Search recent sessions for recurring friction that suggests a skill is missing or an existing agentfile has a blind spot. Produces a short report of candidate skills, command improvements, or AGENTS.md additions.
Find patterns in recent session history that suggest the agentfiles are
incomplete. This is the discovery complement to /skill-review, which improves
existing artifacts. /skill-gaps finds things that don't have an artifact yet.
A gap is a repeated pattern where Jamie had to correct, explain, or manually do something that a skill or instruction could have handled. One-off corrections are not gaps. The same correction across 2-3 sessions is.
Concrete examples:
Cast a wide net, then filter. The goal is to surface candidate patterns, not to be exhaustive.
Search cass for recent sessions containing correction signals. These are phrases Jamie uses when redirecting the agent. Run several searches in parallel, scoped to the last 14-30 days:
cass search "no not that" --robot --limit 10 --days 30 --fields summary
cass search "actually" --robot --limit 10 --days 14 --fields summary
cass search "like last time" --robot --limit 10 --days 30 --fields summary
cass search "remember to" --robot --limit 10 --days 30 --fields summary
cass search "I told you" --robot --limit 10 --days 30 --fields summary
cass search "don't do" --robot --limit 10 --days 30 --fields summary
cass search "wrong format" --robot --limit 10 --days 30 --fields summary
cass search "should have" --robot --limit 10 --days 30 --fields summary
These are starting points. Adjust or add queries based on what you find. The signal is in Jamie's messages, not the agent's.
Search for phrases that suggest the agent is being taught something