dsh-delivery-ledger

You register a timed shell command, it runs, and then there is no record. No audit trail, no redelivery on failure, and process jitter can trigger the same window twice. dsh-delivery-ledger solves all three with a single append-only JSONL ledger: every window execution appends one line with a status of ok, failed, or abandoned; same-window deduplication prevents duplicate runs caused by jitter; and ledger_retry redelivers the failed window using the original dueAt with an incremented attempt counter (Hermes redeliver semantics) rather than re-executing at the current moment.

Three actions cover the full workflow. ledger_create registers a command plus an interval in seconds; the plugin has its own built-in scheduler that fires on aligned interval windows without depending on dsh-schedule. ledger_list queries the ledger and supports filtering by status. ledger_retry redelivers the most recent failed or abandoned window. Everything runs locally with zero network calls and zero LLM — commands execute via /bin/sh -c with a 10-second timeout, the ledger lives as a .jsonl file under $DSH_HOME/ledger/, and the test suite achieves 100% line coverage backed by real headless e2e runs that verify both deduplication and redelivery paths.

Best for developers and ops engineers who need a lightweight recurring shell command inside CI pipelines, containers, or single-machine scripts and want an auditable, redeliverable, deduplicated execution trail. Note that in v1 the task registry is in-memory only and does not persist across processes, making it suited to scheduling within a single process lifecycle.