Parking

Parking context

You execute the parking protocol – record the current state of work so that any future session (or return in the same chat) can restore the context from the stop point.

Why should I?

Switching context costs ~23 minutes to restore focus (Ophir & Nass research). Without a clear point of return, the original task is lost — especially when the context box is clogged with new material. Parking solves this problem: it creates a formalized breadcrumb that survives both this chat and the next one.

Task routing model

Parking works within the task-routing model (see task-routing-methodology-2026-04.md and Rules 10–14 AGENTS.md]). Key invariants:

• plan.md (Goal, Milestones, **Blockers, Drift Guard, Contingency) Parking required changes plan.md §Blockers if the interrupt type is BLOCKED or DELEGATED with milestone-impact. Blockers are the first-class essence of the plan, not the tasks.md section. See §5.4 of the methodology .
• tasks.md - execution queue. The PAUSED marker only comes here if it is the NEED INFO step of the current milestone. Thinking ("you have to deal with X") goes to plan.md as an open question under the milestone, not in tasks.md. Section Blocked in tasks.md net - she moved to plan.md §Blockers.
• 01_now/ops/<contour>/delegations/<slug>.md – If the parking happened because the task was delegated to the employee, the marker goes here. If blocking a milestone, parallel recording in plan.md §Blockers with Type: delegated-work.
• 01_now/personal/tasks.md is a personal external obligation of the owner. Parking doesn’t generate personal strings — it’s a routing event, not a type of interruption.
• 00_inbox/ is only for a “random idea” that has no design or home outline. Again, not parking, but routing.

** S SPAWNED is a categorical error.** The appearance of a new idea while working on the main task is not an interrupt type, but a routing event**: the idea goes to its target file (plan.md of another project, personal/tasks.md, 00_inbox/, delegations/), and work continues without parking. In PAUSED types, this marker no longer exists.

Protocol (strictly follow steps)

Step 1. Define project and task from the context of the conversation

Analyze the current chat and determine:

• **Project: Which 01_now/projects/ project does the work relate to? If it does not apply to any of them, specify “outside the project” and select destination by type (see Step 4): personal → 01_now/personal/tasks.md, curiosity → 00_inbox/.
• **Task **: what exactly was done (not “working on the project”, but “completed 3 out of 5 points of unit economics analysis”).
• **Stop point: what exactly was interrupted - as specific as possible.
• Does the interruption affect the vector of the project? ** If yes, step 3 updates plan.md, not just log.md.

Step 2. Classify type of interruption

Determine the type automatically from the context of the conversation. Do not ask the user if the type is obvious.

BLOCKED vs NEED INFO disambiguation rule: If the type is not obvious, ask one short question: “Do you have an action that will bring the result closer, or are you just waiting?” There's an action - NEED INFO. No action. BLOCKED.

The new idea/task is not parked, but routed to its target file without interrupting the main work (see § "Task Routing Model" above).

Step 3. plan.md is the first-class layer

This is the first layer of the record. The order "slow-protocol.md §5](../../meta/rules/write-protocol.md)" means that plan.md always updates before log.md, tasks.md, delegations/.

By type of interruption:

If the initial approach turned out to be a dead end, the scope changed, the milestone lost its meaning - it is always an edit of plan.md in §Milestones / §Drift Guard / §Contingency, not a blocker.

Step 4. Record a stop frame in the log.md project

Open 01_now/projects/<project>/log.md and add the entry in the format:

- YYYY-MM-DD: <MARKER> PAUSED | Task: <what did>.
  I stopped at: <specific point>.
  Reason: "Why do we switch?"
  Action: "What is done to unlock - to whom he wrote that he delegated, where he wrote the idea."
  Return when: "Return trigger".

A neutral example:

- 2026-03-07: BLOCKED PAUSED | Task: Recalculate the launch model.
  Stopped at: collected all the metrics except one input metric.
  The reason: there is no confirmed data from the responsible person.
  Action: Recorded delegation in the file responsible.
  Return when: I will receive the data or in 3 days.

For the ⏹️ STOPPED type, the format is shorter:

- 2026-03-07: .️ STOPPED PAUSED | Task: <what was done>.
  I stopped at: <specific point>.
  The next step is “what to do when you return.”

Step 5. Create/update the marker task in the target file

The choice of file depends on the type of interrupt:

Convergence rule for delegated-work blockers: if the employee stops blocking milestone (delegation is completed or no longer critical), the blocker is closed according to ritual §5.4.2 methodology - the card moves to ## Blockers — Resolved, the milestone status is removed, the line in delegations/<slug>.md goes to Done (7d).

A neutral example of a blocker card in plan.md:

### B1 - Missing indicator from the responsible
- Open: 2026-03-07
- Type: delegated-work
- Waiting: a confirmed indicator to recalculate the model
- Block: M3
- My action: remind the person responsible in 3 days
- Condition of return: confirmed data received
- Fallback: Take a score from the latest report and mark as an approximation
- Stale check: 7 days

A neutral example of a parallel string in delegations/responsible.md:

- [ ] 2026-03-07 → collect the missing indicator [TRACKER-123](url) • due: 2026-03-10 • link to `plan.md#b1`

Step 6. Confirm the parking.

Respond to the user with one short message in the format:

Parked. [Project X] — recorded a stop frame in log.md, a marker in [target file]. (if applicable) plan.md is updated: [what has changed]. Stopping point: [Short description]. Return trigger: [condition]. We can switch.

Example of delegation:

Parked. 2026-example - log.md, delegation to ops/2026-example/delegations/responsible.md. Stopping point: collected all metrics except one metric. Return trigger: Responsible sends data or 2026-03-10. We can switch.

Limiting the depth of switching

If the user is already in the first level of switching (moved from the main task A to service B) and wants to dive deeper (task C), block:

“This is the second level of switching. Write the task [C] in tasks.md/inbox – let’s go back to it separately. Now we finish [B] or park and go back to [A].

Maximum permissible depth: 1 level.

Long chat protocol

If the current chat has already had 2+ parking or the dialogue is clearly long and multi-theme – warn:

“The chat became long. I recommend continuing in the new session - I will pick up the context from log.md when calling /resume.

Proactive parking

If the user starts talking about another project / task without an explicit parking command, and there was meaningful work in the current chat - suggest:

“I see we're switching. Do you want to make a stop frame of [the current task] so as not to lose the point of return?

Do not impose – if the user says “don’t” or switching trivial (a short question that does not require deep immersion) – skip.