[Roadmap][Phase 2] codegraph-vba Event declarations — P1 sub-prompt within round-2 #21

Goal

Ship Event declarations modeling as a high-priority sub-prompt within the broader codegraph-vba round-2. Specifically: model Event ... RaiseEvent declarations AND find every subscriber / handler across the project (WithEvents, Set X = ... patterns). This is the P1 slice of round-2 — the maintainer is invited to ship it first or standalone if the broader round-2 scope is too large for a single PR.

Why this is P1

The orchestrator consumes codegraph-vba across multiple Access projects where Event ... RaiseEvent is heavily used:

• gestion_riesgos (around 12 Event declarations across forms and classes)
• no_conformidades (around 8 Event declarations)
• expedientes (around 6 Event declarations)

Without Event modeling, the orchestrator's static reachability analysis over those projects is incomplete. Today's workaround: grep for the event name in .cls files — slow, error-prone, no caller coverage.

Consumer context

Once shipped, this unblocks the new skill vba-event-tracer (ardelperal/VBA_TOOLKIT_BENCH#9), which takes an event name and returns: list of RaiseEvent call sites + list of handlers attached via WithEvents or Set X = .... That skill is on the orchestrator's Phase 1 backlog.

This also feeds vba-handler-backtrace (ardelperal/VBA_TOOLKIT_BENCH#10), which traces the chain control event → handler → helpers → DAO calls.

Acceptance criteria

• Event declarations (e.g., Public Event AlgoChanged(ByVal NewValue As String)) modeled as event nodes with module + line + signature attached
• RaiseEvent invocations modeled as edges (caller → event node)
• WithEvents X As ClassY declarations modeled as subscriber edges (variable → event node)
• Programmatic Set X = New ClassY + later X_SomeEvent handler wiring detected via heuristic (for VBA's implicit WithEvents after Set, depending on the project's coding convention)
• codegraph_explore "Event NombreEvento" returns: declaration site + list of RaiseEvent callers + list of handlers
• RED + GREEN tests against Gestion_Riesgos.accdb (in ardelperal/VBA_TOOLKIT_BENCH) using a real Event like Form_FormRiesgosGestionRiesgo.NotificacionEnviada or equivalent
• Edge cases: event declared but never raised → returns empty raise_sites; event raised but no subscriber → empty handlers (with no false positives from string literals)
• Documented in ardelperal/codegraph-vba/README.md node kinds table

Self-verification (mantainer-side)

Before closing this issue, the maintainer must verify ALL of the following:

• Event node kind exists: codegraph_explore against the bench returns at least one event node for a real Event declaration in Gestion_Riesgos.accdb
• Trace from RaiseEvent works: A specific test calls codegraph_explore with "RaiseEvent X" and returns the declaration site + at least one caller site correctly
• WithEvents / Set X subscribers detected: A specific test verifies a WithEvents declaration is detected as a subscriber edge
• No false positives from comments / strings: A specific test includes an Event name appearing in a string literal + a comment, and confirms zero raise_sites returned for that surface (only real call sites counted)
• Tests verde against the bench: RED + GREEN tests pass against ardelperal/VBA_TOOLKIT_BENCH/Gestion_Riesgos.accdb
• No regresiones: codegraph-vba v1.2.0 existing queries keep working
• Documented in README: The new event kind is in the node kinds table in ardelperal/codegraph-vba/README.md
• Backward compatible schema: If any node kind shape changed vs round-3 (feat(vba-extractor): add VBA / Access language support to codegraph #1) or earlier round, document the migration in the PR description

Only when ALL checkboxes are checked should this issue move to status:approved → status:done. If any is missing, comment which one and why before closing.

Scope vs broader round-2

This issue focuses exclusively on Events. The other round-2 scopes (Type, Declare, huecos 5/3/9) are tracked at ardelperal/codegraph-vba#2 and can ship together OR separately. This P1 sub-prompt lets the maintainer ship Events alone if the rest of round-2 needs more time.

Required side-effect

Closing this issue unblocks vba-event-tracer (ardelperal/VBA_TOOLKIT_BENCH#9) and partially unblocks vba-handler-backtrace (ardelperal/VBA_TOOLKIT_BENCH#10 — full unblock needs Type + Declare too).

References

• Plan: https://github.com/ardelperal/VBA_TOOLKIT_BENCH/blob/main/plans/plan-ambicioso-dysflow-codegraph-vba-2026-06-29.md (Phase 2.1)
• Epic Phase 2: https://github.com/ardelperal/VBA_TOOLKIT_BENCH/issues/2
• Companion round-2 issue (broader scope): fix(vba): resolve parser, SQL, and form control extraction bugs #2
• Round-3 (already OPEN, parallel work): [Roadmap][Phase 1] Round-3: model Enums + Consts + .sql queries in graph #25
• Consumer skill it unblocks: https://github.com/ardelperal/VBA_TOOLKIT_BENCH/issues/9
• Test bench: https://github.com/ardelperal/VBA_TOOLKIT_BENCH
• Maintainer prompt pack format: existing dysflow #611-#616 for the established style