Mods: cortex-bar

mods/cortex-bar is a Claude Code mod (plugin version 0.9.0) that shows this session's Cortex calls above the prompt: a stacked bar with one color per category, the latest call, what the results cost and saved in tokens, how often the agent opened what a code lookup ranked, and where /cs stands. Each cortex tool row in the transcript gets a tag in its category's color.

◆ cortex ████████████████████████████████████████ 9 calls · last plan_quality 1.3s ✓
■ session 2  ■ knowledge 1  ■ memory 1  ■ code 3  ■ quality 1  ■ tasks 1
tokens ~3.3k in · ~9.8k saved est.  │ lookups cortex 3 · grep 9 · 25% · 1 fallback  │ hit@1 1/3 · …
/cs ● session  ● recall  ● changes  ● tasks  │ ● build  ● typecheck  ○ lint  ○ report

Under each cortex tool row:

● cortex-hub - cortex_code_search (MCP)(query: "hybrid search")
■ code · 233ms ✓ · ~700 tok · 3 files ~13k · used #2

It needs Claude Code 2.1.287 or later, in the terminal or the desktop app.

Install

For one session, from a clone:

claude --plugin-dir mods/cortex-bar

Or from this repo as a plugin marketplace:

claude plugin marketplace add lktiep/cortex-hub
claude plugin install cortex-bar@cortex-hub      # --scope project to share it with the repo

Then start a new conversation.

In VS Code, Antigravity and Cursor

The Claude Code extension's chat panel does not draw mod UI. Run Claude Code in the IDE's terminal instead: set "claudeCode.useTerminal": true in the IDE's settings, and the band, the tags and the pane all show.

Commands

Command	What it does
/cortex-bar	Show or hide the band and the row tags (remembered)
/cortex-bar on, /cortex-bar off	Show or hide explicitly
/cortex-bar reset	Clear the call counts and the measures
/cortex-calls	Open a pane with the measures, a per-tool table and the last 50 calls

What it measures

Everything is counted in this Claude Code session. The hub's cortex_tool_stats counts on the server, with its own baseline, so the two differ.

Measure	Meaning
Tokens in	Characters of each cortex result divided by 4: what the result added to the context
Saved	Per lookup: the size of the files it pointed at (each capped at 25k tokens), less the result's own cost, less the files the agent opened anyway. An upper bound
Used / missed	The agent opened one of the files a lookup named (used #n is the rank), or opened other files instead (missed)
hit@k	The share of judged lookups whose best opened file was ranked k or better
Lookups vs grep	Cortex code lookups against the agent's own searches that cortex could have answered. A fallback is such a search within three calls of a cortex lookup
Quality	From each tool's own output: result counts, top score, risk_level, plan score

The /cs checklist reads the same markers the hooks gate on. gate-off shows as ⚠ gates off.

The mod sends nothing anywhere. Arguments whose name looks like a credential (key, token, secret, password, authorization) are never shown.

Colors

Category	Tools
session	session_start, session_end, changes, health, list_repos
knowledge	knowledge_search, knowledge_store
memory	memory_search, memory_store, memory_delete
code	code_search, code_context, code_impact, code_reindex, cypher, detect_changes
quality	quality_report, plan_quality, tool_stats
tasks	task_*
other	any cortex tool not listed

Past sessions

The same measures, read back from Claude Code's transcripts:

node mods/cortex-bar/scripts/lookup-audit.mjs             # this directory's newest 5 sessions
node mods/cortex-bar/scripts/lookup-audit.mjs --last 0    # all of them

The full description is in the mod's README.