--- name: ccusage-agent-wait-time description: "Every Monday: how many times your agents asked you something, the median wait for your answer, and whether the longest waits ended in a merged PR. Use when the user asks for this, or when it runs on a schedule." --- # Where your agents waited on you: session logs and GitHub Every Monday: how many times your agents asked you something, the median wait for your answer, and whether the longest waits ended in a merged PR. For developers and founders who run several agent sessions at once and suspect the agents are faster than they are. Every question an agent asks and every permission prompt it raises sits idle until you look. The session logs on your computer have the timestamps; GitHub has what became of the work. The agent measures each wait from the log timestamps, ties the session to its repository and branch, and checks whether that PR merged. The card shows only counts and times, never what was said. The usual finding is that the long waits cluster on the work that never shipped. ## Two ways to get the data Use the API or MCP route when you have an MCP server, a CLI or an API token for the app. Otherwise, if you can drive a browser the user is signed in to (Claude in Chrome, a Playwright or Chrome DevTools MCP server), use the browser route. Say which one you used. ## Route A: API or MCP server ### What you need - Runs on the computer where Claude Code and Codex ran, with read access to `~/.claude/projects/` and `~/.codex/sessions/` and Node for `npx ccusage@latest`. - The shared key: one branch per piece of work, and a GitHub `origin` remote in each project folder; `gh` signed in with read access to pull requests. If something is missing, say what and stop; do not invent numbers. ### Steps 1. **List last week's sessions** (ccusage): Run `npx ccusage@latest claude session --json --since --until ` and the same with `codex` to get the session IDs worth reading, with `firstActivity` and `lastActivity`. ccusage summarizes per session, so the per-message timing comes from the logs themselves. 2. **Measure each wait in the logs**: In `~/.claude/projects/*/.jsonl`, a wait starts at the last assistant line of a turn that ended on a question to you, or on an Edit or Write tool call whose result came 30 seconds or more later (a permission prompt), and ends at the next line you typed. Tool results also appear as user lines, so skip those, and skip subagent lines. In Codex rollout files under `~/.codex/sessions/`, use the agent message and user message events the same way. Waits over 12 hours count as parked, outside the median. Decide what is a question locally; keep only timestamps. 3. **Join each wait to its PR**: Each Claude Code line carries `cwd` and `gitBranch`; Codex records the working folder at the start of the file. Map the folder to `owner/repo` with `git -C remote get-url origin`, then `gh pr list --repo --head --state all --json number,state,mergedAt,url`. Waits on the default branch or a branch with no PR are listed as "no PR"; when a branch has several PRs, take the one open at the time of the wait. 4. **Read each PR's outcome** (GitHub): Mark each PR merged, closed without merge or still open, from `state` and `mergedAt`. Compare the median wait on merged PRs with the median on the rest. 5. **Post counts and times only**: Read last week's card with `GET https://briefkin.dev/v1/cards/insight-ccusage-agent-wait-time` for the history, then post the median wait, the number of asks answered, and the 5 longest waits with day, project folder name, agent, minutes and PR outcome. Never include prompt, question or code text. ## Route B: the browser ### What you need - Signed in to GitHub with read access to the repositories where your agents open PRs and comment. If something is missing, say what and stop; do not invent numbers. ### Steps Start at https://github.com/pulls. The user is signed in already; never type passwords or codes, and ask if a sign-in page appears. 1. **Find PRs where an agent asked you something**: Search `is:pr involves:@me updated:>=`. Open PRs where your agent posts as a bot account (Claude in GitHub Actions, Codex cloud tasks) and look for its comments that end in a question to you. 2. **Measure each wait**: For each such comment, note its time and the time of your next comment or review on the PR. Hover the relative time to see the exact timestamp. Waits over 12 hours count as parked. 