activity

This skill never installs anything, and it writes nothing anywhere unless asked. It runs one Node process from the plugin directory, makes one or more read-only calls to the configured instance, and prints the answer.

disable-model-invocation: true is deliberate, and matches dashboard. This is a question a person asks about their own data; nothing in a conversation should decide on its own to pull a copy of somebody's memory into the transcript. A disabled skill's description is not always-loaded context either, so it costs nothing until somebody types it.

The distinction that decides which command to run

There are two things this command can produce and they are not the same kind of answer.

• A listing is a claim this client makes. The plugin sends the filters, and then checks the result against them itself: it re-drops derived entries and re-truncates content, and it says so when it had to. A listing is for reading.
• An export is the record the instance holds, byte for byte. /v2/control/activity/export accepts no exclude_derived and no projection at all, so an export is never filtered and never truncated, and nothing here reshapes it. An export is for keeping.

Never describe an export as "filtered activity". If somebody asks for "an export of my non-derived lessons", they are asking for two different operations: an export is everything in scope, and the filtering they want is a listing. Say which one they got.

Reading it

node "${CLAUDE_PLUGIN_ROOT}/bin/activity.mjs"
node "${CLAUDE_PLUGIN_ROOT}/bin/activity.mjs" --exclude-derived --type lesson
node "${CLAUDE_PLUGIN_ROOT}/bin/activity.mjs" --scan --since 2026-08-01T00:00:00Z

The default is one page of the newest activity for the newest run in this data directory. --scan pages through everything that matches, oldest first. --run <id> picks a run and --all-runs asks across every run the key can see.

Filters: --type (repeatable), --since / --until as RFC3339, --agent, --user, and --exclude-derived for the entries the instance promoted for itself rather than ones a client wrote. --full returns every field untruncated; the default is five keys per entry.