Skill Sublation
A governance skill for Hermes, Codex and Claude Code: skill changes pass candidates, validation, review and a user decision.
- 278 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The skill keeps ledger and candidate files in the home directory and, with user approval, changes an agent's working skills.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/skill-sublationDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The skill keeps ledger and candidate files in the home directory and, with user approval, changes an agent's working skills.
This is third-party code. Review the repository files before installing.
What it does
Skill Sublation, version 5.0, governs the life of an agent's skills. A formal skill is not edited on the fly: experience is first recorded as an observation, then as a candidate, validated, reviewed by independent seats and promoted only after the user decides, with an observation window afterwards. Version 5.0 adds a multi-agent collaboration panel, agent auto-wakeup intents (off by default), skill classification and shadow routing suggestions. The suggestions execute nothing. The skill triggers only when the user names `sublation` or `扬弃技能` explicitly.
Who it is for. People who maintain their own skill library for several agents and want every change to leave a trace and pass review.
Good fit when
- A skill needs changes after a failed run, but the live version must not change without validation.
- Three agents split the roles of builder, independent verifier and reviewer.
- Before publication a skill set needs cleaning, with reasons for each change recorded.
- You want shadow suggestions of which skill fits a task, without automatic invocation.
Not a fit when
- You need a one-off edit of a single skill without a ledger or review.
- An agent should edit a working skill on its own without a human: the framework forbids that.
- You expect ready automation with cron or launchd: the package installs neither.
Example request
Sublation: log an observation for the skill that gave a wrong result, create a candidate and show a validation plan without touching the live version.Limitations
The README and description are in Chinese, with no Russian or English documentation. Installation is manual: the unpacked ZIP with the `skill-sublation` directory goes into the agent's skills directory. A full check needs Python 3.10 or newer and `jsonschema`, and a build shows `Ran 271 tests` and `OK`. A successful build does not mean installed or reviewed. By default three different agents fill the roles; the user must enable single-agent mode explicitly.
How to disable. Delete the `skill-sublation` directory from the agent's skills directory and check in a new session that the skill is gone. Clean candidates and records under `~/.hermes/sublation/` separately.
FAQ
Does the skill edit live skills itself?
No. The live path stays read-only, and writes happen only after explicit approval of a promotion or rollback.
Is automatic agent wakeup on?
No, it is off by default. The package installs and registers no watcher, launchd, cron or real host adapter.
Where do candidates live?
Per the skill description in `~/.hermes/sublation/candidates/<skill>/<candidate-id>/`.
How do I run the package tests?
Install `requirements-test.txt` and run `PYTHONDONTWRITEBYTECODE=1 PYTHONPATH=scripts python3 -B -m unittest discover -s scripts -p 'test_*.py'`.
What does the collaboration panel do?
It holds review seats locally per project; a seat in the panel does not grant review authority, and a queued intent does not mean an agent replied.
Discussion
No questions or reviews yet.
Related
SkillsA meta-skill that splits a task into stages, picks a profile and runs multi-agent work in Codex and Claude Code.
A catalog of nearly two hundred skills: code delivery, review, debugging, AI governance, product work and agent coordination.
31 skills for Claude Code and Codex: cross-vendor review, task handoff on low quota, a morning brief and a nightly retrospective
Runs sub-agents from Markdown files in the .agents folder on backends such as Codex, Claude Code, Cursor CLI and Gemini.
Custom skills bundle for Hermes Agent
hermes-custom-skills
A monorepo of five skills for Hermes Agent: agent security hardening, skill auto-updates, skill search, cross-border expansion and Tencent Yuanbao chat
Skills Constitution makes the agent check the skill index and search for a capability before refusing
README in short
The README sets the boundaries: the panel does not replace review, routing suggestions execute nothing, autowakeup is off and writes to live skills need approval.