Skip to content

Timesheet

Bucket: Productivity · Slash command: /zsl:timesheet · Source: skills/productivity/timesheet/SKILL.md

When this skill activates

Claude Code matches this skill against the trigger text below. You can also invoke it explicitly with the slash command.

Summarize recent Claude Code session histories into copy/paste-ready timesheet bullets, grouped by project. Use when the user asks "what did I do today", wants a timesheet entry, daily standup notes, or a summary of recent Claude Code sessions.

Build a copy/paste-ready Markdown summary of Claude Code work over a recent window (default 12 hours). The script extracts raw structured session data; Claude synthesizes the bullets. No regex parsing of shell commands lives in the script — Claude reads the bash commands, user prompts, and files-touched data directly and infers outcomes.

Resolve the script (deterministic gate)

The digest script renders the deterministic parts of the timesheet for you — each project's duration label (Xh/X.Yh/Xm) and the window/timezone header line — so you copy them verbatim rather than re-deriving minute→label arithmetic or timezone conversion by hand. Resolve its absolute path once (this works whether the skill is the installed plugin, a personal skill, the remote ~/.claude/skills symlink, or this repo's own checkout):

DIGEST=$({ ls "$PWD"/skills/*/timesheet/scripts/digest_sessions.py 2>/dev/null
           ls "$HOME/.claude/skills/timesheet/scripts/digest_sessions.py" 2>/dev/null
           ls -d "$HOME"/.claude/plugins/cache/zsl-superpowers/zsl/*/skills/*/timesheet/scripts/digest_sessions.py 2>/dev/null | sort -Vr; } | head -1)
[ -n "$DIGEST" ] && echo "resolved: $DIGEST" || echo "zsl-gate: digest_sessions.py unresolved — see Fallback"

Use python3 "$DIGEST" … for every invocation below.

Fallback (if $DIGEST is empty): the gate didn't resolve — find the script under your install (it lives at skills/productivity/timesheet/scripts/digest_sessions.py in the repo) and invoke it by absolute path. If you cannot run it at all, compute the two renders by hand with these exact rules:

  • duration_label from a project's integer active_minutes: if < 60, "{minutes}m"; else hours = minutes/60 → "{int}h" when whole, else "{h:.1f}h" (e.g. 270 → 4.5h, 90 → 1.5h, 180 → 3h, 45 → 45m).
  • window_header: convert window_start/window_end to local time and render "_{start:%Y-%m-%d %H:%M} → {end:%H:%M} {TZ}_". window_phrase: "last {n} hour{s}" — singular hour only when n == 1 (e.g. last 1 hour, last 12 hours).

Required workflow

Never render the final timesheet without first asking which repos to exclude. This applies even when the user says "show me the timesheet" — they will likely want to trim repos out, and unsolicited renders waste their attention.

  1. List candidates. Show projects with active hours and full path:
python3 "$DIGEST" --list

  1. Ask which to exclude. Wait for the response. Suggested phrasing: "Any of these to exclude before I build the timesheet?" Do not render anything yet.

  2. Extract structured data. Once filtered:

python3 "$DIGEST" [--exclude PATTERN] [--only PATTERN] [--merge-nested]

Output is JSON. Top level: window_header (the pre-rendered _<start> → <end> <tz>_ line) and window_phrase (e.g. last 12 hours). Per project: active_minutes, duration_label (the pre-rendered Xh/X.Yh/Xm), and sessions[]; per session, user_prompts, files_touched, bash_commands (deduped, truncated to 300 chars). Copy duration_label, window_header, and window_phrase verbatim — don't recompute them.

  1. Synthesize the timesheet. Read the JSON and write outcome bullets following the rules below. Print the result with the standard header.

  2. Offer to copy. Once the user signals approval (or stays silent on a clean render), offer: "Copy this to your clipboard?" — and on yes, pipe the same text via printf '%s' "<bullets>" | pbcopy (macOS) / wl-copy / xclip -selection clipboard / clip (Windows).

Skip the list+ask step only when the user's request already names the exact projects (e.g. "timesheet for spark-asset-iq, last 4 hours"). When in doubt, list and ask.

Synthesis rules

Build the bullets from the JSON. Apply these rules in order:

  • One bullet per delivered outcome. Find git commits in bash_commands (look for git commit -m, git commit -am, heredoc forms — Claude reads these natively, no regex). Pull the commit subject as the bullet text.
  • PR opens count too. gh pr create --title "..." patterns produce bullets prefixed Opened PR: <title>.
  • Drop redundant signals. gh pr merge is implied by the prior open — don't bullet it. git push is plumbing — don't bullet it.
  • Drop projects with no outcomes. If a project's sessions show only file edits with no commit / PR, omit the project entirely from the timesheet (don't write "in progress" lines).
  • Collapse WIP sequences. If three commits all advance one outcome ("wip", "fix typo", "Add foo"), collapse into one bullet with the outcome subject.
  • Dedupe within a project. Identical commit subjects appear once.
  • Tense. Keep commit-message-style imperative ("Add X", "Remove Y") to match how the messages are written.

Output format (paste-ready Markdown). Group bullets under a per-repo heading, repos sorted by total active time descending. The title line is ## Timesheet — <window_phrase> and the second line is window_header, both copied verbatim from the JSON. Each per-repo heading shows that project's duration_label next to the name (### <name> · <duration_label>) — copied verbatim, not recomputed from active_minutes:

## Timesheet — last 12 hours
_2026-05-09 12:00 → 00:00 NZST_

### spark-asset-iq · 4.5h
- Migrate Cognito user/identity pools to ap-southeast-6
- Polish READMEs with cross-references and updated seed-data layout

### zsl-superpowers · 3h
- Add timesheet skill for Claude Code session summaries
- Bump to 0.4.0 for timesheet skill

Common flags

  • --hours N — window size, decimals OK (--hours 1.5, --hours 24). Default 12.
  • --list — print the project picker (basename, active hours, session count, full path) instead of JSON.
  • --only PATTERN — include only projects matching PATTERN. Bare patterns match basename (case-insensitive substring); patterns containing / match the full cwd. Repeatable.
  • --exclude PATTERN — exclude matching projects (same rules). Ignored if --only is set.
  • --merge-nested — collapse projects whose cwd is a subdirectory of another project's cwd into the parent (useful for monorepos). Active time is unioned, not summed.
  • --include-noise — keep ClaudeProbe / CodexBar health-check sessions that the default filter strips. Implicit when --only is set.
  • --projects-dir PATH — alternative to ~/.claude/projects (rare).