Attendance data can balance and still be wrong. One export stops before page two. A fallback-hour shift loses sixty minutes. A manager edits a punch without approval. Duplicate events inflate time. Held-out overtime quietly slips back into payroll.
Attendance Payroll Handoff turns source attendance, roster, schedule, leave, correction, and approval evidence into a controlled payroll input. It freezes the period and population, records source coverage and time zones, preserves raw events, reconstructs employee-days, and applies only the supplied policy revision. Missing rules stay missing; the workflow does not invent rounding, overtime, leave, pay, tax, or labor-law decisions.
The output separates payroll-ready regular time, approved leave, approved overtime, and unresolved held time. Every correction and exception keeps its source references, owner, required evidence, and resolution state. Bidirectional controls reconcile roster to attendance, leave to approvals, corrections to authorizations, employee-days to period totals, and period totals to payroll rows.
Release states are deliberately strict. Held-out time prevents an unsupported payment, but it does not resolve the exception. A period remains blocked when time-affecting corrections, overtime, policy conflicts, source gaps, or mapping issues are open. Ready for approval means the final reviewer's signature is the only remaining gate.
The package includes the core workflow, an attendance reconciliation contract, exception codes, source and employee-day field definitions, payroll handoff fields, release gates, and Codex interface metadata. It can review exports from Lark, spreadsheets, time clocks, or HR systems. It does not authenticate to those systems, change punches, approve leave or overtime, interpret labor law, calculate statutory amounts without supplied rules, or write payroll without separate authority.
Version 1.0.0 was evaluated on 30 July 2026 through Codex CLI with gpt-5.6-sol in 12 fresh read-only contexts. Across three blinded cases, the candidate scored 97.92, compared with 70.83 for the same agent without the skill and 77.92 for the exact free Lark attendance upstream. It passed every case and led the aggregate comparison by 27.08 and 20 points. The cases covered incomplete Lark pagination and a daylight-saving shift, corrections and capped overtime, and recovery from an ambiguous payroll import. This evaluation measures the supplied cases; it does not prove a real payroll run was changed or approved.
The reviewed package SHA-256 is 6f9757e57789b5945a8651e5f516023462be0f8a94e8281dc19f64053ed8612c. Maintenance and install support are handled by the platform-operated publisher for this listing.