# Sectioned fan-out: your context

Copy this file to CONTEXT.md next to SKILL.md and fill in every blank. Never put passwords, access keys or card numbers here.

## Agent tool

(How your agent spawns a sub-agent, and whether several calls in one message run in parallel in your tool.)
Example: The Agent/Task tool; several calls in one message run concurrently.

## Model tiers

(Which model handles mechanical sections and which handles judgment sections. Use the names your tool accepts.)
Cheap tier (mechanical, clear spec): ___
Strong tier (judgment, security, taste, review): ___
Example: Cheap tier: a small fast model. Strong tier: your most capable model.

## Escalation allowed

(May the agent rerun a weak cheap-tier section on a stronger tier without asking you first? Answer yes or no.)
Escalation allowed: ___
Example: yes, rerun once on the strong tier, then ask.

## Concurrency cap

(How many section agents may run at once. Derive it from your machine's free memory and your plan's rate limits: start low, raise it while nothing stalls, lower it when free memory is tight.)
Cap: ___
Lower the cap when: ___
Example: Cap 4. Drop to 2 when less than a third of memory is free; run one at a time below that.

## Report folder

(Where section agents write their full reports. A folder outside version control or one you clean up after each run.)
Example: ./tmp/fanout-reports/

## Reply summary limit

(The maximum length of the summary a section agent may put in its reply. Pick a size your parent session can take from every section at once with room to spare.)
Example: a short paragraph, at most a few hundred words.

## Usual seams in this project

(The cuts that fit your codebase or work. List subsystems, user journeys and shared files.)
Subsystems: ___
Main user journeys: ___
Shared files that need a single writer: ___
Example: Subsystems: storefront, checkout API, admin panel. Journeys: browse and buy; refund request. Shared: the database schema file.

## Verification per section

(How a section proves its slice works end to end: a test command, a browser run, a query.)
Example: run the journey in a real browser against the staging site and attach a screenshot path.