[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-82762":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":8,"htmlUrl":8,"language":9,"languages":8,"totalLinesOfCode":8,"stars":10,"forks":11,"watchers":12,"openIssues":11,"contributorsCount":11,"subscribersCount":11,"size":11,"stars1d":13,"stars7d":14,"stars30d":15,"stars90d":11,"forks30d":11,"starsTrendScore":16,"compositeScore":11,"rankGlobal":8,"rankLanguage":8,"license":17,"archived":18,"fork":18,"defaultBranch":19,"hasWiki":20,"hasPages":18,"topics":21,"createdAt":8,"pushedAt":8,"updatedAt":22,"readmeContent":23,"aiSummary":24,"trendingCount":11,"starSnapshotCount":11,"syncStatus":13,"lastSyncTime":25,"discoverSource":26},82762,"sublation","Sven-Mirana\u002Fsublation","Sven-Mirana",null,"Python",108,0,54,2,16,50,11,"MIT License",false,"main",true,[],"2026-06-12 02:04:27","# Skill Sublation (技能扬弃)\n\nA structured governance framework for AI agent skills — observe, candidate, audit, review, promote, and observe again.\n\n## What is Sublation?\n\nSublation (Aufhebung \u002F 扬弃) is a governance pipeline that turns agent execution experience into auditable skill improvements, without letting agents freely edit production skills.\n\n```\nObservation → Candidate → Audit → Review → Promotion → Observation Window\n```\n\n**Core principles:**\n- **Formal skills are read-only** — agents cannot modify active skills directly\n- **All changes go through candidates** — sandboxed copies with full audit trail\n- **Promotion requires review** — cross-agent review + user approval before merging\n- **Post-promotion safety net** — rollback, path verification, smoke test, fallback check\n\n## Quick Start\n\n```bash\n# Create an observation from a skill execution failure\npython3 scripts\u002Fobserve.py \u003Cskill-name> \\\n  --skill-path ~\u002F.hermes\u002Fskills\u002F\u003Ccategory>\u002F\u003Cskill-name> \\\n  --session \u003Csession-id> \\\n  --classification defect \\\n  --reflection-type SKILL_DEFECT \\\n  --step \"script execution\" \\\n  --evidence \"error message or observed behavior\" \\\n  --summary \"What went wrong\"\n\n# Create a candidate from the observation\npython3 scripts\u002Fcandidate.py create \u003Cskill-name> \\\n  --source-path ~\u002F.hermes\u002Fskills\u002F\u003Ccategory>\u002F\u003Cskill-name> \\\n  --candidate-type spec-patch \\\n  --agent hermes\n\n# Audit the candidate\npython3 scripts\u002Faudit.py ~\u002F.hermes\u002Fsublation\u002Fcandidates\u002F\u003Cskill>\u002F\u003Ccandidate-id>\n\n# Check system health\npython3 scripts\u002Flifecycle.py health --warn-after-days 7\n```\n\n## Key Numbers\n\n- **22 audit checks** (10 base + 12 strict) with `passed | conditional | failed` resolution\n- **Closed self-governance trail** across runtime, lifecycle, audit, cross-skill, release, and hardening candidates\n- **Production sample candidates closed** across NPL, Canghe, legal, GBrain, and briefing skills\n- **32→11 skill consolidation** via merge-driven sublation (-66%)\n- **4 candidate types**: spec-patch, script-enhance, infra-fix, tooling\n- **3 promotion modes**: human_patch, user_delegated_agent_patch, rollback\n\n## Capabilities\n\n| Capability | Description |\n|---|---|\n| Lifecycle Management | 9-state lifecycle (active→closed) with health scanning |\n| Cross-Skill Absorption | Donor→target absorption without donor modification |\n| Merge-Driven Sublation | Multi-skill consolidation with review checklist |\n| Darwin Evaluator Adapter | External evaluator integration (read-only, proposal-only) |\n| Post-Promotion Safety Net | Rollback, path verification, smoke test, fallback check |\n| Observation Window Policy | Mandatory production observation before closure |\n| Legacy Migration | Plan-based migration from v2 manifests to v3 |\n| Rights & Provenance | License tracking, expression copying audit |\n\n## Governance Trail\n\nCandidate manifests are internal runtime data and are not included in the public repo. See [CHANGELOG.md](CHANGELOG.md) for the version evolution timeline and [RELEASE-v1.0.md](RELEASE-v1.0.md) for the v1.0 release report.\n\n## Status\n\n**v1.0 — Maintenance Mode.** The framework is complete. Future changes only from real skill practice exposing cracks — no feature development for its own sake.\n\n## License\n\nMIT\n","Skill Sublation 是一个用于AI代理技能的结构化治理框架，通过观察、候选、审核、复审、提升和再次观察的过程，将代理执行经验转化为可审计的技能改进。项目核心功能包括正式技能只读、所有变更需经候选流程且带有完整审计轨迹、提升前需跨代理复审及用户批准、以及提升后的安全网措施如回滚、路径验证、冒烟测试等。适合需要严格控制AI技能更新过程，并确保每次更新都经过充分测试与审查的场景使用。该项目采用Python语言编写，遵循MIT许可协议。","2026-06-11 04:09:09","CREATED_QUERY"]