plat 220

← Pulse platform issue index

PLAT-220 — consolidated two more Pulse tool pairs: focus agenda folded into get_pulse_state, the two migration-reconciliation tools merged into one scoped tool

2026-09-10 update: PLAT-306 supersedes mandatory Markdown checkpoint/report maintenance with optional SQLite review notes on the existing result tool. Typed lifecycle records remain authoritative; historical design discussion below is retained. No reporting-only turn is required.

Coordination	Value
Assigned agent	unassigned
Ticket state	implemented
Last synchronized	2026-08-29

• Priority: P3 — tool-surface simplification, not a defect fix. Filed for the same reason PLAT-tickets track PLAT-055's typed-tool migration: a durable record of why the Pulse tool surface looks the way it does.
• Owner: N/A — implemented directly.
• Related: the concurrent, independently-landed a8aaaa012 "Simplify Pulse maintenance and harden lifecycle handling" commit removed complete_pulse_review entirely (findings now persist without a completion handshake) in the same file this ticket touches; both changes were reconciled via a git stash/rebase/pop merge, not a rewrite of either side's intent.

What changed

User-requested, explicitly scoped consolidation of the Pulse agent-facing tool surface, after a design discussion concluding the tool count (not the underlying SQLite structure) was the actual friction:

1. get_pulse_review_focus_agenda folded into get_pulse_state(view="focus_agenda"). get_pulse_state already served three views (module, backlog, review) behind one tool; the focus-agenda read fit the same one-tool-many-views shape. readPulseFocusAgendaView wraps the existing getPulseReviewFocusAgenda and returns the same {"focuses": [...]} payload; route_scope and limit were added as get_pulse_state parameters, gated to this view.

2. record_pulse_lifecycle_reconciliation and record_pulse_actionable_backlog_reconciliation merged into record_pulse_migration_reconciliation(scope="lifecycle"|"actionable_backlog"). ReconcilePulseActionableBacklog already called ReconcilePulseFindingLifecycle as its own first step and embedded the result — the two were never independent capabilities, only two names for "run this migration, optionally with the superset step." An invalid or missing scope is rejected outright rather than silently defaulting to one migration.

Both consolidations are pure surface changes: no SQLite schema, no migration behavior, and no reviewer-facing semantics changed. All known callers were updated — workflow_version_upgrades.go's three contract-upgrade prompt strings (v1.0.32, v1.0.33, v1.0.34), three guidance templates (engineering-review.md, ops-review.md, pulse-review-fixer.md), and the test suite (pulse_review_tools_test.go, pulse_tool_surface_test.go, workflow_schedule_contract_test.go).

Explicitly not done

Two other pairs were considered during the same design discussion and deliberately left alone:

• resolve_run_concern vs record_pulse_result — different write targets (a lifecycle-status flip on an existing concern vs. recording a new, possibly evidence-bearing result); merging them would hide that resolve_run_concern explicitly refuses to close a real finding.
• record_pulse_review_focus vs complete_pulse_review — at design time this pair looked mergeable, but the concurrent a8aaaa012 commit removed complete_pulse_review outright before this ticket landed, so there is nothing left to merge it with.

Markdown-file-based Pulse management (replacing the SQLite-backed tools the way PLAT tickets are managed as files) was proposed and rejected in the same discussion: this session hit a real concurrent git write collision on a PLAT ticket file (PLAT-207, resolved via git checkout --) that SQLite's INSERT ... ON CONFLICT transactions prevent for Pulse writes, and the "stale finding, already fixed, never re-verified" pattern recurred five separate times this session (PLAT-191/195/201/202/212) even with structure in place — a strong signal structure is load-bearing here, not incidental.

Verification

• go build ./... — clean.
• go test ./cmd/server/... and full go test ./... from agent_go/ — all pass, including the pre-existing suite; one new test, TestRecordPulseMigrationReconciliationDispatchesByScope, exercises both scopes plus invalid/missing-scope rejection directly against the merged executor.
• pulse_tool_surface_test.go's exhaustive expected vs registered tool set check (TestPulseToolSurfaceIncludesTypedReviewerWrites) passed after the merge with a8aaaa012, confirming no stale tool name survived from either side of the conflict.
• Grepped the full agent_go/ tree (excluding logs/agent_prompts/, which holds historical run transcripts, not live code or guidance) for the three removed tool names — no remaining references outside the files already updated.