Parse plain-language time reports into structured Notion logs

Time Ledger is conversational time tracking: the user reports time spent in plain language, such as "read papers 2h, gym 1h, did a leetcode," and the skill parses it into structured rows - activity, minutes, date, and an optional compounding tag - written to the user's own Notion database through the official Notion connector. Its core design is what it calls an honesty contract: anything uncertain becomes a To-confirm row and is batch-asked about, never fabricated, because "a log that quietly invents durations is worse than no log." On the first write of a session it uses Notion search to find the database, filtering for type database rather than a page, whose title contains "time-ledger," reading its data_source_id as the parent for all subsequent create-pages/query calls that session - nothing is pasted or manually configured, and it asks the user to disambiguate if more than one match is found. The companion Notion template defines a fixed field schema with controlled select enums that must be copied exactly: Entry (title), Activity (Reading/Coding/Practice/Fitness/Investing/Meeting/Writing/Life/Other), Minutes (number), Date, which must be written as the expanded "date:Date:start": "YYYY-MM-DD" form since a bare Date value fails with HTTP 400, Status (To-sort/To-confirm/Done), Compounding (Compounding/Consuming/Neutral), and Notes (text). Direct-report parsing rules are concrete: reading material maps to Reading, coding to Coding, leetcode to Practice, gym or running to Fitness, markets or research to Investing, meetings to Meeting, docs to Writing, meals/commute/chores to Life; duration cues map "two hours" to 120, "an hour" to 60, "half an hour" to 30, "a while" to about 30, marked To-confirm as an estimate, and "all morning" to about 180; "today"/"yesterday" resolve to the user's local date, defaulting to today when unspecified; multiple activities in one report split into separate rows, defaulting to additive totals plus one To-confirm row to check that assumption; and the Compounding tag is only applied when obvious, Compounding for a reusable asset like learning or building, Consuming only if the user volunteers it, otherwise left blank rather than guessed.

Use it when the user reports time spent in natural language, says "log it," "log my time," "time ledger," or "tidy up my time ledger," or asks where their time went. In batch-reconcile mode, triggered by "tidy up my time ledger," it queries every row with Status in To-sort, To-confirm, or empty, updates the certain ones to Done, and bundles every remaining question into a single message rather than interrogating item by item. It explicitly does not judge the user's Consuming hours - the ledger is described as "a mirror, not a critic" - and by design it is self-report only, deliberately not auto-tracking apps or screens, since an auto-tracker "knows what was open, not why the time was spent." It requires the user's own Notion workspace, a duplicated copy of the companion template database, and the official Notion connector granted access to that specific database - connector and skill setups currently live on paid Claude tiers.

Input is a plain-language time report or the "tidy up" command; output is one Notion page (row) per activity block, written via create-pages with the exact field schema above, or updated via update-page during reconciliation. A worked direct-report example verbatim: "log it: read ML system design 2h, gym 1h, did a leetcode" produces three logged entries, with the LeetCode one marked To-confirm because no duration was stated and a 20-minute estimate was guessed and flagged for confirmation. Four documented pitfalls and fixes: a bare Date property returns HTTP 400, fixed with the date:Date:start expansion; a known Notion MCP bug (notion-mcp-server#121) can silently drop the expanded date field even on a successful create-pages call, so the row should be read back after the session's first create and backfilled with update-page if Date came back empty; a search for "time-ledger" can return example-row pages instead of the database itself, fixed by filtering for type database; and a model-clock-vs-user-timezone mismatch around midnight should default "today" to the user's local date and ask when it's genuinely ambiguous.

Writes and updates rows only inside the single user-granted Notion database via the official Notion connector (MCP) - no shell commands, no network fetches, and no credentials handled directly by the skill. On claude.ai, Notion's write tools default to requiring approval, so the first write in a session popping an approval prompt is expected behavior, not a hang. It names one related, separate skill, @trading-ledger, applying the same parse-plain-language-into-your-own-Notion-database-and-ask-instead-of-guessing pattern to trading journals covering entry thesis, plan, and emotion.

Anyone who wants to log time spent in their own words rather than filling out a structured form, with the guarantee that anything the parser is unsure about gets asked about in one batched message rather than silently invented - useful for a running personal time ledger that stays honest about its own uncertainty.