plat 245

← Pulse platform index

PLAT-245 — Pulse contract-upgrade prompts omitted required workflow identity

Coordination	Value
Assigned agent	Codex
Ticket state	implemented; scheduler restart and LinkedIn retry required
Last synchronized	2026-08-29

• Priority: harness issue, severity high.
• Observed workflow: LinkedIn contract upgrade from 1.0.31 to 1.0.32.

Root cause

The 1.0.32, 1.0.33, and 1.0.34 scheduler-generated migration prompts showed record_pulse_migration_reconciliation and get_pulse_state calls without workspace_path. Both HTTP-backed tools require that field, so the argument validator rejected the migration before it touched Pulse data. The safeguard correctly prevented the workflow contract stamp, but every later schedule would retry the same malformed instruction.

Fix

The three prompt templates now carry an explicit workspace placeholder. The scheduler binds it to the exact ScheduleContext.WorkspacePath before sending an upgrade turn, using a quoted value safe for the displayed tool-call syntax. The workflow-contract status view applies the same binding so the operator sees the actual instruction that will run.

Verification

TestPulseMigrationUpgradeTurnsBindRequiredWorkspacePath covers all three Pulse migration rungs and requires both calls to contain the exact workflow path, while rejecting an unbound placeholder. The complete go test ./cmd/server/... suite passes.

Reverify

Restart the scheduler onto this build and trigger LinkedIn again. Its next upgrade turn must reconcile Workflow/linkedin, read the resulting backlog, and stamp 1.0.32; the same preflight can then advance the remaining rungs.