Install the current LiveKit agent skills - #108
Conversation
Replace the pre-split livekit-agents skill with the seven skills now in livekit/agent-skills (at 5d7488b), installed with `npx skills add livekit/agent-skills --agent claude-code codex --copy`. That writes each skill to .claude/skills/ and .agents/skills/, and records their sources and hashes in skills-lock.json so `npx skills update` can refresh them. This keeps the starter's skills current until the LiveKit CLI installs them itself. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01U7yszNU7t5aCHJsKLmU4hv
u9g
left a comment
u9g
left a comment
There was a problem hiding this comment.
I started reviewing the skills before realizing that this was just the entire skill repo
| @@ -0,0 +1,134 @@ | |||
| # State, effects, and evidence | |||
There was a problem hiding this comment.
are you sure that this shouldn't just go into the docs website?
There was a problem hiding this comment.
+ should we make a postinstall script in package.json that just installs lk skills globally, so they don't need to live here?
There was a problem hiding this comment.
@u9g this is just temporary. the new version of the lk CLI will auto-prompt to install skills when using lk agent init and then we can pull these out of the repo
|
|
||
| ## Evidence comes from the boundary | ||
|
|
||
| Completion needs evidence from where it actually happens: a committed write, a delivered |
There was a problem hiding this comment.
| Completion needs evidence from where it actually happens: a committed write, a delivered | |
| Action completion needs evidence from where it actually happens: a committed write, a delivered |
| ## Evidence comes from the boundary | ||
|
|
||
| Completion needs evidence from where it actually happens: a committed write, a delivered | ||
| utterance, a runtime lifecycle event. An instruction to do something, a tool that returned a string, |
There was a problem hiding this comment.
why would a tool that returned a strong or a boolean named after the action not be evidence?
|
|
||
| Keep one authoritative state object per session, supplied through the SDK's per-session user | ||
| data. Keep model-supplied facts separate from trusted identity, the clock, message and record ids, | ||
| and operation receipts. Don't recover missing state from a global "last session" variable, and |
There was a problem hiding this comment.
Don't recover missing state from a global "last session" variable
This feels like something current agents wouldn't do. Not a removal suggestion, just a thought
| Keep one authoritative state object per session, supplied through the SDK's per-session user | ||
| data. Keep model-supplied facts separate from trusted identity, the clock, message and record ids, | ||
| and operation receipts. Don't recover missing state from a global "last session" variable, and | ||
| don't manufacture a successful snapshot at export time. |
There was a problem hiding this comment.
and don't manufacture a successful snapshot at export time.
I don't know what this means
| | "No, that's everyone" | Confirm the collection as recorded | Every existing entry and field | | ||
| | "No note for that one" | Record explicit absence on one entry | Its identity, title, all other entries | | ||
| | "Actually I'm not sure about that" | Retract one fact to unknown | The entry and unrelated facts | | ||
| | "It's spelled Okafor" | Change one identified entry's field | Its identity, other fields, other entries | |
There was a problem hiding this comment.
for this table, maybe something about asking for more info if something seems confusing?
| @@ -0,0 +1,47 @@ | |||
| { | |||
There was a problem hiding this comment.
probably we don't need a lock file
|
@u9g thanks for the feedback on the skills, i may improve them in the skills repo. gonna merge this as-is just to get it moving, we'll remove these from the repo once the CLI update ships |
3 participants
What
Replaces the bundled
livekit-agentsskill with the seven skills now inlivekit/agent-skills(at5d7488b):reading-livekit-docsbuilding-livekit-agentsdebugging-livekit-agentstesting-livekit-agentswriting-livekit-scenariosrunning-livekit-simulationsoperating-livekit-agentsInstalled with
npx skills add livekit/agent-skills --agent claude-code codex --skill '*' --copy, which writes each skill to both.claude/skills/and.agents/skills/(the layout this repo already used) and addsskills-lock.json, sonpx skills updatecan refresh them later. Both copies are byte-identical to upstream.Why
The bundled
livekit-agentsskill is the pre-split version, since replaced upstream by these focused skills (livekit/agent-skills#10). Left as is, it steers coding agents toward superseded guidance, such as alk agent simulateform that now just prints help. Skill installation is moving into the LiveKit CLI, but until that ships, the starter should carry the current skills.Notes for reviewers
🤖 Generated with Claude Code
https://claude.ai/code/session_01U7yszNU7t5aCHJsKLmU4hv
Generated by Claude Code