sync.md), [memory-bridge.md](usage/memory-bridge.md)
|
|
169
|
+
See also: [agents.md](usage/agents.md), [doctor.md](usage/doctor.md), [sync.md](usage/sync.md), [memory-bridge.md](usage/memory-bridge.md)
|
|
@@ -60,6 +60,14 @@ sccs/
|
|
|
60
60
|
│ ├── zsh_templates.py # zshrc/conveniences/README templates
|
|
61
61
|
│ ├── claude_to_codex.py # CC → Codex conversion rules (model/effort map, sandbox_mode)
|
|
62
62
|
│ └── toml_write.py # Minimal TOML emitter for Codex agent files
|
|
63
|
+
├── supervisor/ # `sccs agents`: configure supervisor wrappers (configuration only)
|
|
64
|
+
│ ├── tools.py # Detect claude/codex/opencode/agy/pi: installed, version, login presence, default model
|
|
65
|
+
│ ├── models.py # Per-tool model lists, MODEL_PATTERN validation (gemini-* for agy)
|
|
66
|
+
│ ├── wrappers.py # Scan ~/.claude/agents, recognise the thin wrapper shapes (agy, opencode, codex)
|
|
67
|
+
│ ├── render.py # Create new wrapper agents from templates/ (agy.md, opencode.md, codex.md; read-only)
|
|
68
|
+
│ ├── edit.py # Textual set-model and adopt (BOM/CRLF kept byte for byte)
|
|
69
|
+
│ ├── tui.py # Full-screen view (Textual; tool calls run in thread workers)
|
|
70
|
+
│ └── templates/ # Wrapper templates (no Pi template yet)
|
|
63
71
|
├── docs/
|
|
64
72
|
│ └── generator.py # Hub README auto-generation
|
|
65
73
|
├── git/
|
|
@@ -15,6 +15,10 @@ The release history — what each version added — lives in `RELEASE_NOTES.md`.
|
|
|
15
15
|
|
|
16
16
|
`config.yaml` reaches every host through the sync repo, so it is treated like `post_install`: a step SCCS ships runs unattended, a step that only the config defines asks (`--yes` still answers). That covers npx `invocation` (`_is_bundled_npx_tool`: same name, invocation and `bundled_skill` as a default), `marketplace_source` (`_is_bundled_marketplace_source`) and skill-package `source` (`skill_packages._is_builtin`, source only — target and lock paths are host details). The npx `invocation` head is pinned to `npx`/`npm` in the schema; before, `bash -c …` validated. `BundledSkillSpec.target` must be `~/.claude/skills/<name>` and `package_subpath` relative without `..`, because `_sync_bundled_skill` replaces the target with `rmtree`; at execute time the source must resolve under `npm root -g`, the target must not be a symlink and must resolve under `~/.claude/skills`, and a non-empty folder without `SKILL.md` is refused, never deleted. In a transfer bundle the manifest's `zip_path` is validated on its own (`importer._staged_source`) — `_safe_extract` only checks ZIP entry names, and an absolute `zip_path` replaced the staging prefix, so `import`/`deploy install` could copy any local file (an SSH key) into a synced folder. Found by a security audit, 08.11.2026.
|
|
17
17
|
|
|
18
|
+
### The version probe is a step too (v2.72.3)
|
|
19
|
+
|
|
20
|
+
`_resolve_version` in both `NpxToolDetector` and `CliToolDetector` runs `[detect_command or name, *version_args]` on every `doctor check`, with no prompt and no bundled-check — detection is not an install, so the 2.72.2 "config-sourced steps ask" rule never reached it. The schema therefore pins `version_args` to a single switch from `_VERSION_ARGS_ALLOWED` and validates `NpxToolSpec.detect_command` like `CliToolSpec.detect_command`. Do not widen the allowlist to free-form args, and do not add a new unattended probe that takes argv from config; a probe that needs more belongs in `DEFAULT_NPX_TOOLS`/`BUILTIN_CLI_TOOLS`.
|
|
21
|
+
|
|
18
22
|
## Doctor
|
|
19
23
|
|
|
20
24
|
### Optional npx tools (v2.68.0)
|
|
@@ -33,6 +37,8 @@ The release history — what each version added — lives in `RELEASE_NOTES.md`.
