ksk-keying

Orchestrate the KSK client document keying workflow (classify, extract, review, export to PEAK account data) with a parent session and bounded Agent-tool subagents. Use when asked to "run ksk-keying", "key this client", "process this client with subagents", "segment and review this client", "run the new KSK workflow", or move a client from folder inspection to ข้อมูลระบบ/_segments, ข้อมูลระบบ/_doc_groups, and review artifacts.

Stage 3 of the ksk-keying workflow — link approved segment interpretations into same-transaction clusters (deterministic prelink script, then one ksk-sherlock over the residue). Invoked by the ksk-keying orchestrator after interpretation; NOT a standalone entry point. Do not trigger from a user request — run ksk-keying instead.

Stage 0 of the ksk-keying workflow — first-contact client profile (CLIENT.md, coa.csv, coa_usage.json record) plus the deterministic inventory census. Invoked by the ksk-keying orchestrator as the first stage of a run; NOT a standalone entry point. Do not trigger from a user request — run ksk-keying instead.

Compare a finished ksk-keying run against its client's answer key (samples/answer-keys/<client>/) for one or more named clients, and produce a short Thai-language HTML report per client showing what the pipeline got wrong. Use when asked to "check <client> against the answer key", "grade this run vs answer key", "/ksk-keying-answer-check <client>", or to audit accuracy after a run finishes. Dev-repo tooling; needs the gitignored samples/answer-keys/ dataset on this machine.

Stage 5 of the ksk-keying workflow — map each doc group to chart-of-accounts codes (⚡ ksk-poirot wave), build review-data, and generate the human review HTML. Invoked by the ksk-keying orchestrator after grouping; NOT a standalone entry point. Do not trigger from a user request — run ksk-keying instead.

Stage 4 of the ksk-keying workflow — build the doc-group category/VAT tree (deterministic group-skeleton + group-populate scripts) and populate the judgment groups with a ⚡ ksk-marple wave. Invoked by the ksk-keying orchestrator after linking; NOT a standalone entry point. Do not trigger from a user request — run ksk-keying instead.

Stage 2 of the ksk-keying workflow — interpret approved segments (⚡ watson/marple wave), validate shape, audit exclusion claims (⚡ lestrade), merge dispositions, pass the interpret Ledger Gate, then patch CLIENT.md from evidence (Stage 2.5). Invoked by the ksk-keying orchestrator after segmentation; NOT a standalone entry point. Do not trigger from a user request — run ksk-keying instead.

Stage 1 of the ksk-keying workflow — segment the month run root into document/transaction units (manifest.yaml + SUMMARY.md), resolve the policy gate, pass the segment Ledger Gate. Invoked by the ksk-keying orchestrator after Stage 0; NOT a standalone entry point. Do not trigger from a user request — run ksk-keying instead.

Run the agent eval suite (evals/) — dispatch eval cases to a ksk agent under test via the Agent tool, grade deterministically, report vs baseline. Use when asked to "run the watson eval", "/ksk-eval watson", measure an agent, or check a prompt change for regressions. Dev-repo tooling; needs the gitignored samples/evals/ dataset on this machine.