Research a topic and file the result — sweep what we already know (knowledge base, crm, and slack when connected), go to the web for current and external background, then write a clear sourced brief and FILE it into the knowledge base with update_knowledge. Labels what came from the web versus our own notes. Use for "research X and save it", "look into X and write it up", "find out what's known about X and file it", "write up a brief on X".
The owner's morning brief — inbound updates awaiting acknowledgment (via the rundown tool), today's meetings and due-or-overdue reminders from the CRM and the real calendar, the emails worth seeing from Gmail, and the Slack threads that need their eyes — each source folded in when connected and stitched into one short digest. Use for "daily rundown", "what's on today", "morning brief", "catch me up".
Autonomously harden the context agent (or any agent in this repo) — derive probes from the agent's own instructions, run them against the live container, judge the responses and tool calls, and edit the agent until it reliably does what it says it does. Use when the user wants to test, harden, stress-test, or improve an agent's reliability WITHOUT naming a specific change ("make sure context is solid", "harden this agent", "probe it for failure modes"). For a user-directed change use extend-agent; to run the committed eval suite use eval-and-improve.
Repo-wide consistency sweep for @context — diff docs against code (the context agent and its providers wired as documented, every env var documented, every path real, every script behaving as advertised), auto-fix mechanical drift, and flag anything bigger. Use when the user wants to review the repo for consistency, prep a public release, or clean up drift after a refactor ("review the repo", "is everything consistent before we ship", "run the drift sweep"). This is a deliberate full sweep, not a quick lint.
Sweep the knowledge base for what needs attention — stubs with no real content, specs whose status is stale, decisions that were made but never written down, and pages missing from the index — and return a short "what needs attention" list. Read-only, so it reports rather than fixing anything itself. Use for "what in my knowledge base needs attention", "review my knowledge base", "what's stale or unfiled", "knowledge base health check", "tidy up my knowledge".
Run this repo's eval suite (python -m evals), diagnose every failing case, and fix in scope (agent instructions, or the eval case when its assertion was wrong) until all cases pass. Use whenever the user wants to run the evals, check for regressions, or fix a red eval suite ("run the evals", "why is a case failing", "make the eval suite green"). Targets the committed cases in evals/cases.py; for ad-hoc probing of one agent's live behavior use improve-agent instead.
Brief the owner before a meeting or conversation with a person, company, or project — sweep what we already know (crm + knowledge, plus slack, calendar, and gmail when connected), anchor to any upcoming meeting with them, and for people or orgs we don't know, widen to a web search for public background. Use for "prep me for my meeting with X", "brief me on X before the call", "what do we know about X".
User-driven loop to change the context agent (or any agent in this repo) — add a source/tool/MCP server, add a capability (knowledge base, memory, sub-agent, scheduled task), refine its instructions, or fix a specific known bug, verifying each change against the live container. Use whenever the user names a concrete change ("add a Google Drive source", "make the daily rundown include the calendar", "the agent keeps doing Y, fix it"). For autonomous hardening with no specific change in mind use improve-agent.