|
|
|
33
37
|
|
|
34
38
|
`sccs/doctor/skill_packages.py` manages skill bundles fetched by the `skills` CLI (`npx skills add <owner/repo>`), opt-in via `doctor.skill_packages` — bundled is `hyperframes`. Three rules. **The `skills` lock file (`~/.agents/.skill-lock.json`) decides membership**, not a glob: HyperFrames' 20 skills share no prefix (`figma`, `slideshow`), so sync excludes, profile parking and the detector all read entries by `source`. The bundled `known_skills` list is only a fallback — and for sync excludes only for an *enabled* package, so a private `figma` keeps syncing on a host without HyperFrames. Lock-derived names are excluded regardless of opt-in, like `gsd-*`. **Install and update are one command** (`skills add … --agent claude-code --skill '*' --yes --copy`), verified to replace each skill directory wholesale; `--copy` is load-bearing, because the CLI symlinks by default and a symlink is skipped by the sync scan and parked as a bare link. **No update signal is invented**: the lock stores a folder hash, not a version, so `check` shows the lock date and `update` refreshes unconditionally. Missing/symlinked skills fail `check`; an unloadable `SKILL.md` (`skill_limits`) is reported yellow but does not, since only upstream can fix it. Profile `hyperframes` carries `skill_packages`, and `installable_skill_packages()` mirrors `installable_npx_tools()`.
|
|
35
39
|
|
|
40
|
+
**Retiring a package (v2.74.0)**: `doctor.retired_skill_packages` → `SkillPackageDetector.get_retired_statuses()` → `skill_package_remove_actions()`, spliced into `build_optimize_plan` next to the retired plugins (no `--strict`). Retired statuses travel in the same `skill_packages` list with `retired=True`, so `install`/`update` must keep skipping them. Three rules. **Lock only**: the removal list is what the lock attributes to the source *and* exists in `target_dir` — never `known_skills`, which would delete a private `figma` on a host without the package. **`--skill` once per name**: `skills remove a b --agent claude-code` matches nothing in skills 1.9.0 and still exits 0. **Verify afterwards**: the CLI exits 0 whatever happened, so a second `python_callable` action (`requires_dependencies_executed`) raises when a folder is left; without it a failed removal reads as success. The removal keeps `auto_confirm=False`. Known side effect, accepted: `skills remove` also deletes a same-named folder in `~/.agents/skills/` (the Codex export copy), even with `--agent claude-code`. A package parked by `sccs profile off` has nothing in `target_dir` and produces no status — the parked copy stays.
|
|
41
|
+
|
|
36
42
|
### CAO provider patches (v2.64.0)
|
|
37
43
|
|
|
38
44
|
`sccs/doctor/cao.py` re-applies the extra agent providers that a `cao update` wipes — CAO's registry is a hard-coded if/elif chain, not a plugin point, so a provider must be edited into the *installed* package at six sites. **Since v2.67.1 no provider is bundled**: pi (`pi_cli`), the only one, was retired from the fleet, so `DEFAULT_CAO_PROVIDERS` is empty and the mechanism serves only providers declared in `doctor.cao_providers`. Four rules. **The provider source is never bundled here** (a private `source_dir`) — this package goes to PyPI and GitHub, and only the mechanism belongs in it; the v2.60.0 removal of the CAO agent profiles set that line. **The opt-in is the pairing**, not a config flag: a row appears only when an installed CAO and the synced source both exist. **`anchor_lost` is reported, never repaired** — a half-applied patch leaves a package that imports but cannot launch, so a moved anchor gets a pointer to re-derive it in `DEFAULT_CAO_PROVIDERS`. And **edits accumulate per file, not per site**: `providers/manager.py` carries two sites, and deriving both from the same on-disk text makes the second write discard the first (found against a real install, not by the fixture). Verification completes before the first write, so a broken spec changes nothing.
