# Time Report > Log the time spent on a Jira story during the current Claude Code session as a Jira worklog (mirrored into Tempo Timesheets), computed from the session transcript instead of typed by hand. Use when the user says "log time", "log my time", "time report", "record this in Tempo", "how long did I spend on ", or when a Stop-hook nudge reports unlogged time — and at the natural end of any session in which a Jira story (e.g. PROJ-12345) was actively worked on, offer to run it. Do NOT use it for editing or reporting on Tempo timesheets in general, for estimating future effort, or for stories that were merely mentioned in passing. - Skill: `qili09/time-report` (Agent Skill, multi-file: 6 files) - Install (CLI): `npx skillmds@latest add qili09/time-report` - Raw SKILL.md: https://api.skillmd.com/api/skills/qili09/time-report/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Docs & Writing - Author: qili09 (https://skillmd.com/u/qili09) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/qili09/time-report --- # time-report Turn a working session into a Tempo entry with one confirmation. The transcript is the timesheet: which story was worked on, for how long, and what was done are all derivable from it. The skill proposes; the user confirms; the script posts and records a ledger entry so the same session is never logged twice. Decisions and evidence behind the design: [PLAN.md](PLAN.md). ## Workflow 1. **Locate the transcript.** `~/.claude/projects//.jsonl`. The session id is in the scratchpad path given in the environment; the cwd slug is the working directory with `/` replaced by `-`. If in doubt, pick the most recently modified `.jsonl` in that folder. 2. **Compute the proposal.** ```bash python3 scripts/session_time.py [--issue KEY] ``` Output: `allocations[]` — **one per story the session worked on** — each with `issue`, `active_seconds`, `rounded_seconds`/`rounded_human`, `started`, `ended`, `already_logged`, and `unlogged_*` (the remainder not yet in the ledger). Rules: active time is the sum of gaps ≤ 20 min between consecutive records; time is charged to the story the conversation was on at that moment (a user message naming a story switches it); each story is rounded to 15 min with a 15-min floor. Keys come only from user-authored text (slash-command args weigh ×3); assistant output and tool results never nominate a story. Whole-session totals (`primary_issue`, `rounded_human`) are also returned for reference. 3. **Reconcile with the arguments.** An explicit `ISSUE-KEY`, duration, or description on the command line overrides the computed value (with `--issue KEY` the whole session is charged to that key — the script replaces the per-story split with one allocation). A session that crossed several stories gets **one entry per story** (user rule, 2026-09-09) — never merge them. Drop an allocation only if the user says that story was merely mentioned; if the split looks wrong, show it and ask — do not guess. **Exception — the ticket the session created** (rule, 2026-09-16): when a story's slice is under a minute and its key first appeared because the session *created* that ticket (or the user only named it at the end to ask where to log), do not propose two 15-min entries for one piece of work. Re-run with `--issue ` on the story the work belongs to — normally the created ticket — and propose a single entry. 4. **Draft the description** in one line, past tense, from what was actually done (e.g. “Validated requirements in the redirect solution doc; 13 findings”). No Claude/AI attribution, no markdown. 5. **Draft-first gate — includes the inferred billing key.** The Tempo *Billing Key* is an Account attribute (attribute name and choices in `~/.claude/time-report/config.json`). Infer it from what the story and the session were about (user rule, 2026-09-09), using the team's RUN/BUILD split: - **`-RUN`** — work on *existing* functionality: keeping something running, analysis or migration of current behaviour, defects, operations, support, cleanup, decommissioning, config or infra changes that do not add capability. - **`-BUILD`** — *new* functionality or an *improvement*: new features, new integrations, redesigns, enhancements that change what the system does for users. - **Another product's key** — only when the story belongs to a product line that has its own billing key in the config choices. - **Unsure between RUN and BUILD** (e.g. a migration that also adds capability, or a story whose description is empty) → ask, with the one-line reason for the doubt. Show **one line per story** and wait for a yes; keys are visible so a reply like "yes, but the second one BUILD" overrides them: `Log 3h 15m to PROJ-1234 (2026-09-09) as TEAM-RUN (parity migration of existing redirect): ""?` `Log 15m to PROJ-1239 (2026-09-09) as TEAM-RUN (DNS zone setup for the same migration): ""?` If a story is `already_logged` in this session, propose only its `unlogged_human` remainder (a continued session), or say nothing about it when the remainder is 0m. 6. **Post via Tempo — once per confirmed story** (needed for attributes; the script falls back to Jira only if the Tempo token is missing and then warns that the billing key was not set). Use each allocation's own `started` and `rounded_seconds` (or `unlogged_rounded_seconds` for a continued session): ```bash python3 scripts/post_worklog.py post "" \ --session --attr _BillingKey_= ``` Report the returned `tempo_worklog_id` and `jira_worklog_id` per story. The script appends to `~/.claude/time-report/ledger.jsonl`. `post_worklog.py attributes` lists Tempo attributes. 7. **Undo / repair.** `post_worklog.py list `; `post_worklog.py delete `; `post_worklog.py fix --attr _BillingKey_=` sets attributes on an existing entry through Tempo. ## Nudge (Stop hook) `hooks/stop-nudge.sh` reads the hook's `transcript_path` and prints a `systemMessage` when the session has any unlogged active time on a story (≥ 1 min), at most once per hour per session. Registered in `~/.claude/settings.json` under `hooks.Stop`. It only reminds; it never posts. User rule (2026-09-09): every session that touched a story gets logged; work under 15 minutes is logged as 15 minutes. ## Guardrails - A worklog is visible to the whole project — never post without the one-line confirmation. - Never log time to a key that appears only in tool output, search results, or sibling tickets read for context; the transcript ranking already excludes those sources. - Time is rounded, not padded: report the raw `active_seconds` if the user asks. - Tokens: Jira via the mcp-atlassian OAuth file (auto-refresh; if 401 twice the refresh token is dead — see the mcp-atlassian OAuth memory note); Tempo via `~/.claude/time-report/tempo-token` (personal token, scopes worklogs r/w + work-attributes read; `accounts:view` would additionally let `attributes` list every Account). Never paste tokens into chat — have the user write the file themselves. - Jira-created worklogs do sync into Tempo but arrive with **empty work attributes**; that is why posting goes through Tempo. If the Tempo path fails, tell the user the billing key will need setting by hand before falling back to Jira. ## Learning loop When a run hits friction this file didn't anticipate, append a dated note to `FIELD-NOTES.md` immediately. Offer a retro at the end; codify only what the user approves.