Deep Work

Deep Work

Blocking time on a calendar is the easy half; making the time inside the block convert into output is the hard half. This skill covers both — measuring how badly meetings fragment your week, defending blocks against encroachment, and designing sessions so that protected time produces a finished artefact rather than a warm feeling.

When to use this skill

• Your calendar looks manageable in total meeting hours but no real work gets done
• Focus blocks exist but keep getting moved, split, or quietly cancelled
• You need to argue for protected time and want fragmentation data rather than a preference
• Sessions happen but produce fragments — nothing reaches a finished state
• You want to know whether your deep-work practice is improving or decaying over months
• A 30-minute meeting keeps landing in the middle of your only long block

Inputs the skill expects

• A calendar export as JSON — date, start, end, title, optional fixed flag
• Your working-window hours (defaults to 09:00-17:00)
• Your role, which sets a realistic deep-work ratio target (40-60% IC, 20-30% manager)
• A session log with date and actual_min, ideally interruptions and artefact
• Your current weekly deep-work target in minutes, set from logged history rather than aspiration

Clarify First

Before generating, confirm these inputs. If any is unknown or vague, ASK — do not assume:

• Role and realistic target ratio — the 0.40 default is an IC figure and will flag every manager's calendar as failing, which is noise not signal
• Which meetings are genuinely immovable — the fixed flag determines whether a reschedule proposal is actionable or fantasy
• Working-window hours — a 9-5 window on someone who works 07:00-15:00 produces meaningless ratios
• Whether the goal is diagnosis or advocacy — arguing for protected time leads with recovery cost; personal tuning leads with block length

Stop rule: ask only the 2-3 that most change the output. If the user says "just draft it," proceed and list your assumptions at the top of the artifact.

Workflows

Workflow 1 — Diagnose calendar fragmentation and get a reschedule

Run weekly against the coming week, before it fills up.

1. Export the calendar to JSON with date, start, end, title.
2. Mark genuinely immovable events with "fixed": true — the search skips them.
3. Run the analyser with your role's target ratio.
4. Read the per-day table: any day with a longest block under 90 minutes produced no substantive individual work, whatever your output log says.
5. Make the one move it proposes. One move per week is politically sustainable; a wholesale calendar rewrite is not.

python3 personal-productivity/deep-work/scripts/calendar_fragmentation.py \
  --input personal-productivity/deep-work/assets/sample_calendar.json \
  --target 0.40

For a manager's calendar with a longer working window and a role-appropriate target — the same file scored at the 0.40 IC default reads as failing on 14 of 20 days, which is target miscalibration rather than a real finding:

python3 personal-productivity/deep-work/scripts/calendar_fragmentation.py \
  --input personal-productivity/deep-work/assets/sample_calendar_manager_4weeks.json \
  --day-start 08:00 --day-end 18:00 --min-block 90 --target 0.25

Workflow 2 — Track whether the practice is improving

Monthly, not weekly. Weekly volume is noisy enough that reacting to it produces thrashing.

1. Log each session at close-out using assets/session-log-template.md.
2. Run the analyser with a weekly target set from your own logged history.
3. Read the trend line — a change over 10% between the first and second half of the period is real; anything less is noise.
4. Read the findings, which name specific structural failures rather than general encouragement.
5. Fix exactly one thing. Volume, block length, and defence are three different problems; working all three at once teaches you nothing about which mattered.

python3 personal-productivity/deep-work/scripts/session_log_analyzer.py \
  --input personal-productivity/deep-work/assets/sample_sessions.json \
  --weekly-target 600

Workflow 3 — Make the case for protected time

When you need a manager or team to concede structural change.

1. Export two to four weeks of calendar history and run the fragmentation tool.
2. Lead with the recovery cost figure, not the meeting-hours figure. "Meetings take 9 hours a week" invites a debate about which meetings matter; "fragmentation costs another 6 hours a week in re-immersion, on top of the meetings" reframes it as waste.
3. Show the count of days with zero viable blocks.
4. Propose the specific single move the tool identifies — concrete, small, and reversible asks get agreed; general ones get sympathy.
5. Re-run after four weeks and report the delta. The follow-up measurement is what turns a one-off concession into a standing arrangement.

mkdir -p build
python3 personal-productivity/deep-work/scripts/calendar_fragmentation.py \
  --input personal-productivity/deep-work/assets/sample_calendar_manager_4weeks.json \
  --target 0.40 --format json > build/fragmentation.json

Decision frameworks

Block length vs usable output

Problem-state reload costs 15-20 minutes and is paid on every re-entry. This is why five 30-minute gaps are not equivalent to one 150-minute block — the gaps yield close to zero.

Interruption budget

Each interruption costs roughly 23 minutes of degraded output, not the duration of the interruption. [PROVEN]

Set the budget one step below your measured rate, never at zero — zero is unachievable in most roles and failing it immediately kills the practice. Define the consequence in advance: exceeding the budget means reschedule the block, not push through. Pushing through teaches you that protected time yields shallow work, which is the belief you are trying to disprove.

Role-calibrated targets

Most people sustain 3-4 hours of genuine deep work per day as an upper bound. Beyond that the work continues but the quality does not, and the deficit shows up as needing to redo it.