|
|
@@ -113,6 +119,19 @@ Pi/Codex/Claude Code share the agentskills.io SKILL.md format and all three **si
|
|
|
113
119
|
|
|
114
120
|
`sccs capacity [--json] [--offline]` (`sccs/capacity/`) reports remaining plan quota per agent CLI. Rules that must not be softened: **no number is ever invented** — Claude Code has no on-disk quota cache, so it is reported as `assumed`, not estimated; `_status()` keeps `unknown` distinct from `tight`, because missing data is not evidence of exhaustion and conflating them pushes work onto the billed API for no reason; `is_codex_exhausted()` requires null windows **and** an empty credit balance, since null windows alone are merely absent knowledge; and **Antigravity bills Gemini separately from the Claude/GPT models it resells**, so when Gemini is tight the fallback reviewer is **Codex, never Antigravity on a Claude model** — that would turn cross-review into self-review. Routing lives in code (`derive_routing`), not in a prompt, so it is testable and identical on every host.
|
|
115
121
|
|
|
122
|
+
### Supervisor wrappers (v2.73.0)
|
|
123
|
+
|
|
124
|
+
`sccs agents` (`sccs/supervisor/`) configures the wrapper agents through which a Claude Code session delegates reviews to other CLIs. It is configuration only; the rules below keep a wrapper from becoming a way to run arbitrary commands or to write where it should only read.
|
|
125
|
+
|
|
126
|
+
- **The model is the only editable token, and it is validated.** The Haiku wrapper runs its bash block verbatim, so a model name is a command-injection surface. It must match `^[A-Za-z0-9._:/~][A-Za-z0-9._:/~-]*$` (no leading `-`, no shell metacharacters); the Antigravity template additionally requires the prefix `gemini-`. A Gemini wrapper on a `claude-*` or `gpt-*` model would turn cross-review into self-review — the same reasoning as in `sccs capacity`. The wrapper's own frontmatter (`model: haiku`, `tools: Bash, Read`) is fixed and never offered for editing. Agent names for `new` match `^[a-z0-9][a-z0-9-]*$`; a path-like name or an existing file is refused before anything is written.
|
|
127
|
+
- **Templates are read-only only.** Antigravity `--mode plan --settings "$AGY_RO"`, OpenCode `--agent claude-ro` with the inline `OPENCODE_CONFIG_CONTENT`, Codex `codex exec -s read-only --ephemeral`. Never `--dangerously-skip-permissions`, never OpenCode `--auto`, never `--mode accept-edits`; do not add a write-capable template without an explicit decision. Pi has no template and `new pi` exits non-zero.
|
|
128
|
+
- **Adoption inserts, it does not re-dump.** `adopt` inserts the `sccs:` marker lines textually into the existing frontmatter; it never parses and re-serialises it, so comments, key order, CRLF and a BOM survive and the body stays byte-identical. The stored hash covers the body with the model token replaced by a placeholder, so a model change never looks like a hand edit and any other edit does.
|
|
129
|
+
- **`set-model` refreshes the hash only when it matched.** The edit is surgical (the one model token in the command line). If the file had been hand-edited since sccs wrote it, the edit still happens but the stored hash is left alone, so the file keeps showing as edited (`mode_label` appends `(edited)` in the CLI table and the view alike). sccs never overwrites an existing agent file: `new` refuses an existing name, and the only writes to an existing file are the surgical model edit and the marker insertion of `adopt`.
|
|
130
|
+
- **Config failures never leave a half-written state.** `new` and `adopt` read `config.yaml` before writing the agent file; if recording the excludes still fails, `new` removes the file it just created and `adopt` restores the original bytes, and the error is a `SupervisorError` (JSON error envelope, TUI notice).
|
|
131
|
+
- **Plugin agents and unrecognised agents are never written.** An agent matching no template shape, and every plugin agent (`codex:*`, `gsd-*`, …), is listed and left untouched.
|
|
132
|
+
- **Auth files are checked for presence only.** Their contents are never read into any output; an undeterminable value is shown as `unknown` (models: `unverified`), never guessed; an absent auth file is `unknown`, not `no`, because credentials may live in a keyring or in environment variables.
|
|
133
|
+
- **The core never imports Textual.** `sccs agents list --json` and every other non-TUI command work without it (a test enforces this); the TUI imports the core, not the reverse. All subprocesses (version probes, model lists) go through `sccs/doctor/runner.py` with a timeout and `stdin=DEVNULL`.
|
|
134
|
+
|
|
116
135
|
## Converters
|
|
117
136
|
|
|
118
137
|
### Converter literal-value safety (v2.56.0)
|