debrief-skilllisted

# Debrief Skill Every skill run leaves a record in the session transcript: each tool call, each error, each thing the user had to say. Read that record for **friction**, the calls, context, and user attention the run spent that a better skill would have spared. Trace each piece of friction to a line of the skill, and fix the skill at its source. A healthy skill debriefs clean. The report then says so in one line per run. ## Step 1: Pick the runs List runs with the bundled script, which reads Claude Code sessions in `~/.claude/projects/` and Codex sessions in `~/.codex/sessions/`. Each output line is a timestamp, the skill name, the `transcript:line` where the run starts, and the directory the skill loaded from, newest first. Without a skill name, the script leaves out runs of this skill. ```bash node ${CLAUDE_SKILL_DIR}/scripts/transcript.mjs runs [--session ID] [--days N] [skill] ``` | Arguments | Runs | |---|---| | Empty | The newest run the user invoked in this session: `--session ${CLAUDE_SESSION_ID}`. A skill that another run loaded through a Skill call belongs to that run: skip its line and take the run that loaded it | | A skill name | Its runs in this session. None there: its runs in every project over the last 30 days | | A skill name and a number | Its runs over that many days | | `self` | The runs of `debrief-skill`, found as a skill name, minus the first output line: that line is this run, still in progress. Apply the "None there" fallback after you remove t