Grill With Docs

Interview me relentlessly about every aspect of this plan until we reach a shared understanding. Walk down each branch of the design tree, resolving dependencies between decisions one-by-one. For each question, provide your recommended answer.

Ask the questions one at a time, waiting for feedback on each question before continuing.

If a question can be answered by exploring the codebase, explore the codebase instead.

Domain awareness

This repo is Activepieces. It is split into bounded contexts; each has a CONTEXT.md glossary, CONTEXT-MAP.md at the root lists them all and how they relate, and .agents/features/ is the per-feature registry:

/
├── CONTEXT-MAP.md                     ← lists every context + relationships (read this first)
├── CONTEXT.md                         ← Execution Runtime context glossary
├── docs/adr/                          ← architecture decision records
│   ├── 0001-engine-posts-run-callbacks-directly.md
│   └── ...
├── packages/server/sandbox/CONTEXT.md ← Sandbox Pool context
├── packages/server/worker/CONTEXT.md  ← Worker Runtime context
└── .agents/
    ├── contexts/<context>/CONTEXT.md   ← product-domain context glossaries
    │                                     (automation-core, data-storage, pieces,
    │                                      platform, authentication, ai, eventing,
    │                                      releases, infrastructure)
    └── features/<feature>.md           ← one living doc per feature: the registry

Read CONTEXT-MAP.md first to find which context the current topic belongs to. Create files lazily — only when you have something to write; if a term belongs to a context that doesn't exist yet, add it to CONTEXT-MAP.md and create its CONTEXT.md.

Before grilling a NEW feature: overlap detection (mandatory)

Activepieces forbids redundant features. If the plan introduces something that sounds like a new feature, stop and run feature-overlap detection before grilling the design. Complete all five checks before concluding:

1. Feature docs — read the .md files in .agents/features/; this is the primary inventory of what already exists.
2. Components / services / hooks — Glob packages/**/*<keyword>*.{ts,tsx} and Grep their contents; also check directory patterns like packages/server/api/src/app/<concept>/.
3. Route definitions — Grep packages/server/api/src (glob *.ts) for a route already covering the use case.
4. Shared types — Grep packages/core/shared/src for existing types/enums for the concept.
5. Feature flags / plan limits — Grep packages/core/shared/src/lib for a capability or plan flag that already gates it.

Always present findings before proceeding — never silently skip this:

During the session

Scan for domain terms

As you grill, collect every noun/verb/phrase that names a core entity (Flow, Piece, Step, Run, Connection, Project, Platform), names a process (publish, trigger, execute, sync), carries codebase-specific meaning, or is used inconsistently.

Flag ambiguity before resolving — never resolve silently

Challenge against the glossary

When the user uses a term that conflicts with the canonical language in the relevant context's CONTEXT.md, call it out immediately. "Automation Core defines 'run' as FlowRun and lists 'execution' under Avoid, but you're using 'execution' — which is it?"

Sharpen fuzzy language

When the user uses vague or overloaded terms, propose a precise canonical term. "You're saying 'connection' — do you mean the saved credential (App Connection) or a live socket?" Honour the canonical term + Avoid discipline: pick one canonical word, list the rest under _Avoid_.

Discuss concrete scenarios

When domain relationships are being discussed, stress-test them with specific scenarios that probe edge cases and force precision about the boundaries between concepts.

Cross-reference with code

When the user states how something works, check whether the code agrees; surface contradictions. Keep multi-tenancy and editions in view — Activepieces is Platform → Project → User, and features differ across Community / Enterprise / Cloud.

Update the docs inline — don't batch

When a decision crystallises, capture it right there:

• A term is resolved → update the matching context's CONTEXT.md (find it via CONTEXT-MAP.md) using CONTEXT-FORMAT.md: a one/two-sentence definition plus an _Avoid_: line for retired aliases. If no existing context fits, add a new one to CONTEXT-MAP.md and create its CONTEXT.md. CONTEXT.md files are glossaries — no implementation details, specs, or decision logs.
• A feature is created or its behaviour/files/edition/scope changed → update its feature doc (see below).

Feature documentation registry

Every feature has a living doc at .agents/features/<feature>.md. It's the registry the overlap check reads. Keep it current — every change to a feature's behaviour requires updating its doc — using this structure:

# <Feature Name>

## Summary
One-paragraph description of what this feature does and why it exists.

## Key Files
- `packages/web/src/features/<path>` — frontend components
- `packages/server/api/src/app/<path>` — backend service/controller
- `packages/core/shared/src/lib/<path>` — shared types

## Edition Availability
Which editions support this feature (Community, Enterprise, Cloud).

## Domain Terms
Key terms used by this feature (link to their CONTEXT.md context).

When modifying an existing feature, reflect new/removed key files, changed edition availability, new domain terms, and any scope change to the summary.

Offer ADRs sparingly

Only offer to create an ADR when all three are true:

1. Hard to reverse — the cost of changing your mind later is meaningful
2. Surprising without context — a future reader will wonder "why did they do it this way?"
3. The result of a real trade-off — there were genuine alternatives and you picked one for specific reasons

If any of the three is missing, skip the ADR. Use the format in ADR-FORMAT.md; ADRs live in docs/adr/ with sequential numbering.