| name | wt-schedule |
| description | Estimates a commitment, links it to its project, and blocks time for it on the user's calendar. Use when the user asks to find time for something — "when can I do this?", "find me two hours for the CFP", "block time for that tomorrow", "I want to work on it today" — and immediately after wt-capture files a commitment that needs real hours. Do NOT use for a commitment that needs only a deadline or a reminder, and do NOT use to change what the user committed to — that is wt-update. |
| metadata | {"aos":{"origin":"work-tracker@0.1.0"}} |
schedule
Find real time for a commitment and write it down. This runs when the user speaks, not
overnight — "I want to work on the CFP tomorrow" is worthless if the block appears at
midnight.
1. Estimate
How long is this, honestly? Prefer the user's own number where they gave one; otherwise a
rough estimate beats none, because it decides how much calendar to look for and lets the
steward spot a two-minute job that has sat for three weeks.
Record it as estimate: on the action — 45m, 2h. Keep the granularity from onboarding;
invented precision (37m) reads as noise.
2. Read the calendar before proposing anything
Existing blocks, meetings, and the working windows from the global MOD.md. Then two rules
that come from the user, not from this skill:
- Working hours bound where a block may land at all.
- Sacred time is never scheduled over, and never silently moved. This skill writes to a
real calendar, so that rule is what makes the capability trustworthy rather than something
the user has to check up on. A conflict gets surfaced — "the only two-hour gap is during
choir practice; shall I look at Friday instead?" — never resolved on their behalf.
The global MOD.md's red lines apply here too. Sending an invitation to another person is
acting outward as the user; blocking time on their own calendar is not.
3. Take the slot and write it
wt-capture hands you the page path. Reached directly ("find me two hours for the CFP"),
find it first rather than guessing at the slug — the open set is small:
kb --base commitments find --where type=action --where status=next
kb --base commitments set actions/write-the-cfp.md block=<TIMESTAMP> estimate=45m
kb set validates every key against the base's schema and makes its own attributed commit,
so nothing else is needed. Both fields are declared at install — an undeclared one exits 14
rather than writing a page lint would later flag.
Link the action to its project in the same pass where one exists. The outer quotes are
load-bearing — kb set runs each value through a YAML parser, so a bare [[…]] is read as
a nested list and stored as one, silently and with exit 0:
kb --base commitments set actions/write-the-cfp.md 'project="[[projects/booking-deal]]"'
That link is what lets the steward answer "what is this for" without reading the body — via
kb find --where 'project=[[projects/booking-deal]]', since kb links reads body text only
and never frontmatter.
4. Report the slot back
Name the actual time — "blocked 10:00–10:45 Saturday". A confirmation that does not say
when is the one thing the user cannot verify at a glance.
Where this runs
Where the harness can spawn a sub-agent, run in one. Not for latency — taking ten seconds to
find a good slot is fine, even reassuring — but because finding one means reading the
calendar, the existing blocks and the project page, none of which the conversation needs in
its context afterwards.
Inline is a correct fallback, not a failure. Say which one happened.
Without a calendar
Say so plainly, once: actions, statuses, deadlines and the steward all still work; there are
no blocks. Then set estimate: anyway and leave block absent — the steward's
--without block backstop is exactly the query that finds these, so a commitment with no
calendar is tracked rather than lost. Silently doing nothing is the one unacceptable
outcome, because the user believes time was found.