3. **Read each PR's outcome**: At the top of the PR, note Merged, Closed or Open. List the 5 longest waits with day, repository, wait and outcome. 4. **Sum up**: Give the number of asks answered, the median wait, waits over an hour, and the median wait on merged against unmerged PRs. Note: A browser cannot read the session logs on your computer, so this route sees only the questions agents ask on GitHub. Local Claude Code and Codex sessions, their permission prompts and most day-to-day waits are invisible here; for those, run the skill on the computer itself. ## The report - **Median wait** for your reply this week, against previous weeks. - **Asks answered**, waits over an hour, and waits parked overnight. - **The 5 longest waits**: day, project, agent, wait and whether the PR merged. - **Merged against not merged**: the median wait on each. - One line on what to do first, for example "Turn on accept-edits mode for acme-web: most of its long waits were edit approvals". ## Post it to Briefkin When `BRIEFKIN_TOKEN` is set, or `~/.config/briefkin/env` exists (the installer from https://briefkin.dev/docs writes it), also send the result to the user's Briefkin board as one card. Keep the `item_key` the same on every run so the card updates in place, use a new `event_id` per run, and put today's real numbers in place of the example ones. If the request fails, say so once and carry on. ```bash [ -n "${BRIEFKIN_TOKEN:-}" ] || . ~/.config/briefkin/env curl -sS -X POST "${BRIEFKIN_URL:-https://briefkin.dev}/v1/events" \ -H "authorization: Bearer $BRIEFKIN_TOKEN" \ -H "content-type: application/json" \ -d @- <<'JSON' { "v": 1, "event_id": "ccusage-agent-wait-time/", "item_key": "ccusage-agent-wait-time", "project": "acme-web", "source": "Where your agents waited on you: session logs and GitHub skill", "kind": "metric", "title": "Your 3 longest waits sat on the 2 PRs that did not merge", "summary": "You answered **31 asks**, median wait **42 minutes**. The 3 longest waits sat on PR 412 (closed) and PR 98 (still open).", "metric": { "label": "Median wait for your reply", "value": 42, "unit": "min", "period_label": "Last 7 days", "better": "lower", "period_start": "", "period_end": "" }, "glance": [ { "type": "Grid", "props": { "columns": 3 }, "children": [ { "type": "Stat", "props": { "label": "Asks answered", "value": 31, "delta": 11 } }, { "type": "Stat", "props": { "label": "Waits over 1 h", "value": 9, "delta": -10, "better": "lower" } }, { "type": "Stat", "props": { "label": "Parked overnight", "value": 4, "better": "lower" } } ] } ], "widget": [ { "type": "Table", "props": { "columns": [ "When", "Project", "Agent", "Wait", "PR" ], "align": [ "left", "left", "left", "right", "left" ], "rows": [ [ "Tue 15:10", "acme-web", "Claude Code", "6.2 h", "PR 412, closed unmerged" ], [ "Thu 11:40", "acme-api", "Codex", "4.8 h", "PR 98, still open" ], [ "Wed 16:55", "acme-web", "Claude Code", "3.9 h", "PR 412, closed unmerged" ], [ "Mon 10:05", "acme-api", "Claude Code", "1.7 h", "PR 97, merged" ], [ "Fri 13:20", "acme-web", "Codex", "1.4 h", "PR 415, merged" ] ] } }, { "type": "Callout", "props": { "text": "Median wait on merged PRs: 28 minutes. On PRs that did not merge: 2.9 hours.", "tone": "info" } } ], "size": "md" } JSON ``` Widgets come from the Briefkin catalog (https://briefkin.dev/v1/catalog). Without a token, just give the user the report. Set `period_start` and `period_end` to the window the number covers: the tile names it from them ("Last 7 days", "Today", "This month"). For a count right now (open items, a balance), send `"period_label": ""` so no period shows; for another name, send it as `period_label` (at most 24 characters). ## Run it on a schedule Suggested: Mondays at 8:30 (`30 8 * * 1`). From cron, with the agent signed in on that machine: ```bash 30 8 * * 1 claude -p "Run the ccusage-agent-wait-time skill" # Codex: codex exec, Gemini CLI: gemini -p, Copilot CLI: copilot -p, OpenCode: opencode run ``` Or use the agent's own scheduler (Claude Code routines, Codex automations, Claude or ChatGPT scheduled tasks): https://briefkin.dev/skills/ccusage-agent-wait-time#schedule Source: https://briefkin.dev/skills/ccusage-agent-wait-time