forgeplan activity-stats
forgeplan activity-stats summarises the activity log: one row per tool name, with how many times it ran, how many times it failed, and how long it took (median / 95th percentile / total). It reads the same log files as forgeplan activity but rolls them up so you can answer “where did my session time go?” in a single glance instead of scrolling through hundreds of individual calls.
This is the CLI version of forgeplan_activity_stats - both produce the same numbers.
A note on percentiles: p50 is the median (half of calls were faster, half slower); p95 is the slow tail (95% of calls were faster than this). High p95 with low p50 means most calls are fast but a few are dragging.
When to use
Section titled “When to use”- Start of a debugging session - find the slowest or most-called tool over the last 24 hours.
- A user reports “it feels slow” and you need a starting point for investigation.
- Pre-release check - compare error counts to last week to spot regressions.
- After a long session, confirm that the destructive-operation count matches what you expected.
When NOT to use
Section titled “When NOT to use”- You need to see individual calls (timestamps, arguments) - use
forgeplan activityinstead, it does not aggregate. - You want a methodology gate (block work until something passes) - stats are observability, not validation.
forgeplan activity-stats [OPTIONS]Options
Section titled “Options” --since-hours <SINCE_HOURS> Time window in hours (1..=720, default 24) [default: 24] --json Output as JSON for machine consumption -h, --help Print help -V, --version Print versionExamples
Section titled “Examples”Example 1: Default 24-hour summary
Section titled “Example 1: Default 24-hour summary”forgeplan activity-statsPrints rows ordered by total time spent (highest first) - the most time-consuming tool appears at the top. Quick answer to “where did my session time go?”.
Example 2: Full-month view
Section titled “Example 2: Full-month view”forgeplan activity-stats --since-hours 720Use this for monthly retrospectives or when investigating a slow workflow that may have started days ago. 720 hours (30 days) is the maximum window allowed.
Example 3: Find slow tools (p95 > 1 second)
Section titled “Example 3: Find slow tools (p95 > 1 second)”forgeplan activity-stats --json | jq '.stats[] | select(.p95_ms > 1000)'Filters to tools where the slow-tail (p95) duration exceeds 1000 ms (1 second). A typical first cut when investigating “the agent feels slow today” - these are the tools dragging the experience down.
How it fits the workflow
Section titled “How it fits the workflow”Pair with forgeplan activity for a two-step investigation: run activity-stats first to find the slow or failing tool, then run activity --tool <name> to see the individual calls. If errors are the issue, narrow further with --status tool_err.
See also
Section titled “See also”forgeplan_activity_stats- MCP equivalentforgeplan activity- entry-level drill-downforgeplan undo-last- pair with stats to reverse misfires- CLI overview