advisor-live

skill
Security Audit
Fail
Health Warn
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Fail
  • exec() — Shell command execution in hooks/register.tsx
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

Claude Code mod: a live side pane of the advisor's calls, latency and moments

README.md

advisor-live

A Claude Code mod that shows the advisor's activity live while you work, in a side pane. The pane opens with the session, or with /advisor-live, drawn in btop's style: a link box with the main model, the advisor and the turn's running time; a latency braille graph of recent calls with their max and average; a moments box metering the three moments the advisor wakes up (before the plan locks, same error twice, before calling it done); and a calls list, newest first, down to the pane's foot. A short pane, or the desktop app's minimized widget, folds it to three lines. /clear and /resume start the counts over.

Start a session with an advisor, for example claude --model sonnet --advisor fable.

The pane

The advisor-live pane after five calls from an Opus advisor to Sonnet 5.5: a latency graph, the four moments metered, and the calls listed newest first

From the top, here after five calls from an Opus advisor working with Sonnet 5.5:

  • link: the main model on the left, the advisor on the right. Below them, the main model's state (○ idle between turns, a spinner and the running time while it builds) and how long ago the advisor last answered.
  • latency: shown once there are calls. A braille graph of recent advisor calls, with the slowest and the average.
  • moments: how many calls fell under each moment the advisor is meant to wake up, as a meter, a share and a count. A check row joins them once a call lands in between.
  • calls: one row per call, newest first: the moment, how long it took and how long ago.

Idle advisor

An advisor that is set but not called through three turns in a row that used tools reads idle · N turns in the pane's link box, and the count keeps going until a call arrives. This usually means the advisor is less capable than the main model, for example Fable under Opus 5.5, and so never activates. Pick a stronger advisor, or drop it. Switching either model starts the count over.


Switch either model mid-session with the built-in /model and /advisor; the pane follows. Both save the choice as your default for new sessions.

A mod can't read which advisor the session is using, so the pane works it out: at start from --advisor on the launch command line, else the advisorModel setting; after /advisor, from what it wrote to that setting. An advisor reading Advisor means calls are arriving but nothing named the model.

The advisor runs inside the API request, so the mod sees a call only after it finishes. There is no "consulting" state while it runs. The moment each call counts under is inferred from the turn, not reported by the advisor: the same tool failing twice in a row with the same message is "same error twice", a step that ends the turn is "before calling it done", a call before any file was written is "before the plan locks", and a call in between shows as a check-in in the recent list.

Install

In a Claude Code session:

/plugin install advisor-live --marketplace iskerwin/advisor-live

From a shell:

claude plugin install advisor-live --marketplace iskerwin/advisor-live

Answer y to add the marketplace, then pick a scope.

Update

From a shell, fetch the marketplace again, then update the plugin:

claude plugin marketplace update advisor-live
claude plugin update advisor-live

Restart running sessions to load the new version.

Develop

claude plugin test .
claude plugin validate .

License

MIT

Reviews (0)

No results found