/gaia-history

/gaia history

Display a table of all GAIA benchmark runs that have been stored in the gaia-runs AgentDB memory namespace.

Usage

/gaia history
/gaia history --limit=10
/gaia history --model=claude-sonnet-4-6
/gaia history --level=1

Options

Flag	Default	Description
--limit	20	Maximum rows to show
--level	all	Filter by GAIA level
--model	all	Filter by model ID
--since	all	ISO-8601 date; show runs after this date
--format	table	table or json

What this does

1. Queries the gaia-runs namespace: npx @claude-flow/cli@latest memory list --namespace gaia-runs
2. Parses each stored value (JSON summary blobs saved by /gaia run).
3. Renders a sorted table (newest first):

Run ID               Date        Level  Model               Q    Pass%   Cost ($)  Mean turns
-------------------  ----------  -----  ------------------  ---  ------  --------  ----------
run-20260527-1423    2026-05-27  L1     claude-sonnet-4-6   53   20.8%   $0.47     4.2
run-20260524-0912    2026-05-24  L1     claude-haiku-4-5    53    9.4%   $0.09     3.1
run-20260522-1537    2026-05-22  L1     claude-sonnet-4-5   53   18.2%   $0.41     3.9

4. Highlights the best pass-rate row.
5. Shows a trend line (pass-rate over time) if 3+ runs exist.

Run record schema (stored per run)

{
  "run_id": "run-20260527-1423",
  "timestamp": "2026-05-27T14:23:00Z",
  "level": 1,
  "model": "claude-sonnet-4-6",
  "total": 53,
  "passed": 11,
  "pass_rate": 0.208,
  "est_cost_usd": 0.47,
  "mean_turns": 4.2,
  "adrs": ["ADR-133","ADR-135"],
  "git_sha": "abc1234"
}

Steps Claude should follow

1. Run npx @claude-flow/cli@latest memory list --namespace gaia-runs --limit $LIMIT
2. Parse each entry value as JSON.
3. Apply --level and --model filters if provided.
4. Sort by timestamp descending.
5. Render the table and highlight the row with the highest pass_rate.
6. If 3+ runs exist, print a simple ASCII trend chart of pass_rate over